Meta and Microsoft cut internal Claude usage, Microsoft caps AI spend at $10K per employee
- Meta and Microsoft are reducing internal use of Anthropic's Claude to push employees toward their own coding tools, according to an October 5 report from The Information; Microsoft's earlier estimate of over $1B in annual internal Anthropic spend fell by more than a third.
- Microsoft slashed monthly AI spending ceilings in its cloud and AI division from $100,000 per employee to about $10,000 in most cases, while customer spending on Anthropic models through Microsoft's enterprise platforms keeps rising.
- Meta's Claude Code users dropped from about 60,000 to 30,000 this year, with the move to in-house MetaCode (over 30,000 internal users) and Muse Code (over 6,000) cited as the main cause rather than layoffs.
- Meta still spent over $105 million on Claude Code in a 28-day window, so fewer users did not translate into lower spending.
- Neither company is pulling Claude from customer-facing products, and Anthropic's reported annualized revenue pace is $65 billion.
Hacker News opinions
So they're going to fall even further behind? Buying puts on Meta and Microsoft, or maybe I just hand it off to Fable.
Microsoft is only capping employees at $10k a month, down from $100k. That's still a huge amount of tokens per person.
Microsoft is like $100bn deep into OpenAI. Of course they want everyone on their own stack.
Who's ahead again? Do they have a moat, or just a nice field?
Unpopular take, but I think AI tools should be paid for by the employee. You should know how to do your work without AI.
I could still type all my code into notepad, but nobody charges employees for their IDE. The employer pays for the tools used for the work.
Should I pay rent for my desk too? And in California, if the employee pays for the tool, the employee probably owns the IP. Lawyers hate murky.
If my employer isn't forcing a tool on me and gives me freedom, fine, I'll buy my own. But if they put up a scoreboard for who uses it, they pay for it.
The forklift should be paid for by the employee too then. They should be strong enough to move things without one.
This isn't surprising. Every big lab blocks competitor tools internally. It's a data governance thing, not a verdict on quality.
Sure, but Meta and Microsoft aren't blocking Claude, they're cutting costs and diverting a fraction of usage. Both still spend nine figures a year on it.
Limited to $10,000 per employee per month, lol. That just cuts off a few people doing absurd low-ROI things. Don't read too much into it.
$105 million on Claude Code in 28 days is $1.4B a year, and it's money going straight to a competitor. That's not small even at Meta's scale.
My company took away my Claude because it's too expensive. I feel like a reckoning is coming. The accountants are finally seeing the cost of token maxing.
Sad, opus 5.5 is really good. I'd move to other models but they all lack something. I trust opus, I just hope other labs catch up.
Most people racking up big costs are token maxing, not being efficient. You can do 99% of the work with Haiku in a collaborative setup. Newcomers just love to one-shot garbage.
We just got put on a budget. Our velocity is still twice what it was before Claude, so I doubt we go back, but efficiency is going to be the priority now.
Good luck to the accountant who tells leadership to slash AI spend. OpenAI cut prices by 60% a week ago. You think now is the time to cut?
Taking it away sounds insane. We had basically unlimited tokens bought from AWS, then they moved us back to the $100 sub to save money.