Meta's biggest AI rival might also be its most important supplier.
Meta quietly projected it could spend up to $10 billion a year on Anthropic's AI models — even as Mark Zuckerberg publicly trashed the company. According to sources cited by The New York Times, Meta's internal projections assumed as much as $10 billion in annual spend on Anthropic's models, a figure that sits awkwardly alongside Zuckerberg's very public criticism of his lab rival. The report, published Aug 27, frames the two companies as "frenemies": fierce competitors in the race to sell frontier AI, yet deeply intertwined economically.
The revelation matters because it exposes the gap between the AI industry's public posture and its internal economics. Meta is spending aggressively to build out its own Llama models and compete head-to-head with Anthropic and OpenAI, yet its internal planning already counts on buying the rival's models at a massive scale. That's not some abstract hedge — a $10 billion annual projection makes Meta one of Anthropic's largest customers in waiting, right up there with the hyperscalers burning cash on model access. It also reframes the earlier reporting from July, when the two were reportedly in talks for a separate $10 billion deal under which Meta would lease computing power to Anthropic; the new figure points the other way, with Meta as buyer.
The takeaway is a reminder that model quality still gates the whole stack. For all the talk of self-reliance and homegrown frontier models, even a company spending tens of billions on its own AI buildout quietly budgets for the possibility that it needs a rival's — not its own — models to get the job done. The tension between Zuckerberg's bravado and his spreadsheets is the real headline here.
Australian police arrested two alleged members of TeamPCP, the hacking collective behind supply-chain attacks that hit Mercor, OpenAI, Microsoft, and GitHub. The AFP said the two men — arrested Wednesday morning and set to appear in Perth Magistrates Court — face a combined 14 cybercrime charges, following a joint investigation with the FBI and Western Australia Police. TeamPCP has been linked to a sprawling series of open-source software supply-chain breaches, including the notorious compromise of GitHub that shut down its millions of repositories, and reportedly harvested credentials for some 2,500 organizations.
The arrests matter because they mark real-world accountability for a group that had, in the words of one security researcher, "woke up Microsoft to the fact that they had become negligent in terms of security." TeamPCP is a far cry from a sophisticated state actor — KrebsOnSecurity's send-off notes the alleged leader had terrible operational security, drunk-driving himself into identifiable clues — yet its members breached some of the most valuable targets in software. The episode is a blunt illustration of the supply-chain era's enemy: scattered, under-trained assailants armed with AI-assistance tools and an outsized appetite for disruption, capable of causing damage wildly disproportionate to their skill.
What to watch: whether Meta's $10 billion projection crystallizes into an actual Anthropic contract — and what it means for OpenAI as its two biggest rivals cozy up.
Here's a question worth sitting with: do the biggest AI labs' public jabs match what their own planning documents actually say? Tell us in the comments.
Sources: The New York Times · Techmeme · Krebs on Security · TechCrunch
