Meta Platforms and Microsoft are reportedly scaling back how much their employees rely on Anthropic’s Claude for internal work, according to a report from The Information. The shift, described in reporting published this week, marks one of the first visible signs that two of Anthropic’s biggest corporate customers are steering engineers toward in-house tools and rival models instead of leaning further into Claude.
The numbers are specific enough to matter. At Meta, internal use of Claude Code reportedly fell from roughly 60,000 users earlier in 2026 to about 30,000 now, a drop of half. Microsoft, meanwhile, had projected spending at least $1 billion on Anthropic technology earlier this year. That figure was later cut by more than a third, per the same reporting. Neither company has issued a detailed public statement breaking down the new totals, and Anthropic has not commented publicly on the figures either.
For a company that has built a chunk of its enterprise reputation on Claude Code’s reputation among developers, even a partial retreat by two hyperscaler-sized customers is a notable data point. It also lands at a sensitive moment: Anthropic has been moving toward a public listing, and its IPO filing devoted 80 pages to AI risk disclosures, underscoring how closely investors are watching Anthropic’s customer concentration and growth trajectory heading into any listing.
What The Information’s report actually says
The core claim, attributed to The Information, is that both Meta and Microsoft have been actively reducing employees’ internal use of Anthropic’s Claude tools. That is distinct from either company dropping Claude entirely or canceling an enterprise contract outright. The reporting describes a pullback in day-to-day usage and budgeted spend, not a severed relationship.
At Microsoft, the reported shift involves steering staff toward GitHub Copilot, Microsoft’s own AI tools, and OpenAI’s models instead of Claude. That detail matters because Microsoft is both an investor in OpenAI and a company that has spent years building out Copilot as its flagship AI coding product. Pushing engineers back toward tools Microsoft already owns or has a stake in is a logical move if internal usage patterns are being re-evaluated for cost or strategic reasons.
At Meta, the picture is more specific to coding tools. Meta’s own internal coding assistant, MetaCode, is reported to have more than 30,000 internal users, roughly matching the number Claude Code is said to have lost. Meta has also been building out Muse Code, which began external testing in August 2026 and reportedly already has more than 6,000 internal users. Put together, those figures suggest Meta’s drop in Claude Code usage lines up closely with growth in Meta’s own coding tools, rather than a move toward some unrelated product.
The numbers at a glance
Here is what has been reported so far, organized by company and tool. Figures marked as reported come directly from The Information’s reporting as relayed in subsequent coverage. Anything not independently confirmed is flagged as such rather than stated as fact.
| Company | Tool | Reported figure | Status |
|---|---|---|---|
| Meta Platforms | Claude Code (internal use) | ~60,000 users earlier in 2026 to ~30,000 now | Reported |
| Meta Platforms | MetaCode (internal tool) | 30,000+ internal users | Reported |
| Meta Platforms | Muse Code (external testing since August 2026) | 6,000+ internal users | Reported |
| Microsoft | Anthropic spend projection (early 2026) | At least $1 billion planned | Reported |
| Microsoft | Anthropic spend, revised | Cut by more than one-third | Reported |
| Microsoft | Steered tools instead | GitHub Copilot, Microsoft’s own AI tools, OpenAI models | Reported |
What is not in that table is deliberate. The exact post-cut dollar total for Microsoft’s Anthropic spending has not been confirmed in the reporting available, nor has a precise percentage beyond “more than one-third.” Claims circulating elsewhere that Microsoft plans to end most Claude Code licenses by a specific mid-2026 date are not confirmed by the sourcing here, and neither is any assertion that Meta’s usage drop was primarily caused by its recent round of layoffs. Readers should treat those as open questions, not settled facts.
Why this is happening now
There are a few plausible, non-exclusive explanations, and the reporting does not pin the shift on a single cause. The most obvious one is cost. Enterprise AI spending has ballooned across the industry in 2026, and companies the size of Meta and Microsoft run constant internal reviews of which tools earn their budget line. If an internal coding assistant built in-house can match or approach the usefulness of a third-party model at a fraction of the per-seat cost, finance teams tend to notice.
A second factor is plain competitive instinct. Both Meta and Microsoft have spent much of 2026 racing to stand up credible internal AI coding tools of their own, partly to avoid long-term dependence on a single external vendor for a function as central as software development. Meta’s MetaCode and Muse Code push, and Microsoft’s continued investment in GitHub Copilot alongside its OpenAI relationship, both fit a pattern of hyperscalers preferring to own the tools their engineers depend on most, when they can build something workable.
A third factor, specific to Microsoft, is the OpenAI relationship. Microsoft has invested billions in OpenAI and has strong incentive to route usage toward OpenAI’s models where the two products are close in capability. Reducing Anthropic spend while increasing reliance on OpenAI and Copilot keeps more of that usage, and more of the associated infrastructure revenue, inside Microsoft’s own ecosystem.
Context: how we got here
It’s worth remembering how recently Meta and Microsoft were both leaning into Anthropic rather than away from it. Earlier in 2026, reporting indicated Meta was weighing spending as much as $10 billion a year on Anthropic as Anthropic’s IPO plans advanced, a figure covered at the time in a report on Meta’s potential $10 billion annual Anthropic commitment. That is a sharp contrast with a report of Meta’s internal developer headcount on Claude Code falling by half just months later.
Anthropic’s own coding product has had a turbulent few months on the news cycle regardless of this specific report. The company published an explanation after users flagged a perceived quality drop in Claude Code, which Anthropic pegged at roughly a 3% hit to certain outputs, tied to infrastructure changes rather than a deliberate downgrade. Separately, Anthropic’s flagship model line has kept shipping aggressively: Claude Opus 5.5 launched with a 1-million-token context window aimed squarely at coding tools, and Anthropic has claimed Opus 5.5 beats OpenAI’s GPT-5.6 Sol on several benchmarks at roughly a third of the cost. Those product wins make the Meta and Microsoft usage pullback a more interesting puzzle: Anthropic’s technology has not obviously regressed, even as two of its largest corporate users dial back how much they lean on it internally.
OpenAI has not been standing still on pricing either. GPT-6’s Sol and Luna variants launched at roughly half off list pricing earlier this year, a move widely read as an attempt to undercut Claude on price for comparable coding workloads. If Microsoft is indeed steering staff toward OpenAI models as the reporting suggests, cheaper OpenAI pricing makes that shift easier to justify internally, regardless of how the models compare head-to-head on raw capability.
Competitive comparison: the internal coding-tool landscape
What this report really highlights is how crowded the market for internal developer AI tools has become at the companies large enough to build their own. Three years ago, most big tech firms simply bought seats of whatever third-party coding assistant was best at the time. In 2026, the biggest players increasingly run a mixed fleet: their own tool for routine work, a licensed third-party model for harder tasks, and a rotating evaluation of whichever competitor is cheapest or fastest that quarter.
| Tool | Owner | Reported scale | Primary use case |
|---|---|---|---|
| Claude Code | Anthropic | ~30,000 Meta users (down from ~60,000) | General coding assistant, licensed by enterprises |
| MetaCode | Meta Platforms | 30,000+ internal users | In-house coding tool for Meta engineers |
| Muse Code | Meta Platforms | 6,000+ internal users; external testing since Aug. 2026 | Newer in-house tool, now being tested outside Meta |
| GitHub Copilot | Microsoft | Hundreds of millions of users industry-wide | IDE-integrated coding assistant, Microsoft-owned |
| OpenAI models (internal use) | OpenAI, used via Microsoft | Growing share of Microsoft internal tooling | General-purpose and coding tasks inside Microsoft |
Anthropic is not alone in facing this dynamic. Google, OpenAI, and Meta have all pushed internal tooling for their own engineers for years, partly as a hedge against vendor lock-in and partly because internal dogfooding is cheap marketing when the tool is good enough to show off. What’s new here is seeing two of the most AI-forward companies in the world apparently pull back on Claude specifically, at a scale large enough to be independently reported rather than dismissed as routine internal churn.
What this means for Anthropic’s enterprise business
Anthropic’s enterprise pitch has leaned heavily on Claude Code’s standing among developers as a proof point, alongside broader enterprise API revenue that extends well beyond Meta and Microsoft. A reported drop in internal seat counts at two large customers does not necessarily translate one-to-one into lost revenue, since enterprise contracts are often structured around committed spend, API call volume, or seat tiers that don’t move in lockstep with headcount using the tool day to day.
Still, the optics matter heading into a period when Anthropic is trying to demonstrate durable enterprise demand to public-market investors. Anthropic has publicly said Claude itself handles a meaningful share of its own internal AI research and development work, citing a 26% figure, which the company uses as evidence Claude is good enough to trust with its own roadmap. A simultaneous pullback at Meta and Microsoft complicates that narrative somewhat, even if the underlying cause turns out to be cost allocation rather than a judgment on model quality.
It’s also worth noting that Microsoft’s position is unusual: it is simultaneously a major Anthropic customer, a major OpenAI investor, and a company racing to build its own tools. Any reporting about Microsoft trimming Anthropic spend has to be read with that three-way conflict of interest in mind. The same applies to Meta, which has both invested in building internal tools and, per earlier 2026 reporting, considered committing billions annually to Anthropic. Large companies often run both strategies at once, hedging rather than picking a single lane.
Market impact and investor reaction
Anthropic remains a private company, so there is no stock ticker to move on this news the way a Meta or Microsoft earnings miss would. But the report lands during a sensitive IPO-adjacent window, and institutional investors evaluating Anthropic’s growth story will read customer concentration and retention signals closely. A reported halving of usage at one of the three or four largest technology companies on earth, even confined to one specific internal tool, is the kind of detail that shows up in due diligence conversations regardless of how it’s eventually explained.
For Microsoft and Meta, the market impact is likely to be close to zero in isolation. Both companies run AI budgets in the tens of billions of dollars, and a shift of roughly a third of one vendor relationship’s spend is a rounding error against their total AI capital expenditure. The more interesting signal for investors in Meta and Microsoft is less about dollars saved and more about what it says about each company’s confidence in its own internal AI tooling. If MetaCode and Muse Code are good enough to absorb real internal workload at scale, that’s a data point in Meta’s favor as it tries to prove its AI investments are paying off beyond advertising.
How Anthropic has responded so far
As of this writing, Anthropic has not issued a public statement directly addressing the figures reported by The Information. That is not unusual. Companies frequently decline to confirm or deny specific internal usage statistics attributed to named customers, particularly when those customers are also major business partners. Anthropic has, however, continued to ship product updates at a steady pace, including the Claude 5.5 family of models and continued iteration on Claude Code itself, suggesting the company’s public posture is to keep shipping rather than directly rebut the report.
Neither Meta nor Microsoft has published a detailed breakdown confirming or disputing the specific user counts or spending figures either. Until one of the three companies speaks on record with specifics, the numbers in this story should be treated as reported rather than officially confirmed, even though they come from a publication with a strong track record on enterprise AI spending stories.
What to watch next
A handful of concrete developments would confirm or complicate this story in the coming weeks. Anthropic’s next enterprise revenue disclosure, whether in IPO-related filings or an investor update, would show whether API and seat revenue from large accounts held steady despite the reported internal usage drop. Any public comment from Microsoft about its AI tooling budget allocation across Copilot, OpenAI, and Anthropic would also clarify how much of this is cost discipline versus a strategic pivot away from Anthropic specifically.
Meta’s Muse Code rollout is another thing worth tracking. It only began external testing in August 2026, and if Meta pushes it toward a broader public or enterprise release, that would support the theory that Meta’s internal Claude Code drop reflects substitution by Meta’s own tools rather than a broader retreat from third-party AI coding assistants altogether.
Predictions: where this goes from here
Based on the pattern described in this report and the broader direction of the AI coding tools market in 2026, a few outcomes look more likely than others over the next two to three quarters.
- Anthropic will likely emphasize API revenue and smaller enterprise accounts in upcoming disclosures to offset questions about large-customer concentration, rather than directly address the Meta and Microsoft figures.
- Meta will probably continue expanding Muse Code’s external testing footprint, using internal substitution of Claude Code as evidence the tool is production-ready beyond Meta’s own walls.
- Microsoft is likely to keep steering default tooling toward GitHub Copilot and OpenAI models for new internal projects, while keeping some Anthropic access available for specific use cases rather than eliminating it outright.
- Expect other large AI buyers, not just Meta and Microsoft, to quietly run similar internal cost reviews of third-party model spend in late 2026, given how much total enterprise AI budgets have grown this year.
- Anthropic’s IPO narrative will likely shift further toward API usage growth, developer platform metrics, and model benchmark wins like the Opus 5.5 comparisons, rather than headcount-style usage stats at any single customer.
The bigger picture for enterprise AI spending
Step back from the Meta and Microsoft specifics and this story fits a broader 2026 pattern: the era of large tech companies defaulting to whichever third-party AI model tested best is giving way to a more deliberate build-versus-buy calculation, the same calculation that has shaped enterprise software for decades. Companies with the engineering resources to build credible internal tools are doing so, and using that leverage to negotiate down, or simply reduce, spend with external vendors.
That doesn’t mean third-party model providers lose out across the board. Smaller companies without the resources to build MetaCode-style internal tools will keep buying from Anthropic, OpenAI, and others. But for the handful of companies large enough to build their own, 2026 looks like the year that calculation started tipping back toward in-house tools for at least some workloads, with Anthropic’s Claude Code apparently absorbing the first visible hit.
Frequently asked questions
Did Meta and Microsoft drop Anthropic’s Claude entirely?
No. The reporting describes a reduction in internal usage and planned spending, not a full cancellation of either company’s relationship with Anthropic.
How much did Meta’s internal Claude Code usage drop?
The Information reported Meta’s internal Claude Code users fell from roughly 60,000 earlier in 2026 to about 30,000, a decline of about half.
How much did Microsoft cut its Anthropic spending?
Microsoft had projected spending at least $1 billion on Anthropic technology earlier in 2026. That figure was later reported to have been reduced by more than one-third. An exact revised dollar total has not been confirmed publicly.
What tools is Microsoft using instead of Claude?
Microsoft is reportedly steering employees toward GitHub Copilot, its own internal AI tools, and OpenAI’s models.
What is Meta using instead of Claude Code?
Meta’s internal coding tool, MetaCode, reportedly has more than 30,000 internal users. Meta is also testing a newer tool, Muse Code, which began external testing in August 2026 and has more than 6,000 internal users.
Has Anthropic responded to the report?
As of this writing, Anthropic has not issued a public statement directly addressing the specific user and spending figures reported by The Information.
Does this affect Anthropic’s IPO plans?
It is unclear. Anthropic’s IPO filing has already devoted significant space to AI-related risk disclosures, and large-customer usage trends like this one are the type of detail investors typically scrutinize during due diligence, though the report alone does not confirm any direct impact on IPO timing or valuation.
Is Claude Code still competitive on capability?
Anthropic has continued to promote strong benchmark results for its latest models, including claims that Claude Opus 5.5 beats OpenAI’s GPT-5.6 Sol on certain benchmarks at a lower cost. The reported usage pullback at Meta and Microsoft appears tied to internal cost and tooling strategy rather than any publicly confirmed capability gap.
