Anthropic Passes OpenAI on Revenue: A Lead Built on Code, Not Consumers

Omniscient

Sometime in early April, a threshold was crossed that most industry analysts had not expected to see until late summer at the earliest. Anthropic's annualized revenue run rate hit $30 billion,^[1] surpassing OpenAI's confirmed $24 billion run rate^[5] and completing one of the fastest revenue climbs in the history of enterprise software. The company that analysts - citing Epoch AI modeling - had forecast as a likely revenue peer to OpenAI by August 2026 at the earliest got there four months ahead of schedule, and did it without a single viral consumer moment.

The reversal is worth pausing on. A year ago, Anthropic was at roughly $1 billion in annualized revenue; OpenAI was at $6 billion. The gap looked structural. It wasn't. In the span of twelve months, Anthropic covered five times the distance OpenAI had built up, and it did so almost entirely through business-to-business contracts.

What drove the surge?

The clearest single driver is Claude Code. When Anthropic made the coding agent publicly available in May 2025, it entered a market that barely existed; ten months later, that market was generating $2.5 billion in annualized revenue for Anthropic alone, a figure that had more than doubled just since January.^[2] The product now accounts for an estimated 4% of all public commits on GitHub - a share that Anthropic projects will exceed 20% by year-end - and business subscriptions quadrupled over the same period. The velocity is hard to contextualize in conventional software terms: an agentic coding layer has gone from zero to a multi-billion-dollar category in less time than most software companies take to close a Series A.

Beneath the headline product, the enterprise base that funded the acceleration is striking in its concentration. More than 1,000 business customers now spend over $1 million annually with Anthropic - a figure that more than doubled between February and April alone.^[1] Eight of the Fortune 10 are Claude customers. The company that most people outside B2B circles could not name two years ago now counts the world's largest corporations as its clients.

What does the hardware deal signal?

Alongside the revenue disclosure, Anthropic confirmed a long-term infrastructure deal with Broadcom and Google that will give the company access to approximately 3.5 gigawatts of computing capacity beginning in 2027.^[1] Broadcom, working from Google's tensor processing unit specifications, will produce the chips under a supply assurance agreement running through 2031. The arrangement is explicitly designed to reduce Anthropic's dependence on Nvidia - a strategic hedge at a moment when Nvidia GPU availability is constrained and geopolitically fraught.

The deal is not simply about chips. It is a statement about Anthropic's confidence in its revenue trajectory. The Broadcom filing notes that expanded compute consumption "is dependent on Anthropic's continued commercial success" and that all three parties are in discussions with "operational and financial partners" about the deployment. In other words, the infrastructure build is contingent on the growth curve continuing - and Anthropic is betting it will.

The capital efficiency argument that changes the competitive narrative

Beneath the revenue comparison, the Wall Street Journal's reporting on confidential IPO-stage financials from both companies exposes a structural divergence that may matter more in the long run.^[3] OpenAI's own projections show $121 billion in compute expenditure in 2028, generating an $85 billion loss in that year alone, with a return to profitability pushed past 2030. Anthropic's compute costs in the same year peak at around $30 billion - a quarter of OpenAI's figure - and the company projects crossing into profit in 2028 or 2029.

This matters because the dominant assumption in AI has been that training scale determines competitive position: the company that spends most on frontier model development wins. Anthropic is now running a sustained real-world test of whether that assumption holds. So far, the evidence cuts against it. Anthropic has achieved higher run-rate revenue than OpenAI while spending a fraction of what OpenAI allocates to model training. The implication is not that model quality is irrelevant - Claude's enterprise traction clearly depends on the models being good - but that revenue density per dollar of training spend may be a more durable competitive moat than raw compute expenditure.

There is an important caveat: the two companies count revenue differently. Anthropic books sales through its cloud partnerships with Google Cloud and AWS, while OpenAI does not recognize equivalent pass-through revenue the same way.^[3] The direct run-rate comparison carries some noise. But even accounting for that methodological gap, the capital efficiency differential is real and widening.

The Pentagon lawsuit: a headwind that has not slowed growth, yet

The revenue figures arrive against an unusual legal backdrop. In early March, the Department of Defense formally designated Anthropic a "supply-chain risk" - a label normally reserved for foreign adversaries - after a standoff over the Pentagon's demand for unrestricted access to Claude for autonomous weapons applications and mass surveillance.^[4] The designation requires any company or agency doing work with the DoD to certify it does not use Anthropic's models. The General Services Administration subsequently terminated Anthropic's "OneGov" contract, cutting off access across all three branches of the federal government.

Anthropic sued the Pentagon in both California and D.C. federal courts in March, alleging First Amendment retaliation and procedural violations of federal procurement law.^[4] A federal appellate panel denied the company's motion to lift the designation in April,^[7] meaning the label stands while the litigation continues. More than 100 enterprise customers contacted Anthropic to express doubt about continuing their relationship with the company, according to the firm's own attorney. And yet the $30 billion run rate suggests those doubts, so far, have not translated into cancellations at scale.

Anthropic's chief commercial officer, Paul Smith, offered one explanation: some customers specifically value a vendor that "demonstrates its principles" in its dealings with the government.^[1] That is a remarkable commercial framing - essentially arguing that regulatory confrontation is a brand asset. Whether it holds as the litigation extends into late 2026 and beyond is one of the more consequential questions hanging over the company's IPO preparation.

What the crossover means for OpenAI's valuation story

OpenAI closed a $122 billion funding round in March 2026 at an $852 billion valuation.^[5] CFO Sarah Friar, speaking on CNBC's Mad Money shortly before the round's close, captured the investor sentiment: "It didn't matter where you went, people really believed in this AI revolution and they wanted to put their money to work behind it."^[6] That conviction was accurate when she expressed it - OpenAI's trajectory from $2 billion ARR in 2023 to $24 billion annualized today is genuinely without historical parallel. But the valuation was also priced on an implicit assumption: that OpenAI's revenue lead was durable.

Anthropic's April crossover complicates that narrative. At $30 billion run rate, Anthropic is arguably more valuable than its last funding round implied. More importantly, the crossover happened four months ahead of even optimistic analyst projections, which raises a pointed question about the reliability of the models underpinning both companies' IPO-stage valuations. If Anthropic's trajectory can outrun forecasts by that margin, what else in these projections is underestimated - or overestimated?

Both companies remain deeply unprofitable by choice. OpenAI is burning approximately $17 billion in cash this year against a projected $14 billion net loss. Anthropic's burn is proportionally large relative to its funding base, which topped $60 billion after the close of its $30 billion Series G in February 2026.^[2] The investors backing both - Amazon, Google, SoftBank, a16z - are not confused about this. They are making a specific structural bet: that compute costs per unit of useful AI output will fall fast enough, and revenue will compound quickly enough, that the interim losses become irrelevant.

The scale of individual commitments reflects the same logic. Amazon's position in Anthropic has grown from an initial $8 billion stake into one of the largest single-company AI bets on record: in late April 2026, Amazon announced an additional investment of up to $25 billion, bringing its total Anthropic commitment to up to $33 billion.^[8] A position of that size is not a conventional venture bet. It is a structural claim on which company controls the enterprise compute layer for the next decade - and it was placed after, not before, Anthropic crossed OpenAI on revenue.

Reaching $30 billion in annual revenue took Salesforce roughly two decades of compounding growth, multiple product pivots, and a string of major acquisitions. Anthropic crossed the same threshold before its third birthday as a commercial entity. Whether the underlying economics ultimately hold - whether training cost bets prove sound, whether the Pentagon litigation resolves favorably, whether Claude Code's market share translates into durable switching costs - will shape not just two companies' IPO trajectories, but the structure of the enterprise software market for the decade ahead.

Industry