Anthropic Gains Ground on OpenAI Where It Hurts Most: Corporate Spending
Ten weeks ago, Anthropic and OpenAI were sharing the spending of new companies purchasing AI tools in nearly equal proportions. Today, according to Ramp's corporate spending index, Anthropic captures a remarkable 73% of that segment. This shift did not occur due to a spectacular product launch or a marketing campaign. Rather, it happened because a specific group of buyers—those making their first purchasing decision—began to consistently lean towards one provider over the other.
From the perspective of structural risk analysis, this says more about the revenue architecture of both companies than any valuation metric could.
The 73% Is Not a Headline; It's a Structural Signal
When a corporate spending index shows that companies that never previously paid for AI are now choosing Anthropic in three out of four cases, the explanation lies not in the product itself. It’s about the composition of perceived risk by the corporate buyer.
The areas where Anthropic is gaining ground, according to Ramp’s data, include information, finance, and personal services, with higher concentration in venture capital-backed firms. These sectors share a commonality: high regulatory sensitivity and the need for predictable contracts. A CFO approving AI spending does not want a provider whose business model subsidizes consumer users with negative margins. They want visibility into future costs and a counterparty that isn't burning cash to maintain a base of free users.
Here’s the financially relevant contrast: OpenAI generates approximately $5.5 billion annualized from the consumer segment and sustains structural losses in that area as it scales. Anthropic, in contrast, projects that 86% of its 2025 revenue will come from API sales to businesses, with the remainder from subscriptions to its chatbot, Claude. This isn't just a difference in commercial strategy. It represents a difference in revenue quality: multi-month contracts with documented willingness to pay versus consumer subscribers who can cancel frictionlessly in the first month.
The ratio of API to consumer revenue also defines variability in service costs. When the volume comes from business clients with negotiated contracts, the inference cost is predictable. When it comes from millions of individual users with erratic usage, the infrastructure must be scaled for peaks, not for averages.
What the Training Numbers Reveal
Anthropic projects spending $4.1 billion on model training in 2025. OpenAI allocated $9.4 billion just for training compute last year. The difference is not simply due to one having more capital than the other; it reflects different risk scales they are betting on.
Training expenditure is, in terms of capital structure, a sunk cost with enormous uncertainty regarding returns. No one can guarantee that the model emerging from that investment will be sufficiently superior to the previous one to justify the differential cost. OpenAI is making a concentrated bet: spending more than any competitor in the hope that the scale advantage in models translates into sustainable market dominance. Anthropic is making a different bet: spending less on relative training, charging more for corporate access, and building dependency through integration, not model superiority.
The metric that validates which of these bets has a better chance of surviving a downturn isn't total revenue. It’s the moment when each company reaches positive cash flow. According to available analyses, Anthropic projects that point to be reached by 2028, two years earlier than OpenAI. In a sector where funding cycles can contract without warning, a two-year difference in financial self-sufficiency is a non-trivial structural advantage.
Microsoft has already committed nearly $500 million in spending with Anthropic, partly through GitHub Copilot. Cursor and Cognition also rank as significant clients. These aren’t customers who arrived out of curiosity and canceled in 30 days. They are companies that have integrated the API into productive workflows, creating a real and measurable switching cost.
OpenAI's Pivot and What It Confirms About Its Position
The Wall Street Journal reported that OpenAI is considering reducing its exposure to consumer initiatives, such as video generation, browsers, and hardware, to refocus on the corporate segment. Should this move happen, it indirectly confirms the diagnosis that Ramp’s data already indicated: the consumer segment isn’t generating a return that justifies its capital cost.
OpenAI reports 3 million paying business users, with ChatGPT Enterprise priced at $60 per seat per month. Anthropic does not publish that figure with the same granularity, but its annualized API revenue reaches $3.1 billion, surpassing OpenAI's equivalent segment, which stands at $2.9 billion. This means Anthropic is already earning more than OpenAI in the only segment where both compete directly with comparable pricing models.
The projected business penetration for 2026 shows OpenAI moving from 37.2% to 42%, and Anthropic from 14.5% to 22%. The percentage increase for Anthropic (+9.9 percentage points) more than doubles OpenAI’s (+5.5 points). In terms of market capture speed, this sustained differential over time indicates a convergence that current revenue models are already anticipating.
OpenAI is projecting $25 billion in revenue for this year, while Anthropic anticipates $19 billion. The gap is there, but the direction of the vectors matters as much as the absolute level. A company with $25 billion in revenue but infrastructure costs and consumer subsidies surpassing its operating margin carries a structurally different risk profile compared to a company with $19 billion and 86% of its revenue tied to business contracts with predictable renewal.
The Emerging Market Does Not Reward the Most Powerful Model
Competition in the corporate AI segment is shifting from a benchmarks race to something more akin to what determines which B2B software provider survives a full cycle: deep integration, cost predictability, regulatory support, and switching costs for the customer.
Anthropic has built its corporate reputation partly by taking positions on AI governance issues that fostered trust among institutional buyers, especially in regulated sectors. That reputational capital may not appear on any balance sheet, but it does appear in the purchasing decisions of early adopters that Ramp is measuring.
xAI and Google recorded 1.8% and 1.1% penetration in the corporate market respectively, indicating that the current market concentration between the two leading players isn’t being eroded by third parties but rather internally redistributed. The market is consolidating sooner than many analyses anticipated, and the speed of that consolidation favors those with less reliance on external subsidies for growth.
Anthropic arrives at this point of consolidation with a revenue architecture where the business client finances growth, the training costs are relatively lower, and the path to financial self-sufficiency is two years shorter than that of its main competitor. This does not guarantee the outcome, but it accurately describes which of the two structures is less likely to require an emergency round if capital markets contract.
