Anthropic's $6.5B ARR and OpenAI's IPO Race: Who Wins?

Anthropic dropped a bombshell on investors over the weekend: its annualized revenue run rate hit $6.5 billion as of end of July, a seven-fold increase from a year ago. For context, that is roughly the GDP of Belize, achieved by a company that did not exist five years ago. And it is now comfortably ahead of rival OpenAI, which last reported $4 billion in ARR, according to multiple Bloomberg and CNBC sources.

The sequential acceleration is the part that should make your jaw drop. Anthropic reported a run rate of $4.7 billion in May, meaning it added nearly two billion dollars in annualized revenue in roughly ten weeks. Preliminary Q2 revenue came in at $1.15 billion, a 14x year-over-year increase. OpenAI, meanwhile, is growing too, with Greg Brockman confirming 20% month-over-month revenue growth in July and enterprise revenue up 32%. But Anthropic is the one currently meeting with bankers ahead of what looks like a very serious IPO process.

The AI infrastructure trade is not just hype at this point. Bank of America economists noted that hyperscaler capex plans show no signs of slowing, with growth expected around 35% again next year. Bloomberg Businessweek flagged the industry is adding 20 gigawatts of computing capacity annually at roughly $50 billion per gigawatt, implying close to $1 trillion in annual capex across the sector. That is not a rounding error. That is the whole point.

The bear case deserves airtime though. Fundstrat's Tom Lee crowd was on Bloomberg defending off-balance-sheet AI spending commitments of potentially $3 trillion as normal financial structuring with cancellable contracts. Critics are not so relaxed. And with only 7% of companies fully implementing AI, the gap between infrastructure spend and actual enterprise productivity is still enormous. Someone is going to be left holding a very expensive GPU when the music stops.

Meanwhile, Cerebras Systems surged after news broke that its chips will power OpenAI's ultrafast GPT-4.5 Soul processing mode, with OpenAI committing to 750 megawatts of Cerebras capacity by 2028. The new mode delivers 14x faster speeds than standard processing, targeting latency-sensitive use cases. Cerebras is freshly listed so quant ratings are sparse, but the OpenAI partnership is not a small endorsement.

Two private AI giants are sprinting toward a public market that is already priced for perfection, and whoever blinks first on the IPO filing is going to set the valuation benchmark for the entire sector. No pressure.