The Second Line Item
Meta spent $105 million on Claude Code in a single month. When a second line-item starts competing with employee salaries, what happens to the people on the first one?
Meta spent more than $105 million on Claude Code in a single month.1
Lately I have been thinking about token spend, and what it means for the future of my profession.
For many years the biggest line-item on a tech company's R&D budget was employee salaries. Serving software cost almost nothing, which let software companies run at margins orders of magnitude larger than those of, say, a retail business. Meta earns about $2.8 million of revenue per employee. Walmart manages $340,000.2 When your only real input is people, you can afford to pay a lot for people.
But what happens when a second line-item starts competing with employee salaries? I'm talking about AI, of course.
On October 26 Microsoft offered buyouts to roughly a quarter of its workforce, citing AI and data centre costs.3 I keep seeing reports claiming that AI isn't the cause of layoffs in recent years. But the vibe-shift in the industry is impossible to ignore.
Here's some crude math: say the average tech employee makes $100,000. Once your company hits $100,000 in Claude Code spend, all else being equal, there is room for one fewer employee in the budget. A million dollars of tokens is ten employees who no longer fit on the payroll.
So the question that matters for anyone hoping to still be employed in five years is a ratio: how much will future employers spend on salaries compared to tokens?
For most companies right now I guess that's probably something like 100:1. For an AI-native startup it could be closer to 1:1; maybe some even spend more on AI than on human salaries. What the Microsoft and Meta reporting suggests is that the gap is closing.
$105 million, across roughly 30,000 Claude Code users,4 is about $45,000 a year per engineer for a single tool. Microsoft's Cloud and AI division cut its per-employee ceiling from $100,000 a month to roughly $10,000: $120,000 a year of permitted token spend, an amount comparable to an average salary.5 One Microsoft employee, in a leaked internal spreadsheet, self-reported $28,000 of token spend in a single month, which, annualized, is plausibly more than their base salary.6
Some honest caveats. Budgets aren't strictly zero-sum, and a team that ships more can spend more on tokens and still hire. That $28,000 is an outlier rather than a norm -- the median in the same spreadsheet was about $300 a month, and only around 350 people reported at all.6 Microsoft's total payments to Anthropic also reportedly stayed flat through all of this, because external customer use absorbed the internal cuts.7 The money didn't vanish. It moved. But a CFO watching a new cost line grow like this starts asking which old line it is growing out of.
The spread from 100:1 down to 1:1 isn't a law of physics. It's a spread of engineering choices. But if AIs continue to get more and more capable, I expect the budget ratio to skew in favour of AIs. Does that mean fewer people will be employed in tech? That is still very unclear. I think it will be an interesting question to revisit in five years.
What's the salary-to-token ratio where you work? I suspect most people have no idea, which is itself part of the problem.
Footnotes
- The Information, October 5, 2026, via Tech Startups and RS Web Solutions. ↩
- Visual Capitalist and Bullfincher. ↩
- Deadline, Axios and SiliconANGLE. ↩
- Tech Startups. ↩
- The Information, via Crypto Briefing and Tech Startups. ↩
- Business Insider, via Futurism. ↩ ↩2
- Tech Startups. ↩