Anthropic and OpenAI are growing faster than any software company in history, and the enterprise spending data explains why

Krishna Rao’s disclosure is the clearest data point in what has become a crowded field of large numbers. Anthropic’s annualized run-rate climbed from roughly $9 billion at the start of the year to north of $30 billion by the end of the most recent quarter. That is not a year-over-year comparison. It is one quarter.

Elad Gil places both companies at roughly that level simultaneously, citing rumors that OpenAI and Anthropic are each sitting at approximately $30 billion in annual revenue. The figures are consistent enough across sources to treat the order of magnitude as settled, even if the precise figures remain disputed. Marc Andreessen takes the trajectory further, arguing that Anthropic and OpenAI are currently adding more revenue per month than Meta, Google, or Microsoft, and projecting that the combined run-rate for the two companies could reach $200 billion by the end of 2026. That projection requires a great deal to go right. But the starting point it builds from is no longer speculative.

The enterprise spending numbers underneath the headline figures are where the story becomes structural. Harry Stebbings reports Anthropic’s token growth at approximately 15x in Q1 2026. Nico Laqua discloses that his company is spending $400,000 per month on Anthropic, with zero spend on OpenAI. A speaker on the same program, citing Marc Benioff’s own public statements, reports that Salesforce expects to reach $1 billion in Anthropic token spend within two years. These are not projections from analysts. They are first-party figures from buyers, and buyers at that scale do not redeploy budget on that timeline without strong returns on the spend.

We started the year with about $9 billion of run rate revenue and we ended the quarter with, you know, north of $30 billion of run rate revenue. Krishna Rao

Dylan Patel frames the demand picture at the model-tier level. His estimate is that the economy is currently spending $40 billion on what he calls a “46 Opus tier model,” with that figure rising to $100 billion by year-end. The underlying claim stands on its own terms: economy-wide spend on the most capable available models is compounding fast enough that a rough doubling within a single year is, in his words, “not unreasonable.”

Andrew Feldman provides the structural ceiling that makes those trajectory claims feel less like boosterism and more like arithmetic. With 47 million software engineers worldwide, each spending $50,000 to $100,000 per year on AI tokens, the software-engineering token market alone would reach $5 trillion. The figure is large enough to sound invented, but it follows directly from two inputs, headcount and per-seat spend, that are individually defensible. The implication is that even the current growth rates are early-innings relative to what the addressable market could absorb.

The secondary market is pricing in something similar. Harry Stebbings reports that Anthropic’s secondary market valuation has surged to $1 trillion. Brendan Foody goes further, saying he could see one of the two leading AI companies reaching $10 trillion in market value within five years. Those are forward-looking claims that the current revenue figures alone cannot validate. Lip Bu Tan notes that venture firms are now willing to put $1 billion into a single company, something he describes as previously unheard of in the VC business. The capital formation around these two companies is itself a signal about how institutional investors are reading the trajectory.

Chris Degnan offers the most useful frame for what this growth rate means competitively. In the current AI market, he argues, doubling revenue year-over-year is no longer sufficient for survival. A company that goes from $50 million to $100 million in a year might still go out of business. That is a significant recalibration of what “fast growth” means in this environment, and it clarifies why the $30 billion run-rate figures, and the trajectory Andreessen describes, are treated by market participants as something other than ordinary software success. The bar for relevance has moved. The question now is whether the companies currently clearing it can sustain the pace long enough to lock in the enterprise relationships that make the valuation arguments self-fulfilling.