Anthropic Files, the Craziest Thing in the S-1 Isn't $6.6M Revenue per Employee

On October 26, Anthropic filed its S-1 with the SEC, planning to list on Nasdaq under the ticker ANT.

First, the numbers. Full-year 2023 revenue was $41 million, and the first half of 2026 saw $16.3 billion. Greg Isenberg says no one has an existing mental model for a growth curve of this magnitude.

A few key points:

Revenue per employee of $6.6 million (TTM). A year ago it was $1.4 million. The company has about 5,300 employees. One commenter said, "Half the employees might be named Claude." Another said, "$6.6 million per person, and I can't even get approved for a second monitor."

Net revenue retention of about 900%. Customers spend about 9x their initial amount after a year. SaaS companies issue a press release when they hit 120%.

Large customer growth: Customers spending over $100,000 annually grew from 950 to 11,900 in a year. Of these, 1,900 spend over $1 million per year, and about 250 spend over $10 million. More than half of the Fortune 500 are on the list, but on average they spend less than 0.1% of their total expenses on Anthropic.

First positive quarterly operating profit. Q2 2026 operating profit was $974 million. Compute and infrastructure costs as a percentage of revenue fell from 1,056% in 2023 to 55% in the first half of 2026, and from 251% to 48% within five quarters.

Net loss for the first half was $80.7 billion, of which $72.4 billion was a fair value change on convertible notes—a book item from rapidly rising valuation, not an operating loss. After the IPO, when the notes convert to equity, this line item will naturally disappear.

Cowork is being adopted faster than Claude Code was. Anthropic is already integrating Cowork into Claude.ai. Greg's interpretation: The chat box era is ending, and the "do the work for me" era has arrived. Claude Code proved that agentic workflows work for developers, and Cowork wants to replicate the same thing for all knowledge workers.

48% of revenue is generated through the three major cloud marketplaces. AWS, Google Cloud, and Azure are both channels and competitors. The S-1 also disclosed the run-rate formula: trailing 28-day usage revenue × 13 + subscription monthly fee × 12. This formula will likely appear in every AI startup's deck next week.

The governance structure is the most counterintuitive part. Seven co-founders hold one share of Class F common stock through Founder LLC, with 50.1% of voting power. It has no economic rights, does not participate in dividends, and does not participate in liquidation distributions. If there are fewer than three founders, this share will be redeemed for no consideration. The Long-Term Benefit Trust, meanwhile, has the right to elect a majority of the board.

One netizen commented: "On one hand, they say AI poses an existential risk to humanity; on the other, they spend $750 billion to accelerate it." Sarcasm aside, Anthropic's logic is: precisely because the risk is high, more capital is needed for safety research. The market will price what this stance is worth.

Thousands of employees are about to gain liquidity. San Francisco will likely see a wave of new AI startups and angel investing. Some compare it to Facebook's IPO, but the scale is not in the same league.

The biggest growth opportunity Anthropic itself points to: the gap between Claude's capabilities and the breadth of actual usage. The model runs faster than the market, and a huge layer of implementation work is needed in between. Dentists, landlords, and plumbers haven't adopted it yet. For startups, this is a clear signal.

Risk factors are also written realistically. Over the next decade, compute procurement and lease commitments approach $800 billion; some commenters directly say the math on CAPEX returns doesn't add up. Customer spending drops about 20% around holidays. Q3 growth slowed partly due to cracking down on unauthorized distillation (involving about 5-7% of revenue) and proactive price cuts (Opus was cut by 20%).

This is not a curve that always goes up. But it is indeed a prospectus where every number challenges SaaS common sense. Whether it's worth buying depends on whether you believe the biggest assumption: that the gap between model capabilities and market implementation will eventually be filled. And, given 50.1% voting power, what position you accept.