Microsoft Cut Some AI Caps 90%; Meta Claude Code Users Halve

Microsoft and Meta have been working to reduce their employees’ internal reliance on Anthropic’s Claude models, The Information reported, according to a StockTwits summary of the report published on October 5. Within Microsoft’s Cloud and AI division, individual monthly AI usage caps were lowered from $100,000 to roughly $10,000 in most instances, the summary says: a cut of 90%, by our arithmetic. At Meta, internal users of Claude Code fell from approximately 60,000 earlier this year to around 30,000, it says, a halving by our arithmetic.

The report resurfaced on X on November 8, when Cointelegraph posted it as a new item. We covered two similar moves since: Apple lowered some engineers’ daily AI cap to $200 from $300 and Goldman Sachs limited Claude Code and GitHub to its engineers, both reported by Bloomberg in early November.

Business Pill · Usage-Based Pricing

An evergreen Business Engineer explainer on usage-based pricing. It teaches the general idea only and does not discuss Microsoft, Meta or Anthropic.

What Microsoft Changed

Earlier this year, Microsoft executives anticipated spending at least $1 billion annually on internal staff usage of Anthropic models, and that projected run rate has since been reduced by more than a third, according to The Information as summarised by StockTwits. Company leadership instructed employees to scale back Claude usage, introducing stricter spending limits and steering developers toward Microsoft’s homegrown infrastructure and OpenAI models via GitHub Copilot, the summary says.

Microsoft divisions are actively managing designated token budgets under new directives from senior executives, according to the summary. Microsoft continues to spend heavily on Anthropic models to power customer-facing Copilot features, it adds; the reduction applies to internal usage.

What Was Posted on X

Cointelegraph posted this on 8 November as a new item, citing The Information. The StockTwits summary we read is dated 5 October, and the @kimmonismus post on X shows the report was out by then. The post’s ‘by a third’ refers to Microsoft’s projected internal Claude spending, which the summary says was reduced by more than a third from at least $1 billion a year.

What Meta Changed

Internal usage of Claude Code at Meta fell from approximately 60,000 users earlier this year to around 30,000 recently, the summary says. Spring workforce reductions contributed to part of the decline, it says, while a primary driver has been Meta’s push to move engineers to its own tools: MetaCode, which has over 30,000 internal users, and Muse Code, a Claude Code alternative built on Meta’s own models. We covered Muse Code’s launch in August.

Meta also used Claude while developing its consumer Muse AI agent, then switched to its own internal models for the product’s launch, according to the summary. Our earlier piece on Meta’s internal token use covered The Information’s June report on the same company.

Four Companies, One Direction

Set side by side, the four reported moves each cut a different measure. Microsoft’s Cloud and AI division cut a monthly per-person cap by 90%, by our arithmetic of the summary’s figures. Meta’s Claude Code user count halved, by our arithmetic of the summary’s figures. Apple’s daily cap for some engineers fell by a third, by our arithmetic of Bloomberg’s figures. Goldman Sachs narrowed access to its 12,000 engineers out of 47,500 staff, which leaves about three in four outside the group, by our arithmetic in that piece. These are different measures, so we do not add or average them.

Why It Matters for Anthropic

High deployment costs and privacy considerations have led a range of enterprise clients to evaluate lower-cost alternatives, switch to open-source models, or renegotiate terms, the StockTwits summary says, calling the internal cuts by major corporate clients “notable headwinds” for Anthropic as it prepares for an IPO. The same summary says Anthropic has recorded strong overall revenue growth and has expanded its base of enterprise clients spending $100,000 or more a year.

We have reported on the IPO plan in Anthropic’s $80–100B raise and on its compute commitments in the TPU share of its named compute deals. On the market side, Menlo put Claude Code and Codex at 70% of AI coding spend.

The Structural Read

As we read the StockTwits summary, the cuts target internal staff use, not the products: Microsoft still spends heavily on Anthropic models for customer-facing Copilot features, it says.

As we read the four reports together, large companies are putting budgets and access rules around AI coding tools that were rolled out broadly, each on a different lever: caps, user counts, daily allowances and who gets access.

As we read it alongside our IPO coverage, the summary itself frames these internal cuts by major corporate clients as headwinds for Anthropic as it prepares to list.

Three Implications

FOR ANTHROPIC The StockTwits summary calls the internal cuts by major corporate clients notable headwinds ahead of the IPO, while saying Anthropic has recorded strong overall revenue growth.

FOR MICROSOFT AND META Both are steering staff to alternatives they control or partner on: OpenAI models via GitHub Copilot at Microsoft, and MetaCode and Muse Code at Meta, according to the summary.

FOR WHAT TO WATCH Neither a current dollar figure for Microsoft’s internal spend nor the split of Meta’s user drop between layoffs and the tool switch is in the summary, so the size of the shift is not established.

The Business Engineer Lens

This story maps onto the Business Engineer framework The CFO’s Guide to the Token Economy.

The essay says “As companies move from chatbots to thousands of autonomous agentic workflows running 24/7, the line item that used to be a rounding error becomes the budget.”

As we read it, the per-person caps and user cuts reported at Microsoft, Meta and Apple are that line item being managed. The essay does not discuss Microsoft, Meta or Anthropic.

What Is Not Established

We did not read The Information’s article, which is paywalled; every Microsoft and Meta figure here is as given in the StockTwits summary of it. The StockTwits summary gives no current dollar figure for Microsoft’s internal spend, so the size of the cut in dollars is not established. How much of Meta’s fall in Claude Code users came from the spring workforce reductions rather than the switch to its own tools is not quantified.

Cointelegraph’s November 8 post did not say that StockTwits had summarised the report on October 5. Neither Microsoft, Meta nor Anthropic is quoted in the summary.

The Bottom Line

The Information reported, in a story StockTwits summarised on October 5, that Microsoft cut some internal AI usage caps from $100,000 to about $10,000 a month and steered developers to OpenAI models via GitHub Copilot, while Meta’s Claude Code users fell from about 60,000 to 30,000 as it pushed MetaCode and Muse Code, according to a StockTwits summary. With Apple and Goldman Sachs also capping AI tool use in Bloomberg’s reporting, the companies in these reports cut internal use while, the summary says, Microsoft still spends heavily on Anthropic for customer-facing Copilot.