The ledger does not lie, only the interpreters do. At the end of July, Anthropic's annualized revenue run rate hit $65 billion—roughly $25 billion above OpenAI's $40 billion. These figures, sourced from people familiar with the matter, signal a 622% expansion since late 2025. But rapid growth in a capital-intensive industry like AI is not a proof of health; it is a stress test on structural integrity.
Context: The AI Arms Race as a Financial Engineering Problem
Anthropic and OpenAI are not crypto protocols, but their revenue trajectories share the same mathematical incentives that govern DeFi yield farming: subsidize metrics, attract capital, and hope the flywheel holds. Anthropic crossed $9 billion run rate at the end of 2025, then $47 billion in May, and now $65 billion. Quarterly revenue jumped from $787 million (Q2 2025) to $11.5 billion (Q2 2026), while adjusted operating income turned positive. OpenAI, meanwhile, is on track for $40 billion run rate, doubling its year-end 2025 level. Both companies have filed confidentially with the SEC. Anthropic's IPO is expected this fall, with a $2 trillion valuation floated by investors.
Core: Deconstructing the Growth Curve
Let me apply the same forensic skepticism I used when auditing the 0x Protocol v2 reentrancy vulnerabilities. The 622% expansion in seven months is not a hockey stick—it is a parabolic arc that demands scrutiny. The May-to-July gain of $18 billion (38%) implies a monthly growth rate of about 12%. If sustained, the run rate would exceed $100 billion by year-end. But compounding at that rate requires a linear increase in absolute revenue that becomes geometrically harder to maintain.
During the 2021 DeFi frenzy, I analyzed Curve Finance's gauge voting system and found that incentive distribution favored whale wallets. Here, the same principle applies: enterprise clients and API access are the whales. Anthropic's revenue concentration is not publicly disclosed, but a single large contract from a cloud provider like AWS (which invested in Anthropic) could distort the run rate. The $11.5 billion quarterly revenue is impressive, but it represents only 17.7% of the annualized run rate—meaning the current quarter's pace is already baked into the forward estimate. Any deceleration in new contracts will immediately flatten the curve.
Trust is a bug, not a feature. The run rate metric itself is an estimate that assumes no change in the current revenue pace. Anthropic and OpenAI may not calculate it the same way. In my experience auditing Terra/Luna's algorithmic stability, the difference between a 5% and 10% growth assumption could flip a death spiral into a false sense of security. The 622% expansion is real, but it is a trailing indicator of past demand, not a guarantee of future stickiness.
Contrarian: What the Bulls Got Right
To be fair, the bulls have a point: the growth is not entirely fabricated. Anthropic posted positive adjusted operating income—a rare milestone in AI, where most firms bleed cash. This suggests that the company's cost structure (compute, talent, infrastructure) is scaling slower than revenue, a sign of operational leverage. The $10 billion total revenue in 2025 compared to $11.5 billion in Q2 2026 alone indicates a revenue explosion that is more than just a base effect.
History repeats, but the gas fees change. In the crypto bear market of 2022, protocols that survived had strong unit economics and real user demand. Anthropic's product (Claude models) has genuine enterprise traction, unlike the speculative liquidity mining that evaporated when incentives stopped. The $2 trillion valuation, while eye-popping, is not absurd if the company can sustain 50%+ annual growth for a few more years.
Takeaway: The Accountability Call
Code is law; intent is irrelevant. The numbers say Anthropic is growing faster than OpenAI, but the run rate is a liability disguised as an asset. If the IPO happens this fall, investors will be buying a snapshot of a moment that may not repeat. The 622% figure is a single data point in a system where the only constant is that the ledger does not lie. The question is: can the interpreters read the fine print before the next quarterly report?
