Meta Platforms Inc., led by CEO Mark Zuckerberg, has taken a complex stance toward the artificial intelligence startup Anthropic, reflecting the intricate dynamics of competition and collaboration in Silicon Valley’s AI landscape. Recently, Zuckerberg criticized leading AI labs—including but not explicitly naming Anthropic and its CEO Dario Amodei—for concentrating power while projecting a pessimistic outlook on the future. He argued that such labs could shift power disproportionately toward large institutions rather than individuals.
Despite these public criticisms, Meta has become one of Anthropic’s largest clients. Sources with knowledge of the matter reveal that Meta’s internal estimates at one point projected annual spending of up to $10 billion on Anthropic’s AI models. This amount would account for a significant share of Anthropic’s anticipated annual revenue, which the startup estimated to exceed $65 billion this year.
This paradoxical relationship highlights a broader trend among major technology players, where companies publicly challenge rivals while simultaneously engaging in extensive financial and operational partnerships. For example, Google and Amazon have committed $73 billion in investments to Anthropic even as they develop their own AI technologies. Meanwhile, Microsoft, an early investor in OpenAI, maintains a publicly independent stance, and chipmaker Nvidia collaborates with Meta and Google despite their competing chip projects.
Meta’s reliance on Anthropic’s AI has been especially notable in its development and testing of a forthcoming AI product named Hatch. Zuckerberg has described Hatch as a personal AI assistant designed to operate continuously to support users in various aspects of life, including health, relationships, and finances. While Meta has scaled back some spending on Anthropic’s tools recently, it continues to invest hundreds of millions of dollars monthly in the startup’s technology.
Meta executives are reportedly aware that their purchasing decisions could impact Anthropic’s valuation ahead of its anticipated initial public offering (IPO), which could value the startup as high as $2 trillion. The company’s head of AI product, Nat Friedman, has suggested to employees that using Meta’s own AI tools or those from other providers like OpenAI could reduce Anthropic’s revenue prior to its public offering. Observers liken the situation to competing generals maneuvering strategically ahead of an IPO.
Meta’s internal usage of Anthropic’s AI tools, particularly Claude Code—a popular coding assistant—spiked earlier this year, leading to initiatives such as internal competitions measuring the volume of AI use, known as “tokenmaxxing.” Rising costs later prompted the removal of these leaderboards and a re-evaluation of spending management. Additionally, Anthropic has reportedly sought to purchase up to $10 billion in computing capacity from Meta’s data centers over a two-year period, though no agreement has been publicly disclosed.
In recent months, Meta has introduced updates to its own AI coding tool, Muse Code, which has seen increased adoption by its engineers, partially supplanting Anthropic’s offerings. The forthcoming public version of Hatch is expected to be powered by Meta’s latest AI model rather than Anthropic’s. Furthermore, Meta is developing a new AI model internally referred to as Watermelon, intended to rival the capabilities of Anthropic’s advanced systems. However, the release of Watermelon has been delayed, with certain development phases paused and resumed as recently as July.
Neither Meta nor Anthropic have commented publicly on the nature of their commercial relationship or ongoing AI projects. The situation encapsulates the fluid and often contradictory alliances characteristic of the competitive AI industry, where cooperation and rivalry coexist amid efforts to shape the future of artificial intelligence.