Deep dive · Wednesday, 7 October 2026

Meta and Microsoft Slash Claude Use to Rein in Runaway AI Costs

Compiled by AI for Business and Life editorial - 07-10-2026 - AI for Finance

Meta and Microsoft are slashing internal use of Claude to stop runaway AI bills, offering CFOs a key lesson in controlling pay-per-use software costs.

What happened

Tech giants Meta and Microsoft are sharply cutting back internal employee use of Anthropic's Claude to stop runaway spending.

According to reports from The Information on October 5, 2026, both companies are steering technical workers toward their own internal tools instead.

The scale of these AI bills is massive. Before changing its policy, Microsoft projected its internal corporate spending on Anthropic would reach at least $1 billion for the year.

Microsoft leadership has now cut that projection by more than one-third.

As part of this clampdown, Microsoft cut monthly AI spending limits in its Cloud and AI division from $100,000 per employee to roughly $10,000 in most cases.

Microsoft executives Scott Guthrie and Jay Parikh told engineers to use internal alternatives, including GitHub Copilot and OpenAI models, instead of Claude.

Microsoft also told its Experiences and Devices division that it plans to end most internal Claude Code licenses by June 30, 2026. Staff must use GitHub Copilot CLI (command-line interface) instead.

Meta made similar deep cuts. The number of Meta employees using Claude Code fell from about 60,000 earlier this year to roughly 30,000.

Company layoffs caused part of that drop, but the main driver was a deliberate push by leadership toward internal software.

Meta spent more than $105 million on Claude Code during a single 28-day window. Leadership then stepped in with strict usage limits, spending controls, and internal monitoring.

Today, Meta's in-house assistant, MetaCode, has grown to over 30,000 active users. Another internal tool, Muse Code, supports more than 6,000 employees.

Microsoft still sells Claude to external business clients through Microsoft Foundry and Azure infrastructure. These cutbacks apply only to internal staff.

Both companies decided that paying huge external fees for work their own software could handle was financially unsustainable.

What it means for you

For any CFO (chief financial officer) or controller, the moves at Meta and Microsoft show how fast unmanaged AI usage can drain corporate budgets.

Over the last two years, many finance teams treated AI tools as standard software licenses with predictable monthly costs.

The reality is that AI tools run on consumption pricing. Costs build up every time an employee sends a prompt or runs a query.

A single technical worker testing code can generate massive usage in just one hour. Across hundreds of employees, this pay-per-use model creates explosive operational expenses.

Microsoft faced a $1 billion projection, while Meta saw a $105 million bill in 28 days. If tech giants cannot tolerate open-ended AI bills, your business cannot either.

This shift marks the start of disciplined FinOps (financial operations) for AI. Finance leaders must now demand measurable ROI (return on investment) for every dollar spent on external APIs (application programming interfaces).

There is also a clear build-versus-buy financial tipping point. When your external AI bills reach millions of dollars, running private or in-house models often costs less than third-party licenses.

By moving 30,000 workers to its own MetaCode tool, Meta cuts external costs while keeping valuable training data inside its own corporate walls.

These policy changes also highlight data security and contract risks. Both Meta and Microsoft raised concerns over third-party data retention rules.

When staff send proprietary code, financial formulas, or customer data to outside AI providers, standard 30-day data retention windows create legal and security risks.

Risks and limits to watch

Cutting your AI budget too quickly can hurt team productivity and create hidden costs.

First, employee output can drop. Microsoft engineers grew frustrated when their monthly experimentation budgets dropped from $100,000 to $10,000.

Claude Opus and Sonnet models have a strong reputation for handling complex technical tasks. Forcing staff to switch abruptly to other tools can delay project delivery schedules.

Second, building internal software is not free. Meta saves money on Anthropic invoices, but it spends heavily on servers, power, and dedicated maintenance teams.

Finance leaders must weigh the total cost of ownership rather than comparing an outside software bill to zero.

Third, changing team habits disrupts momentum. Moving staff to command-line tools requires training and support.

If an internal tool produces lower-quality work, the extra labor spent fixing errors can erase the software savings.

Finally, vendor relationships can create strategic tension. Microsoft is an Anthropic partner, a cloud supplier, and a direct rival.

Relying on software vendors who also sell competing enterprise tools can complicate your long-term contracts and pricing talks.

How to use this

You do not need to ban AI tools, but you must establish clear fiscal controls over how your team uses them.

Here is a practical roadmap to bring your company's AI spending under control:

Audit all current AI expenditures across departments. Work with procurement and accounting to review every software seat, cloud invoice, and API (application programming interface) charge. Look for shadow AI where staff expense individual accounts on corporate credit cards without central approval.
Set hard monthly spending caps. Eliminate bottomless usage allowances. Establish budget ceilings for engineering, marketing, and finance teams. Require formal managerial approval and a solid business case before anyone can exceed their baseline limit.
Evaluate the build-versus-buy math. Compare the annual cost of external software licenses against hosting your own open-source models on private cloud servers. If your outside AI spending reaches hundreds of thousands of dollars, calculate the break-even point for private hosting.
Review data retention clauses in vendor contracts. Check the legal terms of every AI service your business pays for. Make sure your agreements include zero-data-retention terms and prevent vendors from training their public models on your proprietary business data or financial records.
Tie AI tool budgets to measurable business returns. Require department heads to prove real productivity gains, such as shorter turnaround times or cost savings, during FP&A (financial planning and analysis) budget reviews. Treat AI like any other capital allocation by demanding clear proof of value.

Key takeaways

Microsoft reduced internal Claude spending projections by more than one-third from an initial $1 billion plan.
Microsoft dropped monthly per-employee AI spending caps in its Cloud and AI division from $100,000 to $10,000.
Meta halved internal Claude Code users from 60,000 to 30,000 after incurring over $105 million in fees in a single 28-day span.
Both technology giants are substituting external vendor models with in-house tools like GitHub Copilot and MetaCode.
Corporate finance teams must shift from uncontrolled AI testing to disciplined FinOps budgeting and data-retention reviews.

Frequently Asked Questions

Why are Meta and Microsoft cutting internal employee use of Claude?

Both companies are seeking to rein in runaway AI token costs that exceeded initial projections, with Microsoft facing a potential $1 billion internal Claude bill. They are also prioritizing their own internal AI coding assistants, such as GitHub Copilot and MetaCode, while addressing data-retention and intellectual property concerns.

Does this mean Microsoft is dropping Anthropic Claude for external enterprise customers?

No, the restrictions apply exclusively to internal employee usage. Microsoft continues to sell Claude models to commercial customers through Microsoft Foundry and Azure infrastructure, where customer spending on Anthropic models continues to expand.

How should CFOs and finance leaders manage escalating employee AI software costs?

Finance leaders should implement hard departmental token and budget ceilings rather than open-ended consumption accounts. They should also audit organization-wide AI subscriptions to eliminate shadow spending and compare vendor licensing fees against the economics of private internal deployments.

Sources

Keep learning

How this brief is made

Daily AI Updates are compiled with AI from public sources and checked against our house style. Every story links to its original source. Our editorial note