Read more

However, as with any business that has spent time in the spotlight and is going from good to better, the market has already rewarded the stock with 71% returns over the past year, leaving limited room for upside to fair value.

Company profile

Theme: AI Advertising, Direction: Buy

Symbol: ZETA, Exchange: NYSE
Sector: Technology, Industry: Software - Infrastructure
Fair intrinsic value: $34.75 (+7.68%), as of October 15, 2026
Market capitalization: $8 280 million
Pricing data: P/S 5.27x, P/E N/A

TYPEFCAPITAL.COM
Consider following me on X and YouTube.

This article is for informational purposes only and does not constitute investment advice or an offer to buy or sell securities.

A toll booth on a discretionary road

It seems I can’t escape researching advertising businesses lately. Recently, I covered Meta, Alphabet, and AppLovin, but Zeta differs from those three. Pure-play businesses primarily own their distribution channels and operate somewhat like walled gardens. Put simply, pure-plays work by an advertiser choosing a budget, a ROAS goal, and creatives, and then an AI decides who to show the ads to, all within their distribution channels. Zeta does not own a social distribution channel, and ads aren’t limited to a specific app.

Zeta has one of the largest consumer data assets in the U.S., and it rents it out through its software. Their identity graph covers more than 245 million individuals in the U.S., and 535 million globally. Each individual has an average of more than 2,500 attributes, plus over one trillion content-consumption signals ingested every month. Enterprise customers plug their customer data into Zeta’s data store and, through the Zeta Marketing Platform (ZMP), decide who to contact, when to contact them, and through which channel. Channels range from straightforward methods like email and text messages to web, social platforms, and connected TVs. The integrated platform can be seen as a complete advertising OS.

Zeta reports a single revenue segment, split into two delivery channels: direct platform revenue and integrated platform revenue. The segments are not separate products; they describe where customer activation happens and who controls the inventory and related channels. Direct platform revenue is revenue from activities run entirely on Zeta’s own platform and channels, which include Zeta’s email and messaging, Zeta’s own demand-side platform (DSP), connected TVs, and so forth. This is the higher-margin segment and accounts for 70-75% of revenue, which is the preferred revenue mix. Integrated platform revenue is when Zeta’s intelligence is activated inside third-party platforms such as Meta’s Facebook and Instagram, Google, TikTok, and other social and search platforms. In this segment, Zeta passes through most media costs to the customer, so margins are lower than on Zeta’s own platforms.

Figure 1: Segmented revenue

Customers tend to onboard through the integrated platform and use Zeta audiences and data to run better campaigns on third-party platforms like Instagram. As the relationship deepens, customers tend to expand into Zeta’s direct channels, like Zeta DSP, CTV, and so forth, which improves Zeta’s margin mix because third-party platforms no longer take a large cut of media spend.

The way money flows to Zeta is similar to a toll booth, a business model I tend to gravitate toward. A corporation, be it a retailer, insurer, or telecom company, runs its customer acquisition, expansion, or retention programs through Zeta. Zeta gets paid a subscription fee for the software and data, a usage fee that scales with volume, and revenue from the media it buys on the client’s behalf and books gross. However, unlike a true toll booth like a credit bureau or ratings agency, customers can avoid Zeta’s tolls because marketing budgets are often among the first line items a CFO cuts. Fighting for a marketing budget is a balancing act, but if Zeta can deliver attractive ROAS, the reasons a customer would cut spend thin out.

Zeta’s revenue sources are well diversified, with concentrations in consumer and retail (24% of revenue), Telecom (10% of revenue), Insurance, and Travel and hospitality (11% of revenue each). Per Zeta’s presentations, they serve 51% of the Fortune 100, and in these concentrated revenue categories, they serve either all the largest Fortune 100 companies or most of them. This creates stickiness as they integrate more deeply into top companies per category and start scaling.

One Zeta

The core driver of the business is the ability to scale its customers. After acquiring a new customer for one use case, the goal is to expand that customer's adoption of more of Zeta’s services. Management calls the strategy One Zeta, but in general terms, it is a land, expand, and scale strategy. The AthenaOS platform is positioned to reduce friction in adopting additional use cases, and expansion is necessary for several reasons.

Expanding customers feed the flywheel: more use cases mean better data and more learning, which improves the model and drives further expansion. More importantly, customers who adopt two or more use cases generate more than 3 times the annual revenue of single-use-case customers, with management citing up to 5 times more.

Zeta provides ample data points to track how customers scale throughout the business. The data helps investors understand two things: the sustainability of growth and how much it can accelerate.

First, Zeta splits its scaled customers into two buckets: regular scaled customers, who have recorded between $100 thousand and $1 million in revenue on a trailing-twelve-month basis, and super-scaled customers, who have recorded over $1 million over the past twelve months. The differences between the two cohorts are vast; a super-scaled customer records 17.4x the average revenue of a regular scaled customer.

That figure sounds massive, but it is a bit deceiving. The $100 thousand to $1 million cohort operates within a fixed range, whereas super-scaled customers have no ceiling. As such, a 17.4x figure on its own does not say much. What does tell us a lot is the trend in how that figure is improving. If ARPU for super-scaled customers increases every quarter, it means customers are getting more value and ROAS from their marketing budgets with Zeta, and they want to keep expanding. Unfortunately, the data is incomplete, as Zeta has not reported ARPU for scaled customers and scaled customers between $100k-$1mm since Q4 2025.

Figure 2: Cohort average revenue per user

While the ARPU difference is about 17 times larger for super-scaled customers, the average number of channels per cohort doesn't differ much. An average scaled customer had ~1.9 channels during the periods reported, while a super-scaled customer had about one additional channel. This is from a pool of mid-teens channels a customer can activate.

Figure 3: Cohort average number of channels

This tells us two things: so far, spending is not as heavily tied to the number of channels that are being adopted by a customer, and it also tells us that there is a lot of room for expansion. If ARPU was hitting a ceiling, and super-scaled customers on average adopted ~10 channels, then there would be a real reason to worry about future growth. As it stands, customers still have ample room for scaling. During the same Goldman Sachs Communicopia conference, management

The final piece of the puzzle is how much of the customer base is scaled, and to what degree. If all of Zeta’s customers were already scaled, there would be few pillars to support future growth. The figures support sustained growth, as customer growth has outpaced scaled customer growth since 2023. As of FY2025, 23% of total customers are scaled, of which 30% are super-scaled. That means that even with no new customer acquisition, the super-scaled customer base can grow 14x before the unscaled customer base is exhausted.

The rest of the research report is for paid subscribers only.

Please consider becoming a paid subscriber to help me do this full-time. That would allow me to deliver more frequent as well as higher-quality research.

Read more

One bad headline can turn the business on its head, and Meta has faced its fair share of bad headlines over its lifetime. The business has dealt with everything from class action lawsuits to regulatory and political pressure. Even at the time of writing, cases against Meta remain open, but it's becoming increasingly difficult to ignore what may be one of the major winners in the AI race.

Company profile

Theme: Personal AI, Direction: Buy

Symbol: META, Exchange: NASDAQ
Sector: Communication Services, Industry: Internet Content & Information
Fair intrinsic value: $942 (+27.3%), as of October 7, 2026
Market capitalization: $1 898 840 million
Pricing data: P/S 8.23x, P/E 27.88x

TYPEFCAPITAL.COM
Consider following me on X and YouTube.

This article is for informational purposes only and does not constitute investment advice or an offer to buy or sell securities.

Acquire your threats

Meta’s family of applications serves over half of the internet-connected humans on Earth, and just when investors think the TAM might be getting exhausted, Meta finds new switches to pull and keeps growing. It all comes down to one key factor that has stayed present since its founding over 20 years ago: execution.

Figure 1: Total revenue and revenue growth

The vast majority of Meta’s revenue comes from its advertising segment, which has accounted for over 95% of revenue since 2015. The advertising franchise came primarily from two acquisitions that it expanded well, along with its ability to copy rivals quickly while leveraging its network. Meta's core advantage from its early days was real identities. Users signed up using real names and connected with their real friends. Together with the Facebook News Feed, the company could produce a social graph and a stream of declared interests that no other platform of its era had access to. Next, it acquired two major threats before they grew large enough to pose a real threat: Instagram in 2012 for ~$1 billion, and WhatsApp in 2014 for ~$19 billion. Both were then scaled on Meta’s infrastructure, identity layer, and ad stack. Fast forward to today, and three of the four largest social and messaging applications outside of China sit on a single ad system run by Meta.

The core of the success comes from execution, and Meta’s record is unparalleled in the space. In 2012, Facebook became public and had close to no mobile monetization, with Mobile accounting for ~11% of ad revenue that same year. By 2018, Mobile accounted for ~92% of ad revenue, the result of one of the fastest platform transitions any company of its size has ever executed. Threats have emerged throughout its history, and when they couldn’t be bought, Meta copied them at scale. Snapchat’s Stories format was the first of its kind and quickly gained traction, but Instagram promptly introduced it and soon surpassed Snapchat’s user base. In 2020, Instagram launched Reels as the answer to the quickly growing video-sharing platform TikTok, which today carries a large and growing share of time spent inside apps among consumers.

Figure 2: Historical mobile advertising as a percentage of total advertising revenue

Every step of the way has brought thorns to Meta’s side, but it somehow emerges stronger from each setback. In 2021, Apple introduced its App Tracking Transparency (ATT), which meant that Meta lost signal and took a visible hit to its ads business in 2022. It then rebuilt how it targets and measures users using its own AI models, resulting in an even stronger product with less reliance on third-party data than before.

In essence, there are three factors that compound Meta’s dominance:

• The applications share one backbone and one ad system. Each addition to the system raises the value of all applications instead of splitting the audience and the attention.

• Meta has a dual-class share structure which gives Zuckerberg control. This means that the CEO can align the business for long-horizon bets without fear of public-market pressure or activist investors causing hindrances. Examples include what Meta paid for Instagram in 2012, which seemed controversial at the time. A present-day example includes AI-related capital expenditures.

• Reach. The distribution across Meta’s family of applications is incredibly vast, and reaches all parts of the world. For example, WhatsApp gives Meta distribution in emerging markets where it serves as the main communications layer among both consumers and businesses.

Competitors typically have one of these advantages, but Meta remains the only company with all of them at scale, which is also why the Family of Apps keeps taking share of the ad market despite already being the largest.

Advertising breakdown

Meta’s family of apps segment accounts for 99% of total revenue, of which 98% is advertising revenue. So how does the advertising machine work? In short, Meta runs auctions for ad impressions across its app feeds, Stories, Reels, Threads, and more recently, WhatsApp Status. Like many recent ad businesses I’ve written about (AppLovin, Google), an advertiser picks a sales, lead, app install, or awareness target, sets their budget and creative assets, and then Meta picks which users to show it to. The price an advertiser pays is set by competition in the auction process. Auction winners result from predictions of the probability that a specific user will take action when presented with the ad. This generates two key metrics for tracking the performance of the ads business: ad impressions, which are driven by users and their time spent on ads, and the average price per ad, driven by advertiser demand and how accurately the system predicts conversions.

Figure 3: Average price per ad growth and ad impression growth

2022 and 2023 price per ad growth shows just how Apple’s ATT cut the signals that Meta uses to measure conversions. The price per ad fell for seven straight quarters, and the business only grew by pushing more impressions through Reels at lower prices. By the end of 2023, average price-per-ad growth began recovering as Meta rebuilt the system with AI algorithms, and growth has stayed strong since, averaging above 10% per year.

Meta’s moat lives inside these two levers. Combined with the largest first-party engagement dataset on earth, Meta developed three distinct AI models that have elevated the whole segment. Andromeda is used for ad retrieval, and GEM for ranking and sequence learning. Meta Generative Recommender, which uses LLMs to reason about ad content and user preferences, is the differentiator across a field of many competitors. Better predictions raise conversion rates, which in turn raise return on ad spend for advertisers and, as a result, how much they are willing to bid in the auction. That self-reinforcing loop is fueling both ad impressions and average price per ad growth to stay above 10% simultaneously, and together it produces the 20-30% advertising revenue growth we’ve seen over the past year.

Figure 4: Advertising revenue and advertising revenue growth

So how big is the market? Using global ad market estimates by WPP Media, it has grown in the high single digits over the past 10-year period on average, reaching ~$1.15 trillion in 2025. Looking specifically at the digital ad market, it grew from ~$188 billion in 2016 to ~$835 billion in 2025, an 18% CAGR, including a massive COVID-19 rebound in 2021 that saw the market expand by 38%. Over that same 10-year span, Meta grew its share of the pure-play digital ad market from 14% to 24%, up 200 basis points from 2024 to 2025. It has taken the most market share over the past 10 years against a strong roster of competitors. Alphabet (Google), which is the largest seller of advertising, shrank its share from 42% in FY2016 to 35% in FY2025.

Figure 5: Implied Meta market share of global digital ad market

The rest of the research report is for paid subscribers only.

Please consider becoming a paid subscriber to help me do this full-time. That would allow me to deliver more frequent as well as higher-quality research.

Read more

Before yesterday’s news, why has FICO been trading down for more than a year? Since FHFA Director Pulte's appointment, he has waged a crusade against FICO, which I believe is a search for a scapegoat for how the housing market weakness is perceived.

Company profile

Symbol: FICO, Exchange: NYSE
Sector: Technology, Industry: Software - Application
Theme: High quality
Fair intrinsic value: $2206 (263%), as of September 30, 2026
Market capitalization: $13 777 million
Pricing data: P/S 5.8x, P/E 16.9x

Previous coverage:

TYPEFCAPITAL.COM
Consider following me on X and YouTube.

This article is for informational purposes only and does not constitute investment advice or an offer to buy or sell securities.

Recap of events

From 2020 through 2025, FICO increased the royalty per score pull by 725% cumulatively, making it ripe to frame FICO as evil in the public eye and win political points by targeting it. FICO has long enjoyed a monopoly among GSEs because of GSE eligibility requirements, and among private actors as they built financial ecosystems that relied on FICO. A summary of the timeline of announcements that have materially hurt the stock price is as follows:

Throughout my coverage of FICO, I have analyzed the narrative, rebuked it with data, and explained my perspective on what’s going on. In essence, I believe the following two things:

1. FICO’s price increases have no impact on the housing market booming or busting.

2. FICO was justified in increasing prices so aggressively.

The first point is the most important to clarify. FICO score pulls make up a microscopic part of mortgages. Rates, home prices, GSE fees, mortgage insurance, and closing costs are the real drivers of affordability, not FICO score pulls. Variables vary by state, lender, borrower, and other factors, but the overall consumer impact comes from loan production costs. Total production expense (TPE) also varies, and for a loan to be profitable, costs must be recouped through rates, direct up-front costs, or a mix of both. For example, if TPE is $10,000, $ 3,000 could be recouped through closing fees and the rest through rates, or it could all be through rates. The Mortgage Bankers Association (MBA) tracks production expense per loan, which can be used as a relative measure to see if FICO pricing increases affect the TPE per loan. It shows no correlation, since FICO Scores are a tiny input, and the data supports this. Cumulatively from 2020, TPE per loan increased by 45% in 2022, while FICO royalties still remained at $0.6 per pull. By 2025, TPE per loan was at 47%, while FICO had increased by 725%. In 2020, a FICO score pull was equivalent to 0.01% of TPE, and in 2025 after all the price increases, it was 0.04.

Figure 1: FICO royalty per score pull and total production expense per loan, cumulatively

I go into more detail on why FICO score royalties don’t materially impact mortgage borrowers as I dissected Senator Hawley’s oversight letter to FICO in this article.

Next, why FICO was justified in increasing its prices. FICO received an estimated 5% royalty per tri-merge report. Given how central the score is to lending decisions, FICO was severely underpriced. The markup from bureaus and resellers was estimated 1850% from 2020 through 2023, when FICO first raised its price. Even after a 725% increase, FICO’s share of a tri-merge report was 14%, with bureaus and resellers having a 641% markup. Credit data from the bureaus and reseller services also have costs associated with them, so this isn’t an exact markup estimate, but it shows a directional share distribution. Since the costs are bundled and appear as a single item, it all appears like FICO is making out like a bandit, something the FICO DLP program changes. FICO was underpriced relative to its value, and the limit of its pricing power is not economic, but political.

Figure 2: Estimated costs per credit score pull and markup from bureaus and resellers

I believe the FHFA’s Director frequently targets FICO because it is easy to point to score-royalty increases for the public, rather than discussing the real drivers of mortgage affordability.

Present-day impact

Two news headlines tanked FICO's stock price in a single day by an amount not seen in decades. Before we dive deeper, it helps to understand how FICO earns score revenue, who is involved, and how a loan moves.

The lenders typically don’t keep mortgages. Once they lend the money, they sell the loan promptly to get their cash back and use that capital to lend more. They gain on the sale of the loan in addition to fees, and often keep loan servicing rights, where they collect payments for a fee.

The loan can be sold to different parties depending on the loan type. Loans that conform to GSE rules go to Fannie or Freddie; GSE loans make up roughly half the market and must meet a specific loan-size limit to qualify. FHA (Federal Housing Administration) and VA (Department of Veterans Affairs) loans go to Ginnie Mae, and Jumbo and non-standard loans stay with the bank or are sold to private investors. Each buyer has its own rules, including which credit scores it accepts. Until 2025, GSEs accepted only Classic FICO scores, which the FHFA has now targeted.

Loans reach a lender through either retail, wholesale, or a correspondent. Retail is direct-to-consumer, where the borrower applies straight to the lender, and the lender pulls the credit. That’s about 57% of Rocket’s volume. In wholesale, the borrower goes to a broker, who submits the loan to a wholesale lender such as UWM or Rocket Pro, and the broker runs the application while the wholesale lender still underwrites, funds, and sells the loan. Correspondent loans consist of smaller lenders making the loan themselves, then selling it to a bigger aggregator, which sells it to GSEs. Each channel has a different answer on who chooses the score, which is why Rocket Mortgage explicitly stated that VantageScore 4.0 is the default for eligible loans.

When the GSEs started accepting VantageScore 4.0, lenders had no reason to switch because it would mean rebuilding their internal systems. Even though approved lenders could deliver VantageScore loans earlier this month, they had no incentive to do so as loans still paid a 20-point penalty on the GSE fee grid. Yesterday, the unified fee grid was released, and the 20-point penalty was removed. VantageScore averages ~13 points higher than Classic FICO for the same borrower, which puts the borrower in a cheaper price cell. Now, lenders have a reason to utilize VS, as cheaper loans win more borrowers.

This does not mean lenders use only VS or only FICO; using both increases the chances of underwriting a mortgage, which is in the lender's interest. They want to qualify as many borrowers as possible.

The problem

Rocket Mortgage did not choose VS4.0 as its default because it is cheaper than FICO scores; it is almost irrelevant. What matters is that VS4.0 often gives a higher number on the grid, which means lower fees. GSEs have to go along with it because they are not independent actors, and the FHFA director, Pulte, who is orchestrating this, appointed himself the chairman of both Fannie Mae and Freddie Mac’s boards.

Fannie and Freddie have been in government conservatorship since the 2008 mortgage crash. The single grid released is an FHFA policy decision, and its implications could cause problems for the economy as a whole. Lowering fees cuts the borrower’s costs immediately, without involving Congress or the Fed. It also suits the political narrative that FICO is the villain.

A unified grid invites gaming, where formulas are incentivized to qualify more people at higher grid brackets. As a result, the system introduces significant risk: a borrower may qualify for a loan using VS4.0 but not using Classic FICO. Credit losses won’t show up immediately, and if defaults escalate, taxpayers are the backstop. Using a grid that doesn’t differentiate or adjust for which score was used, the most optimistic score wins, not the most accurate one.

However, policy reversals can happen as quickly as they were introduced. GSEs set the fee grid themselves, and if VantageScore loans perform worse, they can act. If loans delivered on VS start defaulting worse than their price bucket implies, the offset between the formulas will likely be reintroduced. The FHFA also wants to IPO Fannie and Freddie, and giving away fees while taking on underpriced credit risk lowers the market value. Leadership changes at FHFA can also quickly reverse policies. The bureaus owning both the data and the scoring model may raise eyebrows for regulators as well, especially if per-loan optimization starts showing up in the data.

The FHFA essentially makes the current policy decision, not because VantageScore is better or because royalty prices matter. I believe it is because optics look good in the public eye if they are acting, and a good target is FICO, even though FICO’s impact on mortgage affordability is irrelevant.

Rocket’s decision to default to VS4.0 for eligible loans is a small hit to FICO on its own, but it highlights the real problem, which is the single pricing grid. Rocket will use VS 4.0 by default for direct-to-consumer loans sold to GSEs. However, it still uses FICO for FHA, jumbo, second-home, investment-property, and home-equity loans. Rocket Pro, its broker channel, still offers both and lets the broker choose. The decision can be reversed; it is just more favorable to default to VS4.0 currently, and a default doesn't mean it's exclusive either. An average increase in points doesn't mean it will always score higher than FICO on the grid, so it makes sense to pull FICO as well, which means FICO still gets paid.

FICO 10T is not approved for GSE delivery yet, only Classic FICO. When 10T lands, it might score more favorably on the grid, reversing Rocket’s decision and making VS4.0 obsolete.

FICO, the gold standard.

When lenders are free to choose a model, they have picked FICO for over 20 years. In credit cards, auto, and personal loans, no regulator is forcing FICO. VantageScore has been cheaper or even free for two decades, and still has a 1-5% share. The GSE shift isn’t lenders suddenly preferring VS, it’s the GSEs paying them to use it through the fee grid. That only lasts while the subsidy does. In addition, lenders often pull credit before the loan type is known, so whether a borrower ends up FHA, jumbo, conventional, etc. is often decided after the credit pull. So even though Rocket has chosen VS4.0 as its default, many still order FICO for most applicants.

Cards, auto, personal loans, FHA, jumbo, bank-held mortgages, second homes, investment properties, and home equity all stay on FICO. These models have been built around FICO for decades. I saw a comment stating that VS4.0 is just the beginning, and that more models are coming through AI and other incumbents. These suggestions severely undermine the ecosystem built specifically for FICO, even while competition was available for free. Portfolios would have to be re-rated, regulators would have to get involved, and all players would have to align. That is a very unlikely and costly scenario.

Even in the worst case, if VantageScore takes away not only half of GSE revenue from FICO but half of all score revenue, FICO is still undervalued at these levels ($1125 fair intrinsic value).

Intrinsic valuation model for subscribers:

Read more

Read more

However, it doesn’t have a moat; does it need one to be an attractive investment at these prices?

Company profile

Theme: AI advertising, Direction: Buy

Symbol: APP, Exchange: NASDAQ
Sector: Communication Services, Industry: Advertising Agencies
Fair intrinsic value: $390 (+26.53%), as of September 29, 2026
Market capitalization: $103 886 million
Pricing data: P/S 15.2x, P/E 23.6x

TYPEFCAPITAL.COM
Consider following me on X and YouTube.

This article is for informational purposes only and does not constitute investment advice or an offer to buy or sell securities.

Dominance without a moat

AppLovin has an implied 40% share of measured global gaming user acquisition spend in 2025 (AppsFlyer). But what does the company do, and how much growth is there left in the market for AppLovin to target?

The company sits on both sides of the mobile game advertising auction, similar to what Google does with its Ad-tech stack, where it runs the exchange while also bidding on the buy side. MAX, which is AppLovin’s mediation platform, is where game studios decide which ad network fills each ad slot through auctions. Axon is on the other end, a machine-learning ad engine that is also the most successful bidder in the auction. In essence, AppLovin operates the auction house and also serves as the most successful bidder.

Figure 1: Revenue and revenue growth

The main business is essentially three components. Axon Ads Manager is the demand-side engine where advertisers set a return-on-ad-spend (ROAS) goal. Axon then bids for users predicted to convert. AXON AI is the company’s proprietary model that scores whether a user will convert (install, pay, or buy what is being advertised) in microseconds, and then prices the bid. MAX, the supply-side mediation, runs a unified real-time auction across over 20 networks inside publisher apps. It currently has over a billion daily active users, most of whom are in casual games. These three components work together to match performance advertisers to mobile inventory with the goal of the advertiser’s value exceeding its ad cost.

In short, advertisers invest capital toward a ROAS target. AXON prices the impressions based on predicted conversion value, and then pays the publisher on a CPM-like basis, with AppLovin keeping the spread. The auction has several flywheels, and it is tied to what differentiates AppLovin.

Historically, AppLovin also owned apps, a segment that was divested in 2025. Owning the apps was crucial for the core business, since apps meant users, and users meant data. Gaming studios are not too keen on sharing their user data with third-party companies like AppLovin, so they had to find their own solution. They bought and operated several apps in order to train AXON AI, giving AppLovin a technology edge over other market participants. Once the initial training was complete, AppLovin divested the apps segment, making it a pure AI advertising business.

Figure 2: Segmented revenue

Now, more spending means more auction outcomes, which become training events and improve the model. A better model increases its win rate on MAX, improving ROAS and driving more ad spend. It is absolutely crucial that the model keeps compounding in competence, since barriers to entry are very low and there is essentially no moat besides a technological advantage that delivers better ROAS for advertisers. There are no switching costs in this business; ad spend can be placed on the platform at any time, and it can dry up quickly if ROAS declines.

The global gaming user acquisition market is not expected to grow fast. That poses a problem for AppLovin, since it is already approaching a majority stake. There are no clean third-party estimates available, but Newzoo forecasts global mobile games revenue to grow in the mid-single digits over the coming years. AppsFlyer tracks global gaming UA ad spend, which is roughly 22% of global mobile games revenue, based on historical figures. Using that as a base against my own forecast for AppLovin, they will reach majority market share in 2026.

AppLovin reports revenue net of publisher payouts, meaning gross gaming ad spend through AppLovin is larger than reported revenue. The company does not disclose the take rate, so we have to make our own assumptions to measure market share. I assume a 50% net take rate, which may differ from the actual rate. This essentially means gross gaming ad spend is double the reported revenue. With that in mind, we can look at AppsFlyer’s reported global gaming UA ad spend to measure market share. However, 2022 was rebased, making it impossible to clearly read progress. The only real like-for-like comparison is 2024 to 2025.

Figure 3: Implied AppLovin market share of global gaming UA spend

The rest of the research report is for paid subscribers only.

Please consider becoming a paid subscriber to help me do this full-time. That would allow me to deliver more frequent, higher-quality research.

Read more

Despite that, Alphabet is one of the best-positioned businesses in the world to benefit from and push AI into the next era.

Company profile

Theme: Full-stack AI, Direction: Buy

Symbol: GOOGL, Exchange: NASDAQ
Sector: Communication Services, Industry: Internet Content & Information
Fair intrinsic value: $345 (0%), as of September 18, 2026
Note: equity risk premiums and the risk-free rate are elevated at the time of valuation, heavily pressuring fair intrinsic value and upside.
Market capitalization: $4 275 285 million
Pricing data: P/S 9.59x, P/E 17.51x

TYPEFCAPITAL.COM
Consider following me on X and YouTube.

This article is for informational purposes only and does not constitute investment advice or an offer to buy or sell securities.

A diversified business

Google has been one of the AI winners in 2026, the basket of stocks that have accounted for a majority of S&P 500 returns in 2026. The S&P 500 index excluding AI-related stocks ($SPXXAI) returned less than half of the 11% gains of the overall S&P 500 in 2026, which is also about what stock has returned year-to-date.

Revenue growth has steadily been accelerating after the COVID boom and subsequent bust, largely driven by Google Cloud. However, there are a myriad of different lines of business.

Figure 1: Revenue and revenue growth

Alphabet’s advertising segment has been the core driver of the business, both historically and currently. In essence, it is a two-sided marketplace where advertisers pay to reach people, and Google, along with its partner sites and apps, supply the consumer. Through Google Ads, advertisers buy placements by choosing a goal, campaign type, targeting, creatives, budget, and bid strategy. Goals can range from building awareness to driving sales, leads, or app installs. Alphabet gets paid based on clicks (CPC), impressions (CPM), video views, or sales conversions.

Publishers owning websites, apps, and even YouTube creators sell their leftover space. On sites and apps, they sell space using AdSense, AdMob, and Google Ad Manager. Google then runs an auction, bills advertisers, and distributes a portion of the revenue to the publisher. As an example, Content sites keep roughly 68% of AdSense ad revenue, and YouTube creators keep ~55% of revenue generated from ads shown in videos.

Figure 2: Segmented advertising revenue

Advertising revenue is seasonal because businesses typically set budgets at a point in the fiscal year and distribute them across quarters, with a heavy lean toward the holiday season in Q4. As it has been the backbone of Alphabet’s whole business for as long as anyone can remember, most bear arguments target this segment. There is a fear that AI will disrupt search, which accounts for ~80% of total advertising revenue. The argument is that conversational AI will drive consumers away from querying the web.

If AI were eating away at search queries, Alphabet would need to defend its revenue through pricing. Fewer queries would need to be monetized even harder, since the segment has steadily grown despite the launch of ChatGPT and similar AI interfaces years back. The data shows otherwise: growth in paid clicks across Google’s properties had been decelerating, but is now showing the strongest growth since 2021. Volume is driving segment growth rather than pricing.

Figure 3: Google property paid clicks growth

Google’s own AI overview and AI Mode seem to be expanding the monetizable events. However, Alphabet is notably inconsistent in reporting AI-related figures, making it hard to gauge a general trend. AI Mode passed one billion MAUs globally in Q2 2026, which is up from over a hundred million MAUs in the U.S. and India a year prior. Unfortunately, while the increase sounds impressive at face value, the numbers aren't comparable, as the latter excludes a large swathe of users. In addition, the most common queries AI interfaces handle aren't the most monetizable, since they are conversational rather than directly actionable.

The advertising revenue market as a whole is set to accelerate, with projected global advertising revenue growth of 9% Y/Y in 2026 (up from 8% in 2025), per GroupM/WPP Media. In 2025, Google gained implied market share in the global advertising market compared to 2024, with a 27.3% implied market share versus 26.5% the prior year. However, among the big three global advertising players, Google is consistently losing market share. The composition of the big three is Amazon, Meta, and Google.

Figure 4: Google advertising market share of big three (Amazon, Meta, Google)

Alphabet stock is subject to volatility as the search business faces constant regulatory scrutiny. For example, Alphabet has arrangements with Apple to make Google the default search engine for its browsers, accounting for a large share of mobile queries. In addition, antitrust has targeted Google repeatedly, with a Google Chrome browser divestiture being on the table as recently as Q4 2025, and a recent case that closed just a few days ago relating to its ad tech.

Google owns the dominant tools on both sides of the market, meaning both advertisers and publishers. In the middle, the ad exchange (AdX) platform is used for auctions. Because of Google’s dominance, it was reported that Google took an outsized cut of every advertising dollar, starving digital publishers and news organizations of revenue. A coalition of states and the U.S. Department of Justice accused Google of illegally monopolizing the digital advertising technology market, asking for a ruling that would force Google to divest and sell off its AdX exchange. Judge Brinkema deemed a forced sale too disruptive, and the court instead opted for behavioral remedies.

Google's network advertising segment has declined since 2022, with low single-digit revenue loss each period. Network impressions fell 12% Y/Y in Q2 2026, and cost per impression increased by 13%. That is the opposite dynamic as witnessed on Google properties. The network is not a core engine, but still accounts for ~$30-32 billion every year for the past 5 years, about ~10% of total advertising revenue.

Figure 5: Google network cost-per-impression growth and impressions growth

Search queries and page loads trigger real-time auctions that finish in milliseconds. Google doesn’t choose a winner based on the highest bid. Instead, Google computes an Ad Rank from the eligible ads based on factors that include the bid, the quality of the ad and its landing page, how competitive the auction is, the search context, which includes the user’s device, location, and time, as well as the expected lift from extensions like call buttons, prices, and site links. Google generates a quality score from expected click-through rates, ad relevance, and landing page experience, then uses it in the auction. A more relevant and useful ad can outrank a higher bidder and usually pays less per click. The price isn't an advertiser’s full max bid; instead, it tends to pay just slightly above the second-place advertiser.

Campaign types include tightly controlled Search ads with specific parameters set and AI-driven campaigns that spread the advertiser’s budget across Search, YouTube, Gmail, Maps, and other platforms, letting Google choose placements and bids.

While the backbone of the business remains advertising, the excitement in the markets is surrounding the AI narrative, and for good reason. Growth of the segment has accelerated significantly in recent periods, recording 82% Y/Y growth in Q2 2026.

The rest of the research report is for paid subscribers only.

Please consider becoming a paid subscriber to help me do this full-time. That would allow me to deliver more frequent, higher-quality research.

Read more

Investors can use screeners, read about companies in small, dedicated niche communities, or find a piece of news that gives them a whiff of one of these hidden gems. In today’s short experiment, I will present four different companies with a brief description of what they do, some financial highlights, and then ask you to pick which of these four companies you would be interested in investing in or diving deeper into.

All the data is real, and I will reveal the companies at the end. You can be boring and scroll to the end, or challenge yourself by reading this article in full to play along and choose your favorite of the four businesses.

Each business mentioned below is profitable (cash flow positive, positive operating and net margins) and ranges from mid-cap to large-cap.

Business 1, Government analytics software

This analytics software business operates within government agencies outside the U.S., is growing at an accelerated pace, and continues to expand into more countries. Contract awards vary in size, and it can quickly offer solutions for specific verticals as needed, showing both the adaptability of its software deployment and how opportunistic the company is.

An example is a contract with the UK’s Department for Environment, Food & Rural Affairs (DEFRA). This company was awarded a contract for "Trade Analytics Solutions, where the company’s software was used to model goods trade across the customs border after BREXIT.

The company’s reach is worldwide, and it has won contracts in the EU, Asia, and Australia. In Australia, the Australian Transaction Reports and Analysis Center (AUSTRAC) uses the company’s software for a variety of use cases, including money laundering and financial criminal investigations.

Business 1, Figure 1: Revenue over the trailing twelve months, in Euro

Business 2, MRP software

The business has seen massive growth, accelerated by AI. Use cases include modeling suppliers, bills of materials (BOMs), work orders, lead times, and parts. The software brings together previously siloed data so manufacturing and engineering can work seamlessly and efficiently with the supply chain. Using predictive analytics, the software can model different demand scenarios and algorithmically pinpoint actions to take with high impact in what would otherwise be like finding a needle in a haystack.

A customer testimonial reads that their master scheduler runs demand scenarios against live models of BOMs and supplier lead times, with up to 10x performance improvements over internal solutions. What’s most impressive is that the customer had a fully working solution within roughly six months of signing up for the software.

The growth for the business is not about to slow down anytime soon, with the business seeing a large backlog and strong demand.

Business 2, Figure 1: Revenue, in Japanese Yen

Business 3, Manufacturing OS for the U.S. government and primes

Manufacturing output has long been a segment where the U.S. has lagged behind adversaries such as China, which exceeds the U.S., Japan, and Germany combined, per the World Bank. While U.S. spending is far bigger, the physical industrial output is, by some estimates, up to two orders of magnitude larger in China. In shipbuilding, cited Office of Naval Intelligence estimates put Chinese shipbuilding capacity at about 232 times larger than that of the U.S. The output discrepancy also exists for munitions, drones, and other physical output.

Re-industrialization has been a major topic in recent times, highlighted by both the U.S. government and prominent voices within the defense industry. This business capitalizes on that by enabling real-time insights and operational efficiencies within manufacturing, which is required to start scaling the American industrial base. The company leverages its software to help identify and resolve scheduling conflicts, integration issues, equipment shortfalls, and ties legacy systems to enable seamless production planning.

The business sees tailwinds tied to both the need for re-industrialization and the AI wave.

Business 3, Figure 1: Revenue growth

Business 4, AI model augmented generation

This business has built a platform where their customers can build and deploy AI LLM’s in a safe environment. The platform is model-agnostic, and regardless of which LLM the customers use, the models can be improved by providing superior context. The models are not retrained, which would be costly. Instead, through chunking and embedding, object-relationship retrieval, and indexing of the customer’s entire business, LLM performance can be far more efficient and purposeful because it references real context from the customer’s business.

We have seen many recent examples of IP theft from AI labs, where they use context from their own customers to then launch competing products or advance other discoveries which were not originally in their domain. By deploying on this business’s platform, the customer’s data is isolated, so that neither the model providers nor the business retain the transmitted data; the customer keeps their IP.

This business has grown massively in a short amount of time, and the runway and necessity of the solution see tailwinds for a long time to come.

Business 4, Figure 1: Total commercial customers

Out of these businesses, which seems the most lucrative, interesting, and most importantly, which business has the highest potential?

Well, I have good news. You don’t have to make a decision among the four proposed businesses; they’re all the same company.

^{Business 1: Palantir’s international government business example
Business 2: Palantir’s U.S. commercial business example
Business 3: Palantir’s Warpspeed business example
Business 4: Palantir’s Ontology-Augmented Generation business example}

Palantir has virtually no limits on which industry or sector it can serve. The mentioned examples are vastly different, seemingly four different companies, and yet they are just a tiny fraction of what Palantir software is capable of.

Investors are obsessed with finding the next 100-bagger, so much so that they tend to ignore generational companies that have had a large run-up in stock price, despite the potential for much further upside. Palantir hasn't done a 100x yet, but it seems to have a limitless TAM, is incredibly sticky, has a scalable business, and has immense demand. Keep chasing the next 100x; I know I will, but don’t ignore the potential of companies you already know about, because who knows, you might already have the next 100-bagger in your watchlist.

The final piece to counter that narrative is that the “next X” typically fizzles out and has historically been a weaker business, such as Lucid Motors, which is now teetering on the brink of bankruptcy. Rocket Lab is a credible company with high launch success rates, a clear path to profitability and scalability, and a clear path to business expansion.

Company profile

Theme: Space, Direction: Hold

Symbol: RKLB, Exchange: NASDAQ
Sector: Industrials, Industry: Aerospace & Defense
Fair intrinsic value: $28.1 (-56%), as of September 5, 2026
Market capitalization: $40 463 million
Pricing data: P/S 53x, P/E N/A

TYPEFCAPITAL.COM
Consider following me on X and YouTube.

This article is for informational purposes only and does not constitute investment advice or an offer to buy or sell securities.

The full stack

Rocket Lab, together with SpaceX, is one of only two companies that can design a satellite, build it, launch it on their own rockets, and operate it once it is in orbit. As Rocket Lab describes itself, they are an end-to-end space company. They manufacture everything from solar arrays and star trackers to laser optical terminals and robotic arms. It is a prime contractor for the Space Development Agency (SDA) and the Space Force.

The company manufactures satellites, launches them on its own rockets, and then operates them for its customers. To own the chain end-to-end, it also plans to own and operate its own constellation.

Rocket Lab reports two primary segments with different economics. The first segment is launch services, comprising launches of Electron, HASTE, and, soon, Neutron. The other segment is space systems, consisting of spacecraft and constellation programs, components and subsystems (solar cells and arrays, star trackers, radios, etc.), and related software, services, and payloads.

Figure 1: Segmented revenue

Rocket Lab’s advantage in the industry lies in time and availability. SpaceX might be a far cheaper option in terms of launching a customer’s payload, but it would happen on an orbital plane and date of SpaceX’s choosing. Capacity scarcity in this industry allows Rocket Lab to exercise pricing power. Customers pay for orbit choice, date control, and an established track record, rather than the cheapest kilogram. The broader launch market has gotten cheaper over time, but Rocket Lab’s average revenue per launch has been rising.

What matters in the industry is operating record. Rocket Lab has a 100% mission success rate over the past 3 years, and a 96% cumulative mission success rate as of Q2 2026. Through Q2 2026, Rocket Lab has flown 93 times, with a total of 4 failures. The average time between launches has fallen dramatically, from 61 days in 2021 to a mere 15 as of Q2 2026. Electron was the second most frequently launched orbital rocket globally, only behind SpaceX’s Falcon 9.

Figure 2: Segmented launch missions

In Q3 2021, Rocket Lab lost $9.2 million on a launch. In Q2 2026, they generated $4.7 million in gross profit per launch. This is because fixed costs for manufacturing overhead and direct labor remain the same whether Rocket Lab launches 10 or 20 vehicles.

Another part of the story is more efficient manufacturing: increased automation and better yield over time, less scrap, less rework, in-house structuring, and better supplier lots as volume became predictable. Flying more often on the same infrastructure, charging more as credibility increased, and scarcity for small dedicated launches increasing, and more flights being done for the government.

Figure 3: Cost and revenue per launch

Lastly, the mix of spacecraft launched also affects the segment's gross profit. Currently, Rocket Lab flies Electron and HASTE. Electron is an orbital rocket that is built to give small spacecraft their own ride to low Earth orbit instead of hitchhiking on a larger rocket. Typical uses include:

• Earth-observation and communications satellites for companies like BlackSky, Planet, Synspective, and Capella.

• Intelligence and National-security payloads for Space Force, NASA, DARPA, and NRO.

• Both dedicated and rideshare missions of up to 300 kg to low Earth orbit.

Typically, a customer will reach out to Rocket Lab to get their satellites into an orbit of their choosing, on a date they request. Because Rocket Lab lets customers control the schedule and destination, it can charge a premium over SpaceX flights.

HASTE, or Hypersonic Accelerator Suborbital Test Electron, is a modified Electron that stays suborbital. Since it doesn’t need to reach orbit, it can carry larger payloads (up to ~700 kg) and can push the payload to anywhere between Mach 7 and Mach 20 (or more). Typical use cases for HASTE include:

• Hypersonic weapons and vehicle flight tests.

• Interceptor and missile-defense experiments, such as Golden Dome-related flights.

• Earth re-entry technology testing.

• Guidance and aerothermal tests in high-speed atmospheres when wind tunnel simulations are not enough.

Essentially, the Pentagon needed frequent and cheap hypersonic test shots. Legacy testers are slow to book and a lot more expensive per shot. Because HASTE uses the same factory, engine, and launch pads as Electron, Rocket Lab can fly tests more frequently and at a lower cost than legacy testers. As such, most missions relate to the U.S. government and defense work, with many payloads remaining classified.

Calculating the average revenue per launch for quarters consisting only of Electron launches gives a launch-weighted average baseline of $7.49 million/launch. Backing out the implied HASTE revenue per launch for quarters where HASTE launches are part of the mix gives an uplift in revenue/launch of up to 145%, and as low as 35% from 2024. From Q1 2024 to Q2 2026, a regression cross-check implies that Electron’s revenue/launch is $7.86 million, and $11.86 million for HASTE, implying an uplift of ~51% per HASTE launch.

Figure 4: HASTE implied uplift per launch

Rocket Lab does not publish ASPs for either vehicle. In essence, ASPs vary with orbit, payload, launch site, schedule urgency, and volume discounts. Compared with contracted defense data, HASTE’s implied $11.86 million per launch is roughly in line. The Space Force award implies $22 million per mission at twelve launches, and ~$15 million if all eighteen fly. The Anduril contract implies $10 million using the same arithmetic.

Neutron is the next stage in Rocket Lab’s arsenal that provides vastly different capabilities to Electron. It is a much larger rocket and can carry up to 13,000 kg on a reusable basis, compared to Electron’s ~300 kg and limited reuse. It has not flown yet and is targeted for late 2026, but it has already been pushed back multiple times, making a 2027 debut not unlikely.

While Electron targets niche customers willing to overpay for a dedicated ride, Neutron can target a much larger swath of the market: mega-constellations, SDA and Space Force-class sats, national security launches, and customers competing with Starlink. Essentially, it would become a second U.S. medium-lift option alongside SpaceX. It is also the next step toward operating large constellations through their own vehicles, which ties into the Iridium acquisition target. More on that shortly.

Financially, Neutron is expected to have an ASP roughly 6 times larger than Electron’s at ~$50-55 million. Even with a slow ramp in flights, such as ~7-10 in the first 3 years, it would mean hundreds of millions in launch revenue. However, part of Rocket Lab’s backlog already assumes Neutron’s existence, since some space-systems awards include it as an option on the contracts.

Figure 5: Cumulative mission success rate and average days between launches

Space systems is the larger and faster-growing segment. It includes complete spacecraft, individual satellite parts, and related services. Most revenue comes from products, with a smaller share from services. Rocket Lab can sell a complete design, build, launch, and operate package by pairing the space systems segment with launch services.

There are essentially two businesses inside of the space systems segment. First are merchant components, including solar, reaction wheels, star trackers, software, separation systems, and optical terminals and mechanisms. In a sense, it is a pick-and-shovel revenue stream.

The other part of the business pertains to satellite platforms and full spacecraft, including the SDA transport and tracking layer, GEO domain-awareness satellites, flatellite, and photon-class buses. These are typically large, multi-year contracts delivered over time, at lower margins than the merchant components. Most of the backlog is tied to the satellite platforms and spacecraft part of the segment, and as such, can bring down overall margin as the mix increases.

Figure 6: Space systems revenue recognition

The real value is providing the full stack, end to end. SDA Tranches II and III (goal of building, scaling, and upgrading a resilient LEO satellite constellation for the U.S. military) are worth more than $1.3 billion. Sourcing SADAs (Solar Array Drive Assemblies), gimbals, sensors, optical comms, and other parts in-house raises the captured margin on produced satellites and lets Rocket Lab bid without tracking supplier margins. This allows the company to undercut on both price and schedule, compounding platform wins. Rocket Lab winning SDA Tranche III as prime was tied to owning both bus and payload. Value for the segment lies in winning future tranches, as well as Golden Dome ($1.2 trillion over 20 years, of which $723 is the space-based interceptor layer), and other SDA-like contracts.

Rocket Lab also supplies its own competitors through the Space Systems segment, including other primes competing for the same prime contracts. However, components must be qualified, and once qualified into a satellite bus, they become difficult to displace. That creates a long-lasting, sticky, high-margin revenue stream. As an example, Rocket Lab notes that its hardware and spacecraft have flown on over 1,800 missions over a twenty-year period.

Both operational segments tie together and create more value as a sum, rather than individually. Designing, manufacturing, and launching as one increases win rates on national-security primes and captures margin that would otherwise leak to suppliers and third-party launch suppliers, but the trade-off is concentration risk. A delay on Neutron, or a cost overrun on SDA, could hit the business across seemingly unrelated operations.

Figure 7: Segmented service and product revenue

In essence, Rocket Lab's current launch value is driven by flights, cost per flight, and Neutron's reusability capabilities. Space systems derive value from component margin, manufacturing efficiency, and the duration and durability of government platform wins.

Iridium acquisition

The rest of the research report is for paid subscribers only.

Please consider becoming a paid subscriber to help me do this full-time. That would allow me to deliver more frequent, higher-quality research.

Read more

The other emerging fear among investors is that VantageScore 4.0 is disrupting FICO’s gold standard, which no financial data seems to support. When the market comes around to that realization, the upward correction in the stock quote may be violent, as implied in recent price volatility.

Company profile

August 25, 2026 Follow-up coverage
Direction: Buy
Previous fair intrinsic value: $2116, as of May 2, 2026

Symbol: FICO, Exchange: NYSE
Sector: Technology, Industry: Software - Application
Theme: High quality
Fair intrinsic value: $2269 (98%), as of August 25, 2026
Market capitalization: $26 055 million
Pricing data: P/S 10.9x, P/E 32x

Previous coverage:

TYPEFCAPITAL.COM
Consider following me on X and YouTube.

This article is for informational purposes only and does not constitute investment advice or an offer to buy or sell securities.

Data points to the market being wrong

Uncertainty tends to unequivocally have a downside impact on the stock quote of the targeted company, something evident in the sharp decline in FICO’s stock price. However, the market seems undecided because the company is delivering stellar results each passing quarter, suggesting no disruption is taking place. As a reminder, here is a policy timeline that has led us to today’s status, with bolded events causing direct crashes in FICO’s stock price.

In recent coverage, I have covered a response refuting each point raised by Josh Hawley, as well as discussed FICO’s status as the gold standard and why I believe it won’t be disrupted. Today I want to focus on the actual data from calendar quarter 2’s financial results for the publicly traded bureaus and FICO.

First things first, VantageScore has seen a genuine increase in usage. It appeared in fewer than 5% of TransUnion’s mortgage credit inquiries at the start of the year, to reaching almost 30% by Q2 across more than 900 lenders. Such rapid adoption at face value may seem like disruption of FICO’s gold standard being imminent, but what the headline does not cover is that VantageScore is being sent along for free with FICO score pulls. Meaning, it is pulled alongside FICO, not instead of FICO.

CEO Lansing also notes on the call that measuring market share is difficult, and a first instinct as an analyst is to note the spread between credit score-related revenue among FICO and the bureaus. However, Equifax and TransUnion resell FICO scores after buying them at FICO’s wholesale royalty. They mark it up and book the gross amount as their own revenue. When FICO increases its price, it inflates the growth contribution to the credit-file revenue at both bureaus. Both bureaus have begun disclosing commentary and numbers excluding FICO, and it shows that 50-75% of their mortgage revenue growth is FICO’s pricing power flowing through, at no profit for the bureaus.

Figure 1: Equifax USIS mortgage revenue growth mix (decomposition)

Note: CQ2 data assumes mid-point of 5% per Equifax Q2 earnings call: “USIS mortgage revenue was up 40% and up mid-single digits excluding FICO, with hard mortgage inquiries up only 1%.”

Equifax USIS Mortgage revenue and TransUnion’s mortgage revenue include the bureaus’ own products, which means that the ex-FICO revenue growth could come from sources other than VantageScore 4.0 adoption.

As a reference point, FICO does not disclose exact mortgage score volume growth, but noted in the Q3 earnings call that it grew “low single digit”. Against that backdrop, FICO still grew its mortgage originations revenue by 97% Y/Y, signalling that it still has immense pricing power. VantageScore has waited years for permission, but it turns out to be worth surprisingly little, and that is simply because of who pays for the score pull.

Figure 2: FICO Mortgage originations revenue growth

The marginal cost of pulling a score is passed to the borrower, meaning they don’t feel the impact of exercised pricing power. What they do feel, however, is the risk of a loan failing to qualify, which turns into scores becoming gamed. Lenders pull a free VantageScore next to a paid FICO because it is available, and pulling both means that the lender increases the chances of qualifying a borrower. The data shows that almost no one seems to be paying for VantageScore 4.0.

The CHLA puts the all-in credit report cost at an average of $540 in 2026, and they assume a $10 flat FICO take at a tri-merge rate of $30; it means that bureaus and resellers charge 17 times the FICO take, a 1600% markup. The $540 figure is up from ~$50 in 2022, branding FICO as the villain due to FICO’s price increases.

From another angle, FICO’s DLP performance model that charges a $33 funding fee for classic FICO, or $65 for FICO 10T, reflects a 50-90% reduction in average per-score fees compared to what resellers paid for FICO scores in 2025. This takes the markup out of the bureaus’ hands and lowers the all-in cost to lenders, and simultaneously defuses the political backlash FICO has been facing without cutting FICO’s own take.

The data suggests that FICO is capturing more mortgage dollars, not less, and it is accelerating rapidly. No data point that I can find supports the narrative that FICO is losing share to VantageScore.

Figure 3: Implied U.S. mortgage originations revenue share, TTM

It also seems that the fears of FICO being regulated further may be easing, as the FHFA Director Pulte signals that he is satisfied with recent developments in the credit scoring market.

FICO and Vantage Score are both doing very well. I appreciate the leadership of each company. Likely to have some new news on each soon. Exciting times ahead for the consumer.
- FHFA Director Bill Pulte on X, 21 August, 2026

Business developments

The rest of the research report is for paid subscribers only.

Please consider becoming a paid subscriber to help me do this full-time. That would allow me to deliver more frequent as well as higher-quality research.

Read more

Read more

Company profile

Theme: SaaS, Direction: Hold

Symbol: NOW, Exchange: NASDAQ
Sector: Technology, Industry: Software - Application
Fair intrinsic value: $112.6 (-10%), as of August 19, 2026
Market capitalization: $130 776 million
Pricing data: P/S 8.9x, P/E 78.3x

TYPEFCAPITAL.COM
Consider following me on X and YouTube.

This article is for informational purposes only and does not constitute investment advice or an offer to buy or sell securities.

The ever-present AI fears in the market

In recent times, the SaaS darling of the market has not been spared from the cannon of AI disruption fears. However, none of the reported financial metrics suggest that ServiceNow is being disrupted by AI.

Figure 1: Revenue and revenue growth, TTM

There are two angles to AI disruption fears: one to the business itself, and one to the business model. Anthropic sparked a majority of fears in the market when it started shipping products across a wide variety of categories, including creative and visual design, cybersecurity, and enterprise productivity and automation. The notion is that AI will be able to create software products within any given vertical and disrupt existing software companies. In regard to targeting the business model, the fear is that AI agents will take over seats in a seat-based SaaS, and with that, the revenue associated with the seat will be lost.

In the ServiceNow context, if ServiceNow only offers seat-based SaaS contracts, then agentic AI would be akin to an extinction event for the business, and that seems to be what the market has begun to fear for the company. Both volume and price would be under pressure, but such a scenario is not materializing. Instead, ServiceNow has embraced AI and integrated AI capabilities across its suite of products, which is resulting in increased volume and pricing for the offerings instead. The approach is a hybrid pricing structure of both seat-based pricing for predictability, as well as a consumption-based model for flexibility.

ServiceNow reports some of the most consistent financial figures that I have seen in a business, making it easy to spot any kind of disruption across the business. Renewal rates have stayed consistent, annual contract value (ACV) expansion across customers continues to accelerate through peak AI fear, and even the geographical mix is undisrupted.

Figure 2: Renewal rate

One chart that may seem alarming at first is the number of customers with ACV’s exceeding $1 million. The growth rate has steadily been dropping until it ultimately ceased to be reported on altogether, just as the growth rate was about to drop below 10% Y/Y

Figure 3: Customers with ACV exceeding $1 million

What replaced the metric is customers with an ACV exceeding $5 million, and that cohort of customers has been accelerating over the past year, through peak AI disruption fears.

Figure 4: Customers with ACV exceeding $5 million

Among the cohort of customers with an ACV exceeding $5 million, the average ACV per customer has steadily been increasing throughout the AI fear narrative.

Figure 5: Average ACV of customers with an ACV exceeding $5 million

To recap, AI disruption would show up in both volume and pricing, ultimately impacting growth. What we are seeing is consistent and healthy renewals, accelerating revenue growth, and expanding ACV both in terms of average ACV as well as the number of customers signing large contracts. ServiceNow’s own AI solutions are seeing rapid adoption, with a majority signing consumption-based pricing, and customers spending over $1 million of ACV on ServiceNow’s AI solution grew 130% Y/Y in Q1 2026, signaling expansion rather than contraction. If AI agents were displacing ServiceNow’s solutions, the outlook would look quite different.

ServiceNow’s software

ServiceNow and its “Now Platform” is originally an IT service management (ITSM) system covering the core incident, problem, and change service requests. Since then, it has expanded into covering wider parts of an enterprise’s operations, including:

• IT operations and asset management.

• HR and employee service delivery.

• Customer service and CRM workflows.

• Security, risk, and compliance.

• Low-code application development.

• Generative AI and agent platform.

These types of workflows and platforms are described in a myriad of acronyms: ITSM, ITOM, ITAM, CMDB, and so forth. However, it’s the last one, the configuration management database (CMDB), that matters competitively. A CMDB is a centralized repository that stores the enterprise’s IT assets and infrastructure components, as well as their relationships and dependencies. Configuration Items (CIs) that populate the CMDB include hardware, software, virtual resources, business services, and virtually any asset within the enterprise.

The purpose of a CMDB is to act as a single source of truth for the whole IT environment, which enables teams to understand the infrastructure of the business and its services. For example, the team can see that a specific application runs across these particular servers, in this data center, dependant of this database, which is maintained by that team, and during an outage, these applications and services will be impacted, which affects these teams, and so forth.

Understanding the importance of a CMDB is also a key to understanding ServiceNow’s competitiveness in the market, as well as its moat. Because of the CMDB serving as a single source of truth for the whole enterprise, the switching costs become massive as years of accumulated populating and mapping of a CMDB would be lost, quickly turning into an operational nightmare. It also helps ServiceNow to expand within existing accounts and sell more platform capabilities, as each module makes the entire platform deployment more valuable and more efficient. For example, if an enterprise’s security team purchases ServiceNow’s Security Operations module, they already have a CMDB dependency map to rely on to see which services, servers, and applications are vulnerable to threats.

The business

The revenue can be divided into two disaggregated revenue lines: subscription revenue, which encompasses the software itself, and a professional services segment, which is a small and limited part of the business. The professional services segment accounts for a mere ~3% of total revenue, which is on purpose as ServiceNow relies on a broad network of implementation partners. By outsourcing a majority of implementation, configuration, advisory, and training to partners such as Accenture, Deloitte, EY, KPMG, and others, ServiceNow can keep reporting high software-like margins and scalability. The professional services revenue is recognized as the work is performed, unlike the ratable recognition of subscription revenue.

Another reason to push professional services to partners is that a large consulting ecosystem gets built, and all of the world’s top consultancies have a reason to keep recommending and pushing the platform. In essence, it can be seen as a customer-acquisition infrastructure.

The rest of the research report is for paid subscribers only.

Please consider becoming a paid subscriber to help me do this full-time. That would allow me to deliver more frequent as well as higher-quality research.

Read more

The market can be quite generous at times to those who strive to understand businesses, and Robinhood at the time of writing is one such generous gift. What Robinhood has been busy doing has escaped surface-level investors, seeing the company as a mere brokerage that keeps adding products. The truth is a lot deeper, and a lot more exciting.

Company profile

August 10 2026 Follow-up coverage
Direction: Buy
Previous fair intrinsic value: $155.29, as of May 18, 2026

Symbol: HOOD, Exchange: NASDAQ
Sector: Financial Services, Industry: Capital Markets
Theme: Growth
Fair intrinsic value: $160.99 (73%), as of August 10, 2026
Market capitalization: $85 080 million
Pricing data: P/S 19x, P/E 45x

Previous coverage:

TYPEFCAPITAL.COM
Consider following me on X and YouTube.

This article is for informational purposes only and does not constitute investment advice or an offer to buy or sell securities.

The rotation in transaction revenue

As we are all aware, Robinhood is still a cyclical business and still primarily relies on customer activity in the markets to generate a majority of its revenue and profits. Q1 was a particularly weak quarter for Robinhood, and it traded down substantially after posting a mere 15% Y/Y revenue increase, after growing 100% just two quarters prior. However, Q2 sees a reacceleration to record revenue, and Robinhood grew its total revenue by 32% Y/Y, more than double the rate of Q1.

Figure 1: Total revenue

Over the past quarters, Robinhood has been on a systematic campaign to own the venue layer rather than just routing through third parties. MIAXdx was acquired in early 2026, which turned into Rothera, a CFTC-licensed exchange and clearinghouse through a joint venture with Susquehanna. Bitstamp became an owned cryptocurrency exchange, while Robinhood Chain became an owned settlement layer. Robinhood Banking is an owned deposit relationship in place of a sweep, and TradePMR and the related advisor network are an owned custody channel. Most recently, Robinhood Ventures became an owned issuance vehicle. We will cover all of these and their relevancy to the future of Robinhood in this report.

As most are aware, Robinhood discloses monthly results for two of three months in a quarter, which provides a lot of insight into what the earnings results might look like. I believe the primary cause for the stock quote’s recent collapse is the drop in cryptocurrency volumes. For a long time, crypto and options have represented the primary drivers of revenue and profits for the whole business, which is why a drop in activity can cause a lot of concern. However, crypto has been largely replaced by event contracts, an avenue with more impressive growth and way better take-rates.

Figure 2: Cryptocurrencies and event contracts yield

Robinhood entered prediction markets in Q4 of 2024 as a distributor by partnering up with ForecastEx, and then scaled the business from Q1 2025 with Kalshi. Robinhood provided the distribution channel, and Kalshi provided the platform; a partnership that quickly became a strategic vulnerability. By Q2 2025, more than half of all Kalshi volume was routed through Robinhood. As a response, in Q1 of 2026, Robinhood acquired MIAXdx, a CFTC-licensed designated contract market, derivatives clearing organisation, and a swap execution facility. Instead of going through the hoops of regulatory applications, Robinhood simply acquired it, which enabled the launch of Rothera in a short amount of time. Within a single month from its launch, Rothera became a top-three designated contract market.

Figure 3: Robinhood volume and share of Kalshi event contracts

Of a total of $156 million from event contracts, $139 million was still Kalshi-routed and $17 million went through Rothera. Rothera is not controlled by Robinhood, so the economics accrue partly outside of consolidated revenue since it goes through the joint venture and non-controlling interests. The yield fell 10 basis points to 1.1%, hinting at the unit economics not necessarily being more favorable through Rothera. However, only 15% of total event contracts volume went through Rothera, making it too early to draw conclusions. Robinhood’s estimated market share of the prediction market is ~12%, down from its peak in Q3 2025 of 26%.

However, Robinhood does not participate in the wider categories of predictions that more open platforms do, and as such, will keep losing share. The upside of Robinhood’s strategy of only offering more serious prediction markets is that they will have far fewer regulatory issues down the line. Prediction markets are quickly becoming more and more obscure, with contracts being offered for events that are subject to heavy manipulation, and some are downright luck-based, such as predicting the number of cars that will pass a crossing in an allotted timeframe.

Figure 4: Prediction markets market share

There are now ~2 million customers cumulatively that have used prediction markets on the platform, up from 1.5 million in Q1 2026. The increase is likely due to the FIFA World Cup taking place in Q2, and as an event-driven segment by construction, active events matter a lot for growth. In Q3, the football season will be big, and in Q4, the U.S. midterm elections will play a big role in growth.

The headline could be read as Robinhood swapping a 0.25% take-rate asset for a 1.1% take-rate asset without having to win a single new customer. However, it’s not like Robinhood dropped the ball on cryptocurrencies. Total cryptocurrency exchange volume has fallen off a cliff since Q4 2024, and declined 37% Y/Y in Q2 2026. On the contrary, Robinhood nearly doubled its market share of total cryptocurrency exchange volume Y/Y, from 0.8% to 1.5%. It is not Robinhood underperforming so much as the asset class having fallen out of favor, as is the reality of a cyclical asset.

Figure 5: Total cryptocurrency exchange volume

Speaking of market share, Robinhood is on the rise in the segment that arguably matters the most: options. Options is the segment that offers the most spread, and as such, is the most profitable. Robinhood now has an 8.6% market share of total U.S. options volume, an all-time high. In addition, Robinhood grew its equity market share of U.S. equities notional volume to 1.4%, another all-time high, eclipsing even the meme-mania bubble of 2021 of 1.3%.

Figure 6: Segmented market shares

Q1 2026 showed signs of genuine cyclical pain, only growing transaction-based revenue by 7% Y/Y. However, Q2 is a massive rebound, recording 44% Y/Y growth, primarily driven by event contracts, a segment where Robinhood is gearing up to be a major player for a long time to come.

Further diversification, the most important driver

The rest of the research report is for paid subscribers only.

Please consider becoming a paid subscriber to help me do this full-time. That would allow me to deliver more frequent as well as higher-quality research.

Read more

Palantir has always championed the concept of creating value through its software, and then charging its customers a portion of that value. In a sense, that concept is infinitely scalable: Create $100 million for a business, charge them $30 million for the software. Create $10 billion for a business, charge them $3 billion.

What has been transpiring outside of the Palantir context has been the opposite: charge customers for tokens, without caring about the value the token spend results in. The term associated with this is “tokenmaxxing”, which CEO Karp has mentioned in relation to how the AI labs conduct their business.

In addition, the results have once again surpassed analyst expectations, my own expectations, and Palantir’s own guidance, by quite a margin. The financial results are in a class of their own.

Company profile

August 4, 2026 Follow-up coverage
Direction: Buy
Previous fair intrinsic value: $268.5, as of May 9, 2026

Symbol: PLTR, Exchange: NASDAQ
Sector: Technology, Industry: Software - Infrastructure
Theme: AI Software
Fair intrinsic value: $356 (124%), as of August 4, 2026
Market capitalization: $409 039 million
Pricing data: P/S 66x, P/E 135x

Previous coverage:

TYPEFCAPITAL.COM
Consider following me on X and YouTube.

This article is for informational purposes only and does not constitute investment advice or an offer to buy or sell securities.

Note: Type-F sells an aggregated Excel sheet of Palantir’s historical financial data, containing ~1000 rows of both company-reported metrics, but also calculated KPIs, contract data, and more. The sheet also includes my valuation model, a charting tool (internal in Excel and on the web), and a dashboard overview of the business.

Available at: https://typefcapital.com, with free quarterly sheet updates.

Stop tokenmaxxing

In a research note, I broke down a recent statement made by CEO Alex Karp in regard to the business being poised to hit $15-18 billion in FCF in two years. The implications from that statement would see Palantir grow its revenue by at least ~520% from its FY 2025 base by 2028, a massive undertaking which no one has even begun imagining (Wall Street consensus for 2028 is a mere $16 billion).

That statement implies an ~83% CAGR, and comes with massive implications for the intrinsic value of the business. Even more impressive, during a CNBC interview in relation to the Q2 earnings result, Karp stated that they are on a trajectory to grow at “this” pace over the next 18 months. During the Q2 earnings call, Karp clarified that the pace that he is referring to is the U.S. Commercial growth, which is 150% Y/Y.

Figure 1: Segmented revenue

The total revenue growth pace that Palantir showed in Q2 is what is required of the business to be able to live up to the statements made by the CEO in recent interviews. A trajectory of >150% growth across the business would far surpass any previous revenue assumptions derived from the FCF target of $15-18 billion in two years.

As investors and analysts, it is important to understand where that growth is coming from. For example, U.S. commercial revenue growth has accelerated for every passing quarter since Q1 of 2024, a feat which many thought would not be possible once that growth started exceeding 100% Y/Y. Now, Palantir has a full year of >100% Y/Y U.S. commercial revenue growth, and each quarter has managed to push the growth further and further.¹

Figure 2: U.S. Commercial revenue and revenue growth

¹ Q1 2026 saw a customer get reclassified from U.S. Commercial to U.S. Government. Without that impact, revenue growth would have been 143% for the segment during the period.

In order to achieve sustainable and robust growth for an extended period of time, there has to be scalability to the revenue. That ties directly to the notion of Palantir creating value for its customers, and as value is unlocked, so too should the contract value per customer increase. Net dollar retention has been consistently accelerating each passing quarter since the launch of AIP, which has usage-based pricing. If AIP is creating value, the use of the product should increase among existing customers (>12 months with Palantir), thus creating scale.

While the average customer is increasing their spend, as evident by NDR rising every quarter, so too is the cohort of Palantir’s largest customers. 2% (top 20) of customers account for 40% of total revenue, each providing >$120 million in annual revenue. That is over 3400%(!) more in annual revenue compared to the average customer outside of the top 20. For a year and a half, this massive cohort, which is already paying massive amounts to Palantir annually, is outgrowing the average customer base that has been with Palantir for over a year. That is strong evidence for the scalable nature of Palantir’s software.

Figure 3: Top-20 cohort and existing customer base average revenue growth (TTM)

Michael Burry has been one of the prominent voices criticizing Palantir for not being an AI company due to not having their own large language model (LLM), while CEO Karp made it clear from the very beginning that models are going to be commoditized.

We have been seeing evidence of the commoditization of LLMs at an alarming pace as of late, where open-source models distill frontier models (which also appear to be distillations to some degree) and replicate them at a fraction of the cost. In many cases, these distilled models, like Kimi K3, score better benchmark results than the frontier models.

At the same time, frontier AI Labs have been criticized for stealing IP with partners; a recent example includes Figma partnering with Anthropic, upon which Anthropic launched a competing product and Anthropic’s CPO, who was a Figma board member, resigned from the board.

The way it ties back to Karp’s criticism of tokenmaxxing is obvious in hindsight, but still caught investors and analysts off-guard. Palantir launched a partnership with Nvidia to deliver sovereign AI infrastructure, meaning that enterprises and governments can run models inside secure environments, with no risk of labs stealing IP. In addition, Palantir has started fine-tuning models for use in an enterprise context.

Palantir engineer Chad Wahlquist comments on the following observations from enterprise customers across the country:

• Customers are frustrated with having to assume risk for software products not working, while locking in the customers.

• Customers don’t want to subsidize AI labs that will ultimately launch a competing product.

• Customers are frustrated with having to exhaust tokens without receiving value in the output. For example, a PhD is overkill on a factory line where things are merely placed in a box. There is a need for task-specific models that are trained on the craft.

• Customers want to own the guardrails in their own environments; otherwise, the company risks not operating.

While models are being commoditized, and which frontier model reigns as king of the hill changes intermittently, Palantir has found a different approach to solve customer issues, provide value, and escape commoditization. They are fine-tuning models for enterprise-specific use cases, and their fine-tuned models outperform frontier lab equivalent models.

Palantir is uniquely positioned to fine-tune models because their software platform features Ontology, which is Palantir’s operational layer for the whole organization. Within the Ontology, the data of the organization is turned into objects, properties, links, and actions. The goal is to create a complete picture of the organization, and as such, provide an accurate context of the business to models.

Figure 4: Simplifed Ontology example

While the fine-tuning approach is rather new, through commentary, the demand for such solutions seems massive. Palantir’s software is model-agnostic, and it has already differentiated itself as a business that puts customer value-creation first, as evident by the sheer growth in existing customers. The fine-tuning approach is a smart way to build further on top of an already superior AI business.

Further sustainability of the massive growth

The rest of the research report is for paid subscribers only.

Please consider becoming a paid subscriber to help me do this full-time. That would allow me to deliver more frequent as well as higher-quality research.

Read more

Neoclouds’ soaring demand is the result of compute scarcity due to AI, but what the playing field looks like once first-term contracts are up could spell doom for the entire business model.

Company profile

Theme: Neocloud, Direction: Hold

Symbol: NBIS, Exchange: NASDAQ
Sector: Communication Services, Industry: Internet Content & Information
(legacy GICS classification)
Fair intrinsic value: $179.3 (-6%), as of August 02, 2026
Market capitalization: $58 831 million
Pricing data: P/S 67x, P/E 72x (inflated by revaluation gains)

TYPEFCAPITAL.COM
Consider following me on X and YouTube.

This article is for informational purposes only and does not constitute investment advice or an offer to buy or sell securities.

Business overview

Typically, my research reports contain a vast amount of financial data in order to understand core drivers of a business, its discipline, its strengths and weaknesses, its margin profiles, and the list goes on. However, with Nebius, there is not much to look at, at least not yet in terms of its current business.

Nebius has historically been an engineering organization that built Yandex, a search, cloud, and self-driving platform at technical depth few firms outside of the US have matched, but by using a fraction of the budget of a giant like Google. It was highly successful, and the growth in the business was impressive, reaching a peak of ~$10 billion (800 billion RUB) in revenue in 2023.

Figure 1: Legacy business revenue growth

In 2024, the company carved out the Yandex part of the business and emerged on the Nasdaq without a legacy business and a pile of cash. The company retained its Finnish data center and the cloud infrastructure teams, along with Avride, Toloka, and TripleTen. Nebius then built the most vertically integrated of the neoclouds by owning both land and power, self-designing its data centers and racks, and building a hyperscale cloud platform with a software layer above it. From a near-zero base, revenue growth has been in a vertical growth phase as it rides the secular tailwind that is AI.

Figure 2: Segmented revenue

Before we get into the intricacies of the neocloud business, it is worth highlighting all the business segments. In essence, Nebius Group consolidates one dominant business and two consolidated subsidiaries. The AI cloud business accounts for 98% of total revenue as of Q1 2026, making the other two immaterial both in terms of size and opportunity.

The Avride stake comprises an autonomous vehicle tech company, focused on developing and operating self-driving delivery robots and robotaxis. It was originally part of Yandex’s self-driving division, and now partners with platforms like Uber to provide driverless food deliveries and robotic rideshare services. It more than doubled its AV-capable fleet YTD, and launched robodeliveries in Philadelphia. It passed 500k cumulative robot deliveries in early April, and has received strategic investments from Uber.

TripleTen is an education tech platform that provides bootcamp program in technology fields such as software engineering, data science, and cybersecurity. The segment grew 10% Y/Y, with about 5000 new students in Q1.

The way to approach TripleTen and Avride in relation to Nebius Group is as free optionality. Simply, focus on the AI cloud business when looking at the business, and whatever the equity stakes produce, you get for free.

It is also worth noting that Nebius has two deconsolidated equity stakes in ClickHouse (real-time analytics database) and Toloka (AI data for frontier labs). The stakes matter both as assets and also serve as a GAAP reporting distortion. Nebius reported positive net income in FY2025, but it is the result of a revaluation artifact rather than due to the companies’ operating results.

The neocloud business

While revenue growth has been impressive across the neocloud space, the true economics and the attractiveness of the business are quite obfuscated. A neocloud converts MW into recurring revenue, and when looking at the unit economics, it seems as if neocloud contracts merely return the approximated capital costs of the assets, and nothing else.

Figure 3: AI Cloud segment growth

Looking at CoreWeave’s contracted power, the backlog of $99.4 bn revenue against its >3.5 GW of power yields about $28mm per contracted MW, which is the lower end of what an MW costs to build. The first contract is the financing instrument that pays for the whole asset, and not where the actual money is. For the economics of neoclouds to make sense, the second contract has to land. However, at years ~5-8, the hardware is already substantially depreciated, just for when the contract renewals are coming up.

The renewal is uncertain

The question to be asking is: does the second contract exist, and at what price? The neocloud revenue largely exists on a few balance sheets, primarily concentrated at hyperscalers like Microsoft, Meta, and Google, as well as AI labs. However, while hyperscalers underwrite an overwhelming majority of neocloud demand right now, there’s a big question as to whether they will continue to do so since they are building out their own data center capacity.

┌────────────┬────────────────────────────────────────────────────┐
│ Buyer      │ Neocloud commitment                                │
├────────────┼────────────────────────────────────────────────────┤
│ Microsoft  │ ~$60B (CoreWeave, Nebius, Nscale)                  │
│ Meta       │ ~$62B ($21B CoreWeave, up to $27B Nebius)          │
│ Google     │ Lease guarantees (Cipher, TeraWulf)                │
│ OpenAI     │ Expanded by up to $6.5B in 2026 (CoreWeave)        │
│ Anthropic  │ Multi-year agreement (CoreWeave Q1 2026)           │
└────────────┴────────────────────────────────────────────────────┘

The four largest hyperscalers are guiding for roughly $700 billion in combined capital expenditures in 2026, all while simultaneously being the largest renters of third-party capacity. The reason for this is simply time, and not technology. It takes power and time to build out capacity. If a hyperscaler wants to build GW capacity in 2027, it must have secured its position to do so back in ~2023. For example, Microsoft has described in commentary that they can’t serve their full Azure backlog due to power constraints.

Renting out capacity at hyperscalers eases its construction pains, and as mentioned, they are throwing a lot of money at the problem. By the time the leases expire, will there be a second contract? The demand could be large enough to warrant another contract, but will the hardware still be up to par, and what price will it fetch? Neoclouds will need to continuously build out capacity, and at a projected lower $ per MW.

The reason is spelled inference, and is a double-edged sword. Inference is a reliable source of recurring workload and has a higher margin profile compared to training. Inference accounted for roughly a third of AI compute in 2023, which became about half by 2025, and now in 2026, it accounts for two-thirds (per Deloitte). However, cost per unit of inference is quickly collapsing.

Per the Stanford AI Index 2025, the cost of inferencing a GPT 3.5-equivalent model has fallen from ~$20 per million tokens in late 2022 to $0.07 in 2024, a 280-fold reduction. Inference is the reason the overall market keeps growing, but for neoclouds, it is also the reason why revenue per MW will keep falling.

Figure 4: GPT 3.5-equivalent inference cost

Recent developments see Meta building its own cloud business to sell compute, while simultaneously being the sector’s second-largest customer with roughly $62 billion in committed capital across CoreWeave and Nebius. That adds a lot of fuel to the fire that is renewal risk.

Nvidia has begun underwriting GPU residual values under a new six-year backstop. Nvidia will rent back unused GPUs at a pre-agreed price in exchange for a share of cloud revenue. It removes the capital constraint that has been rationing supply in the sector, making it good for volume, but bad for pricing and the revenue/MW which neoclouds depend on for their economics.

Nebius cloud

The Nebius stack is built in several layers, but before we dive deep into Nebius neocloud, it is important to distinguish between the different kinds of neoclouds that exist. Operators, which are the category that Nebius and CoreWeave fall into, sell compute and carry risks in terms of GPUs becoming obsolete, utilization, and price risk. Landlords, which are companies like Applied Digital, Cipher, TeraWulf, etc., lease powered shells for up to 25 years at a much lower rate. They lease them to investment-grade or investment-grade guaranteed tenants, and transfer the demand risk in exchange for a fraction of the revenue accrued.

Generally, there is roughly 4-6 times less revenue per MW that landlords collect compared to operators, but that is simply the price of risk transfer. Landlords contractually shed demand, utilization, and obsolescence risk. There is also a third category, the aggregators, which contract capacity from landlords and resell it, carrying even less risk.

Published estimates of the price per AI megawatt vary depending on what is included.

• Conventional data-center shell and core

  • $10-12 million per MW

• AI-optimized shell (liquid-cooled, high-density)

  • $15-20 million per MW (higher power density, liquid cooling loops, heavier electrical work)

• Tenant-adapted shell with GPUs and networking

  • ~$20-25 million per MW (GPUs alone exceed $20mm/MW)

• All-inclusive operational AI megawatt

  • ~$28-38 million per MW

Operator revenue run rates are roughly $8-12 million per MW at current pricing (CoreWeave $8.3bn annualized against >1 GW, Nebius $1.92bn ARR against ~200 MW active power), which is also expected to drop over time. With that in mind, gross revenue payback becomes >3 years, and that is before any dollars of operating cost. Once power, hosting, staff, SG&A get layered in, it’s more looking like ~5 years until payback. The weighted-average contract per CoreWeave’s disclosures is about 4 years, which means the contract approximately repays the asset, and that’s at current rates.

Nebius monetizes its physical assets in a variety of ways, each with a different margin profile.

• AI factories, including land, power, and DCs

  • Monetizes capacity. Is capital-intensive, but has a cost advantage layer since Nebius has in-house DC design.

• Reserved GPU capacity

  • Monetizes hyperscalers and AI labs through visibility and prepayments. Generally priced below spot, but finances the overall build.

• On-demand and short-term AI cloud (Aether)

  • Monetizes startups and enterprises across a variety of verticals. Has a higher rate per GPU-hour, but still retains utilization risk.

• Token Factory (managed inference)

  • Serving custom and open-source models on a per-token basis (Revolut, monday.com, etc). Has the same margin characteristics as a software business, and sits on top of the same GPUs. Becomes strategic for post-scarcity pricing power.

• Agentic and software layers

  • Monetizes search and inference optimization.

• Other stakes and businesses

  • Avride (AV/Robotics), TripleTen (edtech), ClickHouse, and Toloka stakes. Serve as optionality and are largely immaterial to revenue, but material to asset value.

Nebius has been aggressive in adding new locations and increasing the capacity potential globally. As of Q1 2026, there are 13 disclosed sites, with guidance of 16 total sites by the end of 2026. As of 2025, sites are operational in Finland, Kansas City, Vineland, Iceland, the UK, and Israel, with 2 under construction and the rest just being announced.

Figure 5: Site disclosures

A major expansion is needed in order to acquire credibility for future contracts. As of Q1 2026, contracted power is guided to be >4 GW by year-end. Of that, ~1 GW is expected to be connected power, a guide that has strangely stayed unchanged since 2025 while contracted power quadrupled.

Figure 6: Contracted power guidance, end-of-period 2026

While there are limited KPIs and financial data to observe in Nebius current state, there are a lot of metrics to track over time to ensure a proper monitoring framework for the sector.

• Contracted power, connected power, and active power conversion

  • Connected power hits guidance targets and active power tracks it within one or two quarters.

• Backlog conversion rate

  • These should be growing together to indicate a healthy reading.

• Revenue per active MW

  • Given the capital-intensive nature of the business, $mm/MW will be an early indicator of sustainability.

• Hyperscaler CapEx revisions

  • If capital expenditure continues to be revised up at hyperscalers, it signals an increase in demand. Once it starts getting cut, it indicates a sector-wide top signal.

Nebius ARR is defined as last-month AI-cloud revenue multiplied by 12, making it a run-rate snapshot and not contracted forward revenue. This is the prime metric for capturing the momentum of the business, but it won’t tell us about duration or churn, and will stay ahead of trailing revenue while the business ramps.

Figure 7: ARR

The rest of the research report is for paid subscribers only.

Please consider becoming a paid subscriber to help me do this full-time. That would allow me to deliver more frequent as well as higher-quality research.

Read more

Read more

Company profile

July 21, 2026 Initiated coverage
Direction: Hold
Previous fair intrinsic value: N/A

Symbol: MCO, Exchange: NYSE
Sector: Financials, Industry: Financial Data, Stock Exchanges
Theme: High quality
Fair intrinsic value: $413.13 (-18.5%), as of July 21, 2026
Market capitalization: $89 883 million
Pricing data: P/S 11.42x, P/E 35.98x

TYPEFCAPITAL.COM
Consider following me on X and YouTube.

This article is for informational purposes only and does not constitute investment advice or an offer to buy or sell securities.

AI as an accelerator or disruptor?

Moody’s operates two distinct revenue segments. Moody’s Investor Services (MIS) charges issuers for its credit ratings, similar to S&P Global’s ratings segment. They record a transaction fee whenever a bond, loan, or other financial instrument is rated, and then record additional revenue on the initial rating through fee schedules and monitoring fees for maintaining the rating. Moody’s Analytics (MA) sells workflow tools, risk software, and data feeds on a subscription basis, where close to all of the Analytics revenue is recurring.

Figure 1: Segmented revenue

Moody’s is flat year-to-date, but was down significantly earlier in the year on AI disruption fears. As mentioned, markets like compounders because they are a source of durability, and AI disruption fears left a question mark on that characteristic of Moody’s business. S&P Global made several statements in regard to how much of their data services are proprietary and unreplicable, and Moody’s has done the same throughout the year, though without S&P’s quantification.

There are two angles to the AI threat: one concerns which parts of Moody’s business can be replicated, and the other angle concerns whether it matters if they are.

For example, the credit rating value comes from the embeddedness of the standard; permissions and coordination relating to index eligibility, capital treatment, and acceptance by the bond market. In theory, an LLM could produce a credit opinion, and in some cases it could even be indistinguishable from a Moody’s rating, but that does not mean that it produces a rating that a bond mandate accepts. As a reference, FICO scores have had a competitor for two decades that hands out its credit scores for free, and the market still does not switch from FICO because FICO scores are simply too integrated into the financial system.

As for the more vulnerable part, the data in Orbis might be proprietary, but that does not mean that the raw facts behind the data are exclusive to Moody’s. Where the value lies is the curation layer, where the accumulated labor over decades has built an entity resolution with ownership graphs, identifiers, licensing agreements, and so forth. To use FICO as an example again, owning the raw data doesn’t get the standard, exemplified by FICO’s competitor VantageScore being owned by the big three credit bureaus, which actually own the consumer data that the FICO scores utilize. The very owners of the underlying data could not disrupt or displace the standard sitting on top of their data.

One key metric for evidence against AI disruption is remaining performance obligations (RPO). RPO is contracted rather than assumed, and as such, if AI were eroding demand for Moody’s, the backlog would take a hit before revenue, and it has not. The percentage of RPO that is recognized over the next 12 months is a tell for duration extension, and if it were to start increasing, it would mean that customers are less committed. A falling near-term share means more of the backlog is long-dated, meaning that customers are signing longer contracts.

Figure 2: Remaining performance obligations

Analytics

Perhaps the strongest counterargument is that we are about three years into the generative AI era, and Analytics organic revenue growth has not slowed. In addition, margins have been increasing, showing the increased attractiveness of the segment despite the market’s assumed AI pressures.

Figure 3: Segmented Analytics revenue

What Moody’s sees is that AI drives demand for its data, where customers adopting generative AI solutions from Moody’s grow at 2x the rate of their other customers.

However, the point at which the customers want to consume the data is shifting, and Moody’s has recognized it. Historically, customers would have to go to the data providers’ platforms to get the full breadth of the data, but that is quickly changing with AI implementations. Moody’s largest enterprise customers are building out their own AI infrastructure internally, and the data has to be able to live within those environments. The new consumption demand preference requires change, which is why Moody’s has built an MCP (model context protocol) integration to meet the demand before it becomes a threat.

Moody’s has partnered with both hyperscalers (Amazon, Microsoft) and frontier AI labs (Anthropic, OpenAI) to build a scalable agentic stack. For investors in Moody’s, this change further strengthens the predictable nature of the Analytics revenue as it further drives revenue towards a recurring mix. Moody’s is also deliberately partnering with several major AI platforms, positioning itself as an infrastructure layer, rather than betting on one winner in the AI frontier race. The infrastructure keeps monetization within licenses and stays neutral on which AI interface wins.

Analytics revenue is quickly approaching being rounded up to a full 100% recurring mix, and makes up ~45% of overall revenues.

Figure 4: Analytics recurring revenue mix

The rest of the research report is for paid subscribers only.

Please consider becoming a paid subscriber to help me do this full-time. That would allow me to deliver more frequent as well as higher-quality research.

Read more

AMD has become a second-source specialist, entering huge markets with a dominant vendor. The problem huge markets with a dominant monopolist face is that customers have a structural desire for alternatives, which moderates prices, supply, and pushes the pace of innovation of the whole industry.

Company profile

Theme: AI Accelerator, Direction: Buy

Symbol: AMD, Exchange: NASDAQ
Sector: Technology, Industry: Semiconductors
Fair intrinsic value: $503 (-8%), as of July 10, 2026
Market capitalization: $894 532 million
Pricing data: P/S 24x, P/E 179x

TYPEFCAPITAL.COM
Consider following me on X and YouTube.

This article is for informational purposes only and does not constitute investment advice or an offer to buy or sell securities.

Perpetual second

Even AMD’s original x86 business benefited from IBM’s refusal to let Intel be the sole supplier in 1982, thus forcing a second source to emerge. However, not everything was rosy for AMD; in fact, it was on the brink of bankruptcy before now CEO Lisa Su entered as CEO in 2014. Ruthless capital allocation decisions had to be made to ensure the survival of AMD, and since it lacked the R&D budget to compete against both Nvidia and Intel, Lisa Su hedged AMD’s survival on the CPU side and took the fight to Intel. The GPU division was starved of funding, while Nvidia poured billions into R&D, extending its lead in architecture and efficiency. Therefore, while AMD never seems to be the industry leader, the company has had to carve a path through other means.

AMD, like Nvidia, is a semiconductor designer and outsources its manufacturing. Wafers are manufactured by TSMC , advanced packaging by OSAT and TSMC, and memory by SK hynix, Samsung, and Micron. Since AMD employs a fabless model, its capital intensity is structurally low. AMD is also able to scale free cash flow quickly with revenue growth, but the financial advantages also come with downsides. Supply has to be negotiated, thus becoming a forward-committed resource. Wafers, packaging slots, and high-bandwidth memory must be reserved quarters, or even years, ahead.

The revenue segments are divided by end market, ranging from gaming GPUs (Radeon), client encompassing desktop and notebook CPUs and APUs (Ryzen), to server CPUs (EPYC) and data center AI solutions (Instinct).

Figure 1: Total revenues

Data center will decide everything

The data center segment in particular is what makes or breaks the thesis for AMD, as everything hinges on whether AMD can stay a major player in the AI revolution. EPYC server CPUs and Instinct AI accelerators are sold to hyperscalers, cloud builders, enterprises, and AI labs. In a sense, the segment can be seen as two businesses at different stages of their lifecycles: a mature franchise in EPYC, and a hypergrowth business in Instinct. AMD reports them together, making it difficult to gauge how AI GPUs are performing against the backbone of server CPUs.

Looking at the market share at face value, it may seem like AMD has continuously lost its place in the market. However, that is due to Nvidia’s first-mover advantage in the AI space. Data centers have historically run on CPUs, solving sequential problems, one at a time. However, LLM training during this phase of the AI revolution requires a different kind of processing, namely solving many simpler tasks simultaneously. GPUs excel at parallel processing, and Nvidia was the first to capitalize on the opportunity, extending its dominance through its CUDA platform and NVLink technology. NVLink allows for multiple GPUs to process as a single, efficient cluster. Meaning, whenever a hyperscaler wanted to place a large order of GPUs, they couldn’t mix and match among vendors, allowing Nvidia to take most of the market. During the initial stages of the AI arms race, there was no time to wait for incumbents, and Nvidia’s inbound volumes exploded.

Figure 2: Data center revenues, segmented by vendor

In Q1 2022, AMD held an implied 8% data center market share. In Q1 2026, that market share has diminished to 6.7%. At a glance, it spells disaster for AMD; not only is AMD not gaining market share, but it is actively losing it over time. However, excluding Nvidia from the picture, the market share narrative is turned on its head. AMD has gone from 11% data center market share to overtaking Intel at 54% market share in Q1 2026.

Figure 3: Implied data center market share between Intel and AMD

The CPU business is a qualification business that spans several years. A hyperscaler qualifies a platform, meaning the platform (CPU, firmware, software stack, etc) over half a year to a year, and then it is deployed over several years. That dynamic has protected Intel’s server business for decades, but once share flips, it typically stays flipped for a long time. AMD’s server share, while veiled, has likely not retraced for close to a decade since every EPYC generation has launched on time, while Intel’s advanced nodes have been delayed, and transitions have slipped.

The AI business is starting to shift massively, just as Nvidia saturated most of the market. Nvidia’s had such an immense edge that a second supplier became desperately needed. Unlike CPUs, AI platforms are bought in large-scale commitments, typically sized in gigawatts of data center power rather than units. They are also negotiated at the CEO level, to then be deployed as complete racks, which involve the compute, interconnectivity, networking, and cooling, rather than as mere chips. Nvidia’s rack solutions created vendor lock-ins since there were no alternatives, but AMD is now becoming competitive.

Historically, AMD’s weakness was software, but ROCm, which is the equivalent of Nvidia’s CUDA, is now becoming mature. CUDA and ROCm are software stacks that enable the harnessing of the compute power of GPUs. CUDA, while proprietary, was cleverly made available for free to researchers and universities over a decade ago, resulting in a majority of AI libraries, math frameworks, and learning models being specifically written for CUDA. That in and of itself created a moat, since customers would be reluctant to change vendors, as it would mean having to rewrite code from scratch. ROCm 7 makes it close to seamless to migrate AI/ML frameworks like PyTorch from Nvidia to AMD, and as such, the underlying hardware language now matters less.

Another major hurdle for AMD that is about to be overcome is that NVLink, like CUDA, is Nvidia’s proprietary technology for cluster processing. NVLink became mandatory since there was no alternative for unifying racks, forcing vendor lock-in. However, a coalition including Microsoft, Google, Meta, Intel, AMD, and Broadcom has come together in order to form the Ultra Accelerator Link (UALink) consortium. UALink will enable interchangeability between various semiconductor vendors, tearing down the walled garden that NVLink historically has provided. To further strengthen its rack-scale systems capability, AMD acquired ZT Systems in March of 2025, but divested the manufacturing arm and only kept the rack design and enablement teams.

Like UALink and ROCm, other layers of AMD’s stack are also open standards. AMD uses the OCP (Open Compute Project) open-rack standard introduced by Meta for its data center hardware. Vendors who fear vendor lock-in have strategic reasons to fund the open stacks.

AMD’s key advantage is its price-per-performance, meaning tokens-per-dollar, especially in inference. That is also a key pivotal point in this stage of the AI buildout: large-language models are sophisticated enough to be widely deployed across enterprises, meaning that there is increasingly more demand for inference (using models) rather than just training. Customers, however, are unhappy with the costs of deploying AI, with many companies imposing per-employee token budgets.

Figure 4: LLM inference throughput, full system performance comparison against Nvidia

GPUs are versatile and can be applied to a wide variety of workloads, and while not the most efficient at every task, they are very flexible. As such, even though computational demand gradually shifts from training to inference, GPUs will still stay relevant as a jack-of-all-trades. However, the companies committing the most capital expenditures towards GPUs are the hyperscalers, most of which are now building out their own custom silicon (ASICs). That causes not only an introduction of new competition, but also a loss of revenue from the largest spenders.

The future of the AI data center market carries a lot of uncertainty, and forecasts by third-party experts are continuously revised. However, the consensus throughout the volatility in expectations all points towards a prosperous decade for AI accelerators. It is hard to predict how the near-term shift between GPUs and ASICs will play out, but it is looking like there is enough demand to stay distant from saturation for a long time to come. AMD has already de-risked the MI400 ramp by signing deals with both OpenAI and Meta for 6 gigawatts of AMD GPUs, which will take years to fulfill. AMD themselves target a >60% CAGR in the data center segment over the coming 3-5 years, citing a $1 trillion compute TAM by 2030 (>40% CAGR). That alone shows a lot of confidence in the experienced demand in the near-term.

AMD discloses very limited information about its AI business within the data center segment. Using those clues and vague disclosures, I have tried to put forward an estimated AI data center revenue line. The implied acceleration is astonishing, going from ~2% of data center revenues to >50% in 3 years.

Figure 5: Estimated Data Center AI revenue

The rest of the research report is for paid subscribers only.

Please consider becoming a paid subscriber to help me do this full-time. That would allow me to deliver more frequent as well as higher-quality research.

Read more

However, history has proven Karp’s ambitious claims to come true, even if they seemed extremely unlikely at the time. I believe his most recent outrageous claim will prove true, as well.

Company profile

Previous fair intrinsic value: $268.5, as of May 9, 2026

Symbol: PLTR, Exchange: NASDAQ
Sector: Technology, Industry: Software - Infrastructure
Theme: AI Software
Fair intrinsic value: $340.77 (163.55%), as of July 5, 2026
Market capitalization: $332 420 million
Pricing data: P/S 64x, P/E 145x

TYPEFCAPITAL.COM
Consider following me on X and YouTube.

This article is for informational purposes only and does not constitute investment advice or an offer to buy or sell securities.

The history of Alex Karp’s ambitious claims

I have made the latest Palantir coverage completely free to read by removing the paywall. I highly recommend reading it for a grasp on where Palantir is in its narrative.

It also serves as a free preview of the level of research I publish here on my Substack. You can find the research coverage below:

This article is not a full research piece.
It targets specifically the comments made by Alex Karp during his CNBC interview.

Alex Karp is not shy about sharing his thoughts, be it geopolitical views or forecasts of where he sees his own business over the coming years. When Palantir went public through a direct public offering (DPO) in September of 2020, one of the key financial targets was a compounded annual growth rate (CAGR) of 30% through 2025.

Looking at the financial performance surrounding the DPO, not only did 30% CAGR seem probable, but it even seemed understated. From Q1 of 2020 through Q2 of 2021, Y/Y revenue growth averaged 48%.

Figure 1: Total revenue and revenue growth, Q1-2020 through Q2-2021

However, what followed was a rapid deterioration of growth rates, with an absolute low of 12.7% Y/Y revenue growth rate in Q2 of 2023. During the decline, Palantir quietly stopped reiterating its 30% CAGR target, raising the eyebrows of investors.

Figure 2: Total revenue and revenue growth, Q3-2021 through Q2-2023

During those times, Palantir still had representation of various financial analysts, and Sanjit Singh of Morgan Stanley addressed the elephant in the room during the Q2 2022 earnings call. That was the first quarter that Palantir dropped below 30% Y/Y revenue growth since going public.

Alex Karp’s response was sharp, leaving no room for speculation:

$4.5 billion in 2025 would be a 33% CAGR from FY 2020, the year Palantir went public. Not only reiterating the 30% CAGR goal, but overshooting it. At the time, for each quarter that followed that quote over the next year, the target seemed more and more unlikely, and the stock quote reacted accordingly, hitting single digits and trading there for close to a year.

However, in Q2 of 2023, at the absolute rock bottom of Y/Y growth, Palantir launched AIP. Since then, for three years to date, every single quarter that followed the bottom of 12.7% in Q2 of 2023 has been higher than the previous quarter. A revenue acceleration out of this world, and one that most investors would find absolutely absurd as a claim if you told them that at the time of decelerating growth. Most did not believe Karp’s reiteration of the 30% CAGR, and the following quarters proved it to be more and more unlikely.

2025 came around, and sure enough, Palantir hit Alex Karp’s forecast of $4.5 billion in revenue, however unlikely it sounded at the time. In Q2 of 2022, when Alex Karp replied to the Morgan Stanley analyst, $4.5 billion would require revenues to 2.5x in 2.5 years, and it did.

Figure 3: Total revenue and revenue growth

The seemingly impossible lofty goals

That brings us to today and the new claims made by Alex Karp. History serves as the witness to the credibility of the ambitious CEO’s claims, and if nothing else, that they should not be disregarded easily.

Before we look at the most outrageous goals of them all, let’s rewind to a few months ago, when Alex Karp set a forecast for the company that changed intrinsic value massively.

After the amazing year of 2025, my fair intrinsic value for the business was $205. Not only was revenue growth accelerating for each passing quarter, hitting 70% Y/Y in Q4 of 2025, but margins were simultaneously exploding. Palantir has immense operating leverage as a capital-light business, unparalleled scalability, and sticky revenue. However, in a single quarter, that fair intrinsic value estimate changed to $270 in a single quarter due to a specific quote during the Q1 2026 earnings call:

U.S. revenue is ~80% of total revenue, and seeing two consecutive years of at least 100% growth has massive implications for the business. The fair value estimate of $270 may seem like a delusion at face value, but when one extrapolates the characteristics of how Palantir grows its revenues and combines it with the expanding margin profile that seemingly has no cap, that fair value estimate no longer seems as crazy.

Now, after Alex Karp’s CNBC interview a few days ago, my fair value estimate for Palantir is $341, an unfathomable number. However, the claim by Alex Karp is equally unfathomable, and once one understands what it implies, $341 no longer seems as a crazy fair intrinsic value estimate.

A bomb was dropped, and due to the nature of the intense ~18-minute interview, this quote seemingly passed by without much attention. However, $15 billion in free cash flow would mean a 5x in 2.75 years. A 2.5x in revenues in 2.5 years was impressive, but a 5x in free cash flow, when adjusted FCF margins are already at 57%, the implications are massive. If FCF margins were low, a 5x would be quite achievable, but as it stands, the implications of a 5x are titanic for the business, as it implies a massive increase in revenue. And that is just at the lower end of the target, $15 billion.

Unlevered free cash flow, commonly used in valuation models, which also accounts for reinvestment needs, is slightly lower than adjusted free cash flow. Karp’s target likely refers to adjusted free cash flow, which is adjusted for payroll taxes associated with stock-based compensation.

When I model my expectations for the business while keeping $15 billion in adjusted FCF in mind, the inputs seem fairytale-like. While I have allowed for reasonably probable margin expansion as well, the CAGR rates for the business segments eclipse my wildest imaginations for the company:

• U.S. Government CAGR: 80%
  Non-U.S. Government CAGR: 54%
  Total Government CAGR: 75%

• U.S. Commercial CAGR: 136%
  Non-U.S. Commercial CAGR: 12%
  Total Commercial CAGR: 112%

• Total Revenue CAGR: 93%

Figure 4: Historical segmented revenue and revenue forecast

When these figures are then extrapolated over 10 years, while keeping the sheer scalability and sustainable nature of the growth in mind, $341 fair intrinsic value does not seem that absurd anymore.

Download the valuation model:

The rest of the research report is for paid subscribers only.

Please consider becoming a paid subscriber to help me do this full-time. That would allow me to deliver more frequent as well as higher-quality research.

Read more

Read more