Meta and Microsoft are actively reducing their employees’ internal reliance on Anthropic’s Claude AI models, according to a report by The Information, as both companies push staff toward their own in-house tools to cut costs.
The shift comes as Anthropic prepares for a potential initial public offering, creating headwinds for the AI company despite its strong overall revenue growth.
Microsoft Slashes Claude Budget
Microsoft had projected at least $1 billion in internal spending on Anthropic’s tools earlier this year. That estimate has since been reduced by more than one-third, with company leadership directing employees to limit Claude usage and prioritize Microsoft’s own AI products.
Microsoft’s Cloud and AI division lowered the monthly AI usage cap per employee from $100,000 down to roughly $10,000 in most cases. Developers are being steered toward GitHub Copilot, which routes most queries to OpenAI models. Because of Microsoft’s business relationship with OpenAI, it does not pay usage fees for those models.
Jay Parikh, head of Microsoft’s developer tools business, said in an August internal memo: “Moving more work to OpenAI models delivers greater value from our token investment”.
The internal cuts do not affect Microsoft’s customer-facing products. Claude remains available through Microsoft Foundry, and enterprise demand for Anthropic models through Microsoft’s platform has reportedly held steady.
Meta’s Claude Code Users Halved
Meta’s internal usage of Anthropic’s coding assistant, Claude Code, has fallen from approximately 60,000 employees earlier this year to around 30,000 recently.
While spring layoffs that cut roughly 10% of Meta’s workforce contributed to the decline, the primary driver is Meta’s push to transition engineers toward its own solutions. MetaCode, an internal coding assistant, has surpassed 30,000 users inside the company. Another tool called Muse Code, based on Meta models, is being used by over 6,000 employees and has been tested externally since August.
Despite the drop in internal users, Meta spent more than $105 million on Claude Code over a recent 28-day period, indicating that reduced headcount does not necessarily translate to lower total spending.
Why This Matters for Anthropic
Anthropic is preparing for what could be a record IPO. In its prospectus, the company disclosed that two unnamed customers account for approximately 25% of its revenue.
The internal pullbacks by Meta and Microsoft present a narrative challenge for Anthropic, even if the financial impact is limited. Both companies remain major customers of Claude for their customer-facing products, and Microsoft reportedly pays at least $2 billion annually to use Anthropic models in Copilot features it provides to customers.
The broader dynamic is one of hedging. Large technology companies increasingly run a mixed portfolio: their own tools for routine work, licensed third-party models for harder tasks, and rotating evaluations of competitors as pricing and capabilities shift.
The Bottom Line
Meta and Microsoft are reducing internal employee use of Anthropic’s Claude models to cut costs and promote their own AI tools. Microsoft cut its projected internal Claude spending by over a third and lowered per-employee AI budgets. Meta’s Claude Code users dropped from 60,000 to 30,000 as it expanded MetaCode and Muse Code. Both companies continue to offer Claude to customers, keeping Anthropic’s enterprise business growing despite the internal pullback.
My Opinion
Meta and Microsoft are not cutting Claude because it is bad. They are cutting it because it is expensive and because they have their own products to sell.
Microsoft pays at least $2 billion a year to use Anthropic models in Copilot. It also owns a stake in OpenAI and builds its own AI. Meta spent over $105 million on Claude Code in a single 28-day period while telling its own engineers to use MetaCode instead. These companies are hedging. They will use whatever model works best and is cheapest, and they will switch the moment the math changes.
That is the uncomfortable reality for Anthropic as it approaches an IPO. Its product is excellent. Its revenue is growing. But two of its largest customers are actively trying to need it less. A prospectus that says two unnamed customers account for 25% of revenue is a prospectus that says the business is fragile.
The AI industry is not a market where loyalty exists. It is a market where everyone is building their own version of everything. Anthropic’s challenge is not making Claude better. It is making Claude indispensable. However, at this time, Meta and Microsoft are proving it is not.
