Anthropic and OpenAI are already among the largest software businesses on the planet, and the gap is widening
Annualized run-rate figures from people close to both companies describe growth that has no modern precedent in enterprise software. The revenue is already on the books. The argument is only about what comes next.
Anthropic’s annualized run-rate revenue went from roughly $9 billion at the start of the year to north of $30 billion by the end of the most recent quarter, according to Krishna Rao, who sits close enough to the company’s finances to cite the figures with precision. That is not a trajectory; it is a vertical.
Chamath Palihapitiya places the Anthropic figure even higher. By his account, the company started the year at $10 billion of annualized recurring revenue and crossed $70 billion mid-year, against an internal forecast that targeted $100 billion by year-end. OpenAI’s ARR, he says, rose from $33 billion in May to $41.3 billion in July, with its year-end forecast revised upward from $60 billion to around $75 billion. These figures do not all agree precisely with one another, and they should be read as a cluster of estimates rather than audited accounts. The direction they point is consistent. Elad Gil, noting that OpenAI and Anthropic are each rumored to be roughly around $30 billion in annual revenue, offers a more conservative read, but even that figure would place either company among the fastest-scaling software businesses in recorded history.
Marc Andreessen frames the pace in competitive terms: Anthropic and OpenAI are adding more revenue per month than Meta, Google, or Microsoft. He would not rule out a combined run rate of $200 billion between the two companies by year-end. Brad Gerstner adds that if either company exits a year above $100 billion, a three-to-five-fold increase the following year is plausible. A year ago, Gerstner observes, the field looked like five major labs. Now it looks like a top two and everyone else.
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
The demand-side arithmetic behind these figures is not mysterious. Andrew Feldman calculates that 47 million software engineers exist in the world, and that figure alone implies a $5 trillion addressable market for software engineering token use. Dylan Patel estimates the economy is already spending $40 billion on compute at the tier of a leading frontier model, and projects that figure could reach $100 billion by year-end. Neither number requires heroic assumptions about new use cases. Both rest on current adoption extended forward.
Sam Altman’s framing of the business model is the clearest statement of why the margin structure differs from conventional software. OpenAI, he argues, will have so much usage that it does not need to be a high-margin business to fund model training. So much future compute will be used selling inference to customers that even modest margins on trillions of dollars of revenue will cover frontier training costs. That logic explains why usage-based, inference-heavy pricing is not just a billing preference but a strategic architecture: volume becomes the substitute for margin.
The customer-side evidence supports the supply-side claims. Brendan Foody, whose company expanded relationships with frontier labs, reported adding $300 million in net new annualized recurring revenue in 60 days. Nico Laqua noted spending around $400,000 per month on Anthropic alone, with nothing going to OpenAI. Yasser Elsaid described his company’s model usage as roughly split between OpenAI and everything else, with Anthropic and Google making up most of that second half. Individually, these are anecdotes. Collectively, they describe a customer base that has moved from experimentation to budget-line dependency.
The secondary market has responded. Harry Stebbings noted secondary market prices surging to a trillion dollars for Anthropic. Chamath Palihapitiya, citing investor Gavin Baker, said a post-IPO valuation of three trillion is a live scenario. Jason Calacanis pushed back, arguing Anthropic’s growth rate will slow considerably and that a trillion in revenue is not a near-term outcome. That dissent is worth registering: forecasts compounding at this rate have a way of colliding with capacity constraints, competitive pressure, or simple mean reversion. What is not in dispute is the revenue already on the books. A company at $30-plus billion in annualized run rate, growing at the pace Rao describes, is already one of the largest software businesses on the planet. The question of what happens next is a legitimate argument. The question of whether something extraordinary is already happening is not.