Analysis
Meta and Microsoft are both pulling back internal employee use of Anthropic's Claude even as their broader financial relationships with the company keep expanding, The Information reported October 5. Meta's internal engagement with Claude Code has roughly halved, from about 60,000 employees to approximately 30,000, while Microsoft has cut its planned 2026 Anthropic spending by more than a third, according to reporting corroborated across multiple outlets.
The Bill Got Too Big To Ignore
Both pullbacks are framed internally as cost control rather than a verdict on Claude's quality. Microsoft had budgeted more than $1 billion for internal Anthropic usage in 2026; usage-based token billing on Claude Code pushed actual costs high enough that Microsoft pulled most internal Claude Code licenses and redirected engineers toward GitHub Copilot CLI and its own in-house MAI models. Meta's case is similar: engagement roughly halved, but the company still spent over $105 million on Claude Code in a single recent 28-day window -- meaning the remaining 30,000 users are either heavier individually or Meta hasn't yet throttled total spend to match the headcount cut.
“## The Bill Got Too Big To Ignore Both pullbacks are framed internally as cost control rather than a verdict on Claude's quality.”
The timing matters because both companies are also trying to stand up credible in-house alternatives: Meta's own AI assistant and Llama model family, and Microsoft's MAI models alongside its GitHub Copilot lineup. Pulse has covered Anthropic's rise to a roughly $965 billion valuation on the back of exactly this kind of enterprise coding demand -- Claude Code's usage-based pricing is a feature when it drives revenue and a liability the moment a large customer's engineering org actually reads the monthly bill.
The headline numbers overstate how clean this break actually is. Microsoft's up-to-$5 billion direct equity investment in Anthropic is untouched, and Anthropic's own $30 billion commitment to buy Azure compute capacity keeps growing -- Microsoft is simultaneously cutting Anthropic as an internal tool vendor while deepening its position as Anthropic's infrastructure landlord. Customer spending on Claude through Azure and Amazon Bedrock, as opposed to direct internal employee licenses, is reportedly still growing. This reads as dual-sourcing and cost discipline inside two large engineering orgs, not Big Tech turning against Anthropic.
Anthropic has not publicly commented on the specific usage figures; a spokesperson has previously said enterprise and API revenue through cloud marketplaces continues to grow regardless of any single customer's internal deployment choices, a framing that lines up with Microsoft and Meta's own statements that this is an efficiency move rather than a vendor breakup.
For AI coding-tool startups -- Cognition, Replit and the Anysphere/Cursor team SpaceX now owns -- the lesson is that usage-based pricing at enterprise scale eventually gets a line item and a budget review, no matter how good the product is. Anthropic's IPO prospectus, reportedly disclosing an operating loss near $8 billion against $4.6 billion in revenue, makes Claude Code's enterprise retention numbers -- not just its top-line growth -- the thing diligence should actually chase before the company's expected November listing.
