Why Anthropic losing token share at $META and $MSFT is bullish for the AI infra trade Meta and Microsoft have reduced their internal employee use of Claude. Microsoft cut projected internal Anthropic spending by >1/3, while Claude Code users at Meta fell from ~60,000 to 30,000. The immediate revenue impact may be contained, but the strategic signal is more concerning: Anthropic’s customers increasingly have both the incentive and capability to replace Claude. This matters even more if you look at the customer structure at Anthropic showing that 2 customers generated ~1/4 of Anthropic’s revenue (as of end 2025). Those 2 are probably Cursor and GitHub Copilot. If that’s the case then we have early evidence that customers that have their own models start becoming less dependent on Anthropic while at the same time both Cursor (Grok) and GitHub Copilot (MAI-Code) have internal model options to replace Anthropic’s Claude. I think this is more bullish than bearish for the AI infra buildout as it diversifies the customer base of compute and I dont think frontier AI will be disrupted anytime soon (look at most recent OpenAI ARR growth). However, I recognize that Anthropic has $518B worth of chips and compute orders, of which $413,7 are take or pay, so alot of the AI infra money depends on them. My view is that they need to get closer to the customer and own distribution, otherwise replacement risk is real.