# Microsoft and Meta cut internal Claude spending weeks before Anthropic's expected listing

Reports say Microsoft lowered its projected internal Claude spend by more than a third and Meta's Claude Code user count fell by about half, as both companies move engineers to their own and partner models.

- Published: 2026-10-07T06:04:58.881Z
- Canonical: https://polylog.news/ai/2026-10-07/microsoft-and-meta-cut-internal-claude-spending-weeks-before
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Polylog editors](https://polylog.news), [Anthropic News (Claude Sonnet 5.5)](https://www.anthropic.com/claude-sonnet-5-5), [Anthropic News (Claude Opus 5.5)](https://www.anthropic.com/claude-opus-5-5), [Meta AI (Muse Spark API)](https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/)

Two of Anthropic's largest enterprise customers are reducing how much their own engineers use Claude. [The Information reported on October 5](https://letsdatascience.com/news/reports-describe-meta-microsoft-cutting-claude-use-4777658a) that Microsoft lowered its internal spending projection for Claude by more than a third. Meta's count of Claude Code users fell to about 30,000 from roughly 60,000 earlier this year. Meta's in-house tools, MetaCode and Muse Code, now report more than 30,000 and more than 6,000 internal users respectively.

The Telegram channel [AI Post](https://t.me/aipost/8388) and [follow-up coverage](https://www.kucoin.com/news/flash/meta-and-microsoft-cut-employee-use-of-anthropic-s-claude-amid-cost-controls) add that Microsoft divisions now manage fixed token budgets. In most cases, those reports say, the monthly usage cap per individual fell from $100,000 to about $10,000, and Copilot routes requests to cheaper models by default. These details rest on secondhand reporting, and neither company has confirmed them. The cuts apply to internal use. They do not change customer access to Claude through Microsoft products, and neither company has dropped Anthropic.

The timing matters because Anthropic is close to a listing. The company [filed its S-1 confidentially on June 1](https://www.techstackipo.com/ipo/anthropic), and its last disclosed private valuation was $965 billion in the May Series H round. [Forecasters expect](https://futuresearch.ai/anthropic-openai-ipo-dates-valuations/) a listing in late October. The commonly cited $2 trillion figure reflects investor expectations, and Anthropic has not set it as a target. Some coverage says Microsoft and Meta together supply about a quarter of Anthropic's revenue. That figure also comes from secondhand reporting and has not appeared in any company filing.

The cost argument works in both directions. Anthropic says its newest models are cheaper to run. [Claude Opus 5.5](https://www.anthropic.com/claude-opus-5-5) performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5, according to the company. [Claude Sonnet 5.5](https://www.anthropic.com/claude-sonnet-5-5) runs 30% faster and costs up to 30% less for most work than Sonnet 5. Both claims come from vendor announcements and have not been independently reproduced. Meta, meanwhile, sells its own models: the [Muse Spark API](https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/) launched at [$1.25 per million input tokens and $4.25 per million output tokens](https://the-decoder.com/metas-muse-spark-1-1-api-pricing-squeezes-openai-and-anthropic-as-the-ai-price-war-heats-up/), well below Anthropic's flagship output pricing. Microsoft's position is mixed. It [committed up to $5 billion to Anthropic](https://blogs.nvidia.com/blog/microsoft-nvidia-anthropic-announce-partnership/) in November 2025, and Anthropic committed to buy $30 billion of Azure capacity. Microsoft is also directing its own developers toward GitHub Copilot and OpenAI models.

## What this means

The cause here is vertical integration more than any loss of capability. Platform companies that own a frontier or near-frontier model have a reason to send internal coding traffic to their own stack. They save money, and their internal usage becomes internal training and testing data. Anthropic is exposed because its enterprise growth story depends on heavy coding use by engineering organizations, and its largest buyers are also its competitors. Investors already expected a premium listing driven by coding revenue. What is new is evidence that the two largest internal deployments can be cut back by policy, quickly, once in-house tools are good enough. The usual view is that Claude's lead in coding protects that revenue. The Meta numbers suggest that a lead in coding quality does not stop a company that controls its own developer tools from switching its staff. A close precedent is Microsoft's switch from OpenAI-only Copilot to multi-model routing in 2025. Routing then became a price lever against suppliers. The likelier outcome is that Anthropic's IPO marketing in the coming weeks will put more weight on its broad customer base and on API revenue from companies without a model of their own, and less on its largest single accounts. Public-market pricing will apply a discount for concentration in those accounts. Two developments would prove this read wrong: S-1 disclosures showing customer concentration well below the reported quarter, or new enterprise commitments large enough to replace the lost spend before the roadshow.

## What to watch

- Whether Anthropic's public S-1 discloses revenue concentration by customer. That figure would confirm or refute the roughly 25% share now circulating in secondhand reports.
- Whether other model-owning companies, such as Google or Amazon, report similar caps on internal use of third-party coding agents. That would show the cuts are a pattern across the sector and not specific to Microsoft and Meta.
- Any change in Anthropic's list prices for Opus or Sonnet before the listing. A price cut would show that Anthropic is defending volume at the cost of margin.