Zuckerberg Slams Anthropic's Monopoly, Yet Meta Pays Them $10 Billion Annually
Nashnova编辑部
Zuckerberg publicly attacked Anthropic for "concentrating power," yet Meta internally projected spending up to $10 billion a year on Anthropic's models — This means → Meta waves the open-source flag with one hand and votes for a rival's closed-source product with the other.
How much is Meta actually spending on Anthropic?
Meta internally projected annual spending on Anthropic AI models as high as $10 billion.
Anthropic estimated in July that its full-year revenue would exceed $65 billion; Meta's potential outlay is a significant share.
Meta has cut back some spending this summer, but still pays hundreds of millions of dollars a month for Anthropic tools.
Why were Meta engineers racing to burn tokens?
Earlier this year, Facebook and Instagram engineers began heavy use of Anthropic's AI coding tool Claude Code.
By April, employees were competing on an internal leaderboard to see who could consume the most Anthropic tokens — a phenomenon dubbed "tokenmaxxing."
This means → Claude Code's real-world developer experience was good enough to trigger voluntary usage wars.
In June, Meta killed the leaderboard as costs surged, telling staff that AI spending this year would hit the billions.
Zuckerberg's next big product runs on Anthropic under the hood?
Meta is using Anthropic models to power and test an upcoming AI product called "Hatch" — a personal agent that works around the clock on a user's behalf.
Zuckerberg has positioned Hatch as Meta's next breakout product.
In plain terms = he calls Anthropic a monopolist on stage, then uses Anthropic's technology to build his most important new product backstage.
Hatch will switch to Meta's own models at public launch; Anthropic is the stopgap.
Can Meta's in-house tools replace Anthropic before its IPO?
Meta is developing a new model code-named "Watermelon," aimed at matching Anthropic's frontier capabilities.
Watermelon's pre-training was paused in July; the schedule has slipped, with the earliest release now after October.
Meanwhile, internal adoption of Meta's own coding tool Muse Code is climbing, gradually displacing Claude Code.
This means → if Watermelon slips much further, Meta will struggle to dent Anthropic's revenue figures before the IPO.
What does this "frenemy" dynamic mean for Anthropic's listing?
Meta AI product head Nat Friedman told staff that switching entirely to in-house or OpenAI tools could depress Anthropic's revenue ahead of its IPO.
Anthropic is preparing an IPO that could value the company at $2 trillion.
In June, Anthropic proactively offered to buy up to $10 billion in compute over two years in exchange for access to Meta's data-center capacity; no deal has been announced.
This reflects a deeper vulnerability: Anthropic's revenue growth leans heavily on a handful of mega-clients — Meta's stay-or-go decision directly shapes the IPO story.
Why is every Silicon Valley giant playing "cooperate and compete" at the same time?
Google and Amazon have committed $73 billion in investment to Anthropic, while still aggressively advancing their own AI models.
Microsoft was OpenAI's most important early backer; the two companies now emphasize their independence from each other.
Nvidia has multi-billion-dollar deals with both Meta and Google — both of which are also developing their own chips.
In plain terms = Silicon Valley's AI landscape is "sign the check with one hand, sharpen the knife with the other" — everyone is making sure no single supplier locks them in.
市场有风险，内容仅供研究参考，不构成投资建议。