There are moments when a company’s numbers stop you cold. Anthropic generated more than $11.5 billion in revenue during the second quarter of 2026, up from $787 million in the same quarter a year earlier and $4.73 billion in the first quarter of this year. That is a fourteen-fold year-over-year move. It is also the kind of growth rate that makes intelligent investors simultaneously excited and suspicious.
The Financial Times reported that Anthropic told a small group of shareholders it will post an adjusted operating profit this quarter, marking a second consecutive period of profitability as it prepares to go public. Bloomberg has separately reported the company is trending toward annualized revenue above $65 billion given its current momentum. For a business that was effectively a startup experiment four years ago, those numbers are historically unusual.
But the question worth asking right now is not whether Anthropic is growing fast. It clearly is. The question is what kind of business it actually is once the accounting layers come off.
The Financial Times reported that gross margins exceed 80% before factoring in revenue-sharing payments to partners such as Amazon and training costs for AI models, according to people familiar with the matter. That caveat carries significant weight. Anthropic’s reported margin can exclude two of the largest cost buckets in frontier AI economics: the wholesale revenue splits it owes channel partners and the ongoing GPU and compute infrastructure required to train and fine-tune models. A disciplined investor reading that sentence would immediately ask what the margins look like after those deductions are included.
OpenAI has publicly challenged comparisons that treat Anthropic’s partner-channel revenue the same way as OpenAI’s, noting that the two companies can record partner-distributed sales differently. When Anthropic’s models are accessed through a cloud marketplace, one approach is to book the full end-customer spend as revenue and record partner payouts as expenses. Another approach is to report only the net share as revenue. That methodology choice becomes a public-market disclosure issue at IPO. The S-1, whenever it arrives publicly, will be the first chance for investors to stress-test these figures against audited financials.
The commitment to AWS alone runs to more than $100 billion over ten years, and a Google Cloud deal has been reported as up to $40 billion. These are compute obligations that sit behind every margin calculation. Some industry analyses have said compute costs fell from 71 cents per revenue dollar in the first quarter of 2026 to 56 cents in the second, the primary driver of the margin improvement. Fifty-six cents of every dollar still going to compute is not what Charlie Munger would recognize as a wide-moat software business. It is, however, a significant improvement, and the direction matters.
The enterprise channel is where Anthropic has built its sharpest advantage. The company has leaned heavily into enterprise contracts, particularly in coding assistance and agentic workflows, where customers are willing to sign multi-year agreements and consume tokens at high volumes.
Then there is the peculiar backdrop to this weekend’s disclosures. Dario Amodei published an essay Saturday calling for AI companies to deliberately slow the pace at which they improve their models, arguing that AI advances are moving faster than researchers’ ability to understand and control them. Pausing would save the firm billions in model-training expenses but could help competitors close the distance. A CEO calling publicly for industry restraint while privately telling shareholders to expect a second straight profitable quarter is a tension the public prospectus will need to resolve.
The company has reportedly settled on Nasdaq for its public listing, and reports have suggested the market may be asked to entertain a valuation of more than $2 trillion. At $2 trillion against a $65 billion run rate, the implied multiple demands that Anthropic’s 80% headline margins survive contact with full accounting scrutiny, that compute costs keep compressing, and that enterprise revenue continues compounding at a pace no software company has sustained at this scale.
The growth is real. The October listing will determine whether the business underneath it is as durable as the numbers suggest.
