Written by: Rita
Two leading AI labs both generated $100 in total revenue in 2026; one ended up with $55 in profit, while the other had only $38. The $17 difference does not stem from model capability, but from the business structure itself.
On August 27, Barclays released a report on the U.S. internet industry, systematically breaking down the unit economics of AI labs and hyperscale vendors. API products have significantly higher profit margins than subscription products, and these structural differences account for most of the gap in gross margins between AI labs and hyperscale vendors. As AI labs begin to disclose GAAP financial data, investors need to look beyond revenue figures to understand these structural differences.
API profit margin has exceeded 65%, while subscription products are nearly 30 percentage points lower.
Barclays estimates that the profit margins for paid inference at AI labs have surged to over 50% to 65% by 2026, a significant increase from the low double digits in 2025. API products offer the highest margins, while subscription products are relatively lower, primarily due to some users benefiting from token cost subsidies provided by AI labs to remain competitive.
Subscription plans are typically monthly or usage-based, often with usage limits, allowing users to switch between different models, with significant variations in token consumption rates. Enterprise subscriptions are priced per license and offer higher limits. The API operates on a pure pay-per-use model, where developers (such as Cursor and Figma) embed AI functionality into their products and invoke the API to generate tokens. As model token efficiency improves and API pricing increases, API profit margins continue to expand.
Barclays presents a set of hypothetical data: Laboratory A’s revenue mix consists of 30% from subscriptions, 45% from direct API sales, and 25% from indirect API sales. Laboratory B has the opposite structure, with 80% from subscriptions and only 10% each from direct and indirect API sales. The inferred profit margins between the two diverged from 4% and 8% in 2025 to 55% and 38% in 2026. Barclays notes that API and paid enterprise accounts have higher profit margins, and as the business mix shifts further in this direction, overall margins still have room to grow.
The revenue recognition method for indirect APIs creates a reporting illusion.
The business model of the indirect API is the same as that of the direct API: users invoke the AI Lab's API to generate tokens, but billing and customer relationships are handled through a hyperscale vendor, with no direct interaction between the user and the AI Lab.
The issue lies in the revenue recognition method. AI Lab A recognizes indirect API revenue on a gross basis, and this portion continues to grow. AI Lab B, on the other hand, recognizes related revenue on a net basis, and even does not recognize any revenue when its strategic partner operates the indirect API.
Barclays compares this to Uber and Lyft. While their core businesses are the same, differences in how total revenue and net income are recognized lead to significant disparities in their financial statements. As indirect APIs account for an increasing share of overall AI lab revenue, this recognition method will significantly distort revenue figures, potentially misleading the market’s assessment of the relative progress of each AI lab.
Of every $100 in AI revenue, $35 to $40 flows to cloud providers.
Barclays estimated the flow of AI lab revenues across the supply chain. For every $100 in AI lab revenue in 2026, approximately $35 to $40 ultimately becomes inference revenue for hyperscale providers, who generate about $10 to $20 in operating profit from this, corresponding to an operating margin of roughly 35% to 45%.
Cash flows vary across different business models. Under the direct sales API model, approximately $35 of every $100 in AI lab revenue flows to hyperscale vendors, generating about $11.8 in operating profit with a margin of roughly 34%. For Lab B, where subscription revenue constitutes a higher share, strategic partnership revenue sharing results in $41 flowing to hyperscale vendors, yielding approximately $19.1 in operating profit and a margin of about 47%. Barclays emphasizes that, excluding revenue sharing, the actual profit per token is largely consistent across providers.
Inference profits will eventually surpass training costs, and cloud providers' market share is expected to decline by 2028.
The vast majority of current AI lab revenues ultimately flow to the financial statements of hyperscale vendors, primarily because training costs constitute a very high proportion of total costs. Barclays expects that inference profits will gradually surpass training costs, driving improved profit margins for AI labs themselves.
The landscape is shifting. Starting in 2028, AI labs' own backstopped AI infrastructure projects will gradually come online and become the preferred source of computing power. Barclays expects AWS, Azure, and GCP to continue controlling the majority of AI lab computing spending over the next two years, after which they will gradually be supplanted by in-house infrastructure.
Barclays' conclusion is clear: the unit economics of AI labs are improving rapidly, but significant apparent differences arise due to factors such as business structure, revenue recognition methods, and partnerships. When comparing AI labs, investors cannot rely solely on revenue figures—they must look beneath the surface to underlying variables like product structure, the proportion of API and subscription revenue, and revenue recognition practices. Competition has shifted from model capabilities to business models.
Disclaimer
This article is a compilation and interpretation by Chaoxiang Research of a third-party brokerage research report (Barclays, August 27, 2026), combined with publicly available market information. The ratings, price targets, earnings forecasts, and related judgments cited herein reflect the views of the brokerage’s analysts and represent the position of their respective institution, not the views of Chaoxiang Research, nor do they constitute any investment advice.
The market carries risks; make decisions independently. This article should not be used as a basis for buying or selling any securities.