Anthropic spends $100,000 a month per employee, and its S-1 shows the strategy is paying off

Anthropic spent more than $1.7 billion on personnel from July through September for a workforce of about 5,300, according to an amended S-1 the company filed Monday. The S-1 is the registration document a company files with US regulators before going public, and Anthropic's IPO is expected later this month. Business Insider calculates that the cost averaged more than $322,000 per worker for the quarter. That is above $100,000 a month, when six figures a year is still the classic American salary milestone.

Most of that money is equity, not cash. Stock-based compensation, the accounting value of the shares granted to staff, came to $1.1 billion in the quarter. That leaves roughly $600 million for salaries, bonuses, benefits and payroll taxes combined. Over the first nine months of 2026, personnel costs reached $4.8 billion. The filing also shows more than $32 billion in stock-based compensation not yet recognized as of the end of September. A successful listing would push those costs up in the coming quarters.

Our thesis is that the scary headline is the less important number. The ratio matters more. In 2023, according to the filing, Anthropic spent more than twice as much on people as it earned in revenue. In the first nine months of 2026, personnel costs were just 15% of revenue. By simple arithmetic, that implies revenue on the order of $32 billion over those nine months. Business Insider puts Anthropic's annual revenue per employee at $6.6 million. Estimates for Meta and Google are $3.2 million and $2.5 million. Anthropic employs 5,300 people. Meta has about 67,000 and Broadcom more than 30,000. “Talent density matters more than talent mass,” the company writes in its S-1.

The short-term picture is uncomfortable. A seat at Anthropic has become one of the century's fastest routes to extreme wealth, and the IPO will create a new group of multimillionaires, many of whom joined only a couple of years ago. That deepens the inequality we already track in AI. A small elite of researchers and engineers captures an outsized share of the value, while back-office work elsewhere comes under pressure from the very tools these teams build. The headcount itself grew fivefold between the end of 2024 and September 2026, so the case for a lean company is relative, not absolute.

Our reading: the filing is an early, audited preview of what AI-native organizations may look like. They are small, highly leveraged teams in which software does work that once took tens of thousands of people. That fits our view on jobs, which is that AI changes which work matters more than it simply removes jobs. It also has a less comfortable side. If $6.6 million of revenue per head becomes the standard, the gains from the transition will go first to capital and to a few specialists. Over the long term, that productivity is the raw material for abundance. Spreading it widely will take policy choices, not just stock options. Investors should also remember that the $32 billion of unrecognized equity is a real future cost, even though it does not show up in today's margins.