Meta projected that it could spend $10 billion on Anthropic’s AI
Mark Zuckerberg, CEO of Meta, recently took aim at competitor Anthropic.
Without naming the AI start-up or its chief executive Dario Amodei, Mr Zuckerberg said leading AI labs were trying to consolidate power while painting the future as “full of doom”.
“If these laboratories lead,” wrote va Essay of 6500 words published this month, “then the balance of power favors larger institutions over individuals”.
But even as Mr. Zuckerberg has publicly criticized Anthropic, his own company has privately helped its rival rise to power.
Behind the scenes, Meta, which owns Facebook and Instagram, has become a big user of Anthropic’s AI products — so much so that it has become one of Anthropic’s biggest customers, according to five people with knowledge of the companies.
At one point this year, Meta projected internally that it could spend as much as $10 billion a year on Anthropic’s AI models, said two of the people, who declined to be identified to discuss private information. That would make up the bulk of Anthropic’s annual revenue, which the startup estimated would top $65 billion in July.
Meta and Anthropic have long competed with each other, but the extent to which they are financially linked has not been previously disclosed. Their plight offers a window into the friend-foe relationships permeating Silicon Valley’s AI race, in which tech titans can publicly malign their rivals while giving them billions of dollars behind the scenes, even as they argue they may pose an existential risk to humanity.
Popular AI startups include Google and Amazon, which have pledged to invest $73 billion in Anthropic, even as they develop their own AI models. Microsoft was an early investor in OpenAI, but the companies now emphasize their independence from each other. Chipmaker Nvidia has multibillion-dollar partnerships with Meta and Google, even though those companies work on their own chips.
Meta’s relationship with Anthropic seems particularly complicated.
Meta is so dependent on Anthropic’s AI models that it has used them to power and test an upcoming AI product called Hatch, four people familiar with the matter said. Mr. Zuckerberg described Hatch as Meta’s next breakthrough, a personal “agent” that works “24/7 on your behalf to help you achieve your goals and improve your life, your health, your relationships, your finances.”
Meta scaled back some Anthropic spending this summer, but is still spending hundreds of millions of dollars a month on start-up tools, two people with knowledge of the matter said.
Meta executives are aware that their spending could affect Anthropic as the startup heads toward a successful initial public offering that could value it at $2 trillion, two people with knowledge of the company said. Nat Friedman, Meta’s head of AI product, told some employees that if Meta uses only its own coding tools or tools from OpenAI, it could reduce Anthropic’s revenue ahead of its public offering.
“It’s like watching generals fight on a battlefield,” said Tomasz Tunguz, an investor at Theory Ventures, a venture capital firm. “Everyone is trying to destabilize the IPO”
Meta and Anthropic declined to comment. (The New York Times has sued OpenAI and Microsoft, alleging copyright infringement of news content related to AI systems. Both companies have denied the suit’s claims.)
Meta spent much of last summer trying to lure AI researchers away from rivals with dangling nine-figure compensation packages. While it has hired some high-profile AI researchers from OpenAI and Google, it has largely failed to recruit from Anthropic, three people with knowledge of the companies said.
Meta’s spending at Anthropic increased at the beginning of the year when its engineers began to heavily use Claude Code, Anthropic’s popular AI coding tool. Back in April, Meta employees were already competing on internal leaderboards to see who could use the most anthropic “tokens,” which are AI units roughly equal to one word. The practice was called “tokenmaxxing”.
Around this time, Meta executives discussed a $10 billion spending projection for Anthropic. But as token prices skyrocketed, the Meta removed token rankings. In June, the company told employees it was on track to spend billions using AI this year and would create a better system to manage that spending.
That same month, Anthropic also approached Meta with an offer to buy up to $10 billion worth of computing power from its data centers over two years, The Times previously reported. No deal has been announced.
Meta has since released an update to its Muse Code AI coding tool, which its employees are increasingly using instead of Anthropic’s tools. And while Meta has been testing Hatch with Anthropic’s AI, the product that will be unveiled publicly will be powered by Meta’s latest AI model, one person familiar said.
Mr. Zuckerberg’s company is also developing a new artificial intelligence model that is said to be as powerful as Anthropic’s high-end models. It’s unclear whether this model, internally called Watermelon, will catch up to Anthropic’s.
In July, while developing Watermelon, Meta suspended and later resumed a phase of AI development called “pre-training,” delaying its release until at least October, four people with knowledge of the matter said. The meta has not announced when Hatch or Watermelon will launch.
Erin Griffith and Sri Muppidi contributed reporting from San Francisco.