🔍 Read the full analysis: The Price Of Switching AI Models After Meta And Microsoft Pulled Back From Claude on ThorstenMeyerAI.com
Get monitors, keyboards and dev gear delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
TL;DR
The Information reported on Oct. 5 that Meta reduced employee use of Claude Code and Microsoft lowered its projected internal spending on Anthropic technology. The reported moves reflect cost controls and alternatives the companies already operate, not a stated finding that Claude performs worse. For other businesses, switching models can bring engineering, evaluation, and productivity costs that large technology companies are better equipped to absorb.
Meta and Microsoft have reportedly pulled some of their employees’ AI work away from Anthropic’s Claude tools, according to a report by The Information on Oct. 5. The reported shifts point to cost management and the availability of in-house or alternative products; neither company is reported to have said Claude delivered poorer results. The development matters to businesses weighing AI vendors because changing models can require substantial work beyond replacing one software subscription with another.
Meta reportedly reduced the number of employees using Claude Code from about 60,000 to about 30,000 after earlier uptake this year. The report says the company has been steering staff toward its own coding products: MetaCode, which has more than 30,000 internal users, and Muse Code, with more than 6,000. These figures describe reported internal use, not sales to outside customers.
Microsoft had reportedly projected more than $1 billion a year in internal spending on Anthropic technology, including Claude Code, Claude models used in Copilot, and Claude Mythos. The company has since cut that projection by more than a third and is directing employees toward GitHub Copilot and OpenAI models, according to the report. Another detail cited in the source account is that some monthly team budgets may have fallen from roughly $100,000 to roughly $10,000; that figure is attributed to a single report and is not established as a company-wide change.
The reported decisions concern employees’ internal tools and budgets. They do not establish that either company has ended access to Claude or stopped using Anthropic products for customers. Microsoft is reported to continue spending on Anthropic models for customer-facing Copilot features, while customer use of Claude through Microsoft platforms is reportedly growing. The companies’ reasons, as described in the reporting, include token costs, tighter spending controls, and a preference for tools they own or support.
Meta and Microsoft pulled back from Claude. Here’s what switching actually costs.
The Information reports both companies steering their own employees away from Claude. Read as a verdict on Claude, it misleads. Read as a demonstration of switching — and who can afford it — it’s the most useful enterprise-AI signal this month.
Staff steered to GitHub Copilot and OpenAI models; stricter token budgets. One unconfirmed report: some team budgets ~$100k → ~$10k/month.
Microsoft reportedly still spends heavily on Claude for customer-facing Copilot — and that spending is reported to be growing.
Reported drivers: rising token costs and owned alternatives. Neither company is reported to have called Claude worse.
Meta builds coding tools; Microsoft owns Copilot and backs OpenAI. This is ordinary vertical integration.
Keep a second vendor live on real work.
A few hundred tasks with pass criteria.
Logic, prompts, tools in your layer.
Tokens are the cheap half.
Know what you’d rebuild.
On the evidence reported, Meta and Microsoft didn’t reject Claude. They brought spending in-house where they could and kept buying where they couldn’t — Microsoft remains a large Anthropic customer for the products it sells. The signal is the mechanism: the most sophisticated buyers treat models as interchangeable suppliers behind a layer they control.Meta could halve its Claude usage because it had built somewhere else to go. Build somewhere else to go.
The Cost of Changing Models
The report highlights a distinction between the price of model access and the cost of switching. Moving a workflow to another model can require teams to rerun evaluations, adjust prompts and tool definitions, and test whether the replacement performs adequately on their own tasks. If a company lacks a representative evaluation set, it may have no dependable way to compare quality before making the change.
There can also be less visible costs. A new coding assistant may lack the integrations and team conventions built around the old one, while engineers need time to adapt. If a replacement produces more errors or weaker results on a company’s work, additional review and rework can offset lower usage charges. For agent-based workflows, changes in cached context and related pricing may also affect total costs. These are general switching risks described in the source material, not reported measurements of the impact at Meta or Microsoft.
Large firms with engineering resources and existing substitutes may be able to absorb this work and make a change worthwhile. Smaller buyers may face a different calculation: savings on tokens do not automatically mean savings overall. The practical lesson is not that businesses should leave Claude, but that they should understand the costs and quality trade-offs of moving any important workload.
As an affiliate, we earn on qualifying purchases.
Why These Companies Had Alternatives
Meta and Microsoft are not typical enterprise customers. Meta develops its own models and coding products. Microsoft operates GitHub Copilot and has a major partnership with OpenAI. That means both companies have alternatives available internally and commercial interests connected to products that can compete with a third-party supplier. Their reported choices may reflect those circumstances, rather than a general verdict on Anthropic’s products.
The source account distinguishes internal employee use from customer-facing services. It says Microsoft continues to use Anthropic technology in some Copilot features and that customer spending on Claude through Microsoft platforms is growing. Those reported details complicate a simple reading that Microsoft has abandoned the supplier. The development is better understood as a change in where some work is routed and how internal spending is allocated.
The account also cites a recent SemiAnalysis finding that AI subscription limits can change without prominent notice and may vary by account. That observation is separate from the Meta and Microsoft report, but it reinforces why buyers may want options when pricing, usage limits, or product terms change. It does not establish that Anthropic changed the terms of service used by either company in this case.
enterprise AI model comparison tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What the Report Does Not Establish
The available account does not provide direct statements from Meta, Microsoft, or Anthropic confirming the reported user counts, spending projections, or reasons for the changes. It is also unclear how the companies define an active user, which teams or tasks are included, and over what period the figures were measured. The reported Microsoft spending figure is a projection, not a confirmed final bill.
There is no reported evidence here that either company concluded Claude was inferior for its work, nor enough information to compare output quality across Claude, Meta’s tools, GitHub Copilot, or OpenAI models. The extent to which internal shifts affect Anthropic’s revenue is also unclear, particularly given the report that Microsoft continues to use Claude for customer-facing features. The specific $100,000-to-$10,000 team-budget example comes from a single account and should not be treated as a broad policy without further confirmation.
AI development cost management software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How Buyers Can Prepare to Switch
The next useful evidence would be clarification from the companies on the scale, timing, and scope of the reported changes, alongside any further information about Microsoft’s ongoing use of Anthropic models in Copilot. Until then, the figures should be treated as reported internal measures rather than a full account of either company’s AI spending.
For other organizations, the immediate question is how to make a future change measurable and manageable. The source account recommends keeping more than one model family in production, even if the second handles only a limited share of work; maintaining an evaluation set of representative tasks; and keeping business logic, prompts, and tool definitions in a layer the company controls. Buyers can also track spending per task and review or rework required per accepted result, rather than comparing token prices alone. These steps do not eliminate switching costs, but they can give teams better evidence about whether a move is worth making.
AI model evaluation tools for businesses
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
Did Meta and Microsoft stop using Claude?
The report does not say that. It describes reduced or redirected internal employee use. Microsoft is also reported to continue using Anthropic models in some customer-facing Copilot features.
Why are the companies reportedly moving work?
The reported reasons include higher token costs, tighter spending controls, and available alternatives that the companies own or support. The account does not report either company saying Claude performed worse.
Does this show that Claude is worse than other AI models?
No. The reported changes alone do not establish comparative performance. The available account attributes the moves to spending and product strategy, and it provides no head-to-head results for the companies’ tasks.
Why can switching AI models cost more than expected?
Teams may need to repeat evaluations, revise prompts and integrations, retrain users, and account for additional review or rework. A lower model price may not produce lower total costs if those expenses or quality differences are significant.
What can a business do before changing AI providers?
Keep a second model tested on real work, maintain representative evaluation tasks, and separate core application logic from provider-specific settings where possible. Track quality and review effort alongside usage costs so a decision reflects the full workload.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
