Apple Secretly Runs on Claude but Chose Gemini for Siri: $500M Is Why

Apple runs on Anthropic. That's how bluntly Mark Gurman, the Bloomberg journalist who has broken more Apple leaks than anyone in the past decade, dropped the news. He said it on a podcast, almost casually, as if it were a minor detail. It's anything but.

Let me break this down: the world's most valuable company uses the product of a startup it rejected as the provider for its flagship assistant. Apple has custom versions of Claude running on its own servers for product development and internal tools, but when it came time to choose who would power the new Siri, it said no to Anthropic and signed with Google. The reason comes down to a very specific number: $500 million per year in savings.

This is the complete story of how Apple is playing both sides of the AI game, why Anthropic lost the biggest consumer AI contract in history, and what it all means for you.

Gurman's Revelation: Apple Runs on Claude Under the Hood

On January 30, 2026, Mark Gurman appeared on the TBPN podcast and revealed something nobody expected: Apple currently "runs on Anthropic." Not as an investor. Not as a minor partner. As an actual day-to-day dependency.

Think of it like this: if Apple were a restaurant, Siri would be the dish served to customers, cooked with Google Gemini ingredients. But in the kitchen, the chefs use Anthropic knives to prepare everything. That's literally the situation.

What Claude Does Inside Apple

According to Gurman, Anthropic is "powering a lot of the stuff Apple's doing internally in terms of product development and a lot of their internal tools." Specifically:

• Xcode with AI: Since May 2025, Apple has been working with Anthropic on an AI-powered version of Xcode based on Claude Sonnet. It includes a programmer chat, code generation, interface testing, and bug detection. Claude Sonnet 4 is now publicly available in Xcode 26.
• Internal development tools: Custom versions of Claude running on Apple's own servers. They don't use the public API; they have dedicated implementations.
• Product development: Gurman confirmed that Claude powers multiple internal tools without specifying which ones.

The trick is understanding why Apple needed Anthropic in the first place. Apple had announced Swift Assist, its own AI coding assistant for Xcode, but it never shipped. Engineers internally discovered it "made up information" and "could slow down app development." Instead of pushing through with their own solution, Apple did something very un-Apple: admitted someone else did it better and signed with Anthropic.

The Negotiation That Went Wrong: Why Anthropic Lost Siri

Here's where the story gets really interesting. Because Apple didn't just want Claude for Xcode. Originally, Apple "was going to rebuild Siri around Claude." Think about that for a second: every iPhone, iPad, and Mac on the planet with Claude as Siri's brain.

But Anthropic, according to Gurman, "was holding them over a barrel." And here come the numbers:

What most guides won't tell you is that Anthropic didn't just ask for more money than Google. They demanded that the price double every year for the next three years. If year one cost $1.5 billion, year two would be $3 billion and year three $6 billion. We're talking over $10 billion total.

Google, meanwhile, offered ~$1 billion per year. But there's a detail that makes this figure even more attractive: Google already pays Apple over $20 billion annually to be the default search engine in Safari. The Siri deal can be structured as a partial offset of that existing flow. In practice, the net cost for Apple of using Gemini in Siri could be close to zero.

What Google Brought vs What Anthropic Demanded

It wasn't just about money. Google brought a 1.2 trillion parameter model to the table compared to Apple's internal model of 150 billion parameters (an 8x improvement). Plus, Google offered to run Gemini on Apple Private Cloud Compute servers, with data isolated from Google and no visible branding. Siri would still be Siri, just with Gemini under the hood.

Tim Cook confirmed in Apple's Q1 2026 earnings call: "We determined that Google's AI technology would provide the most capable foundation for AFM" (Apple Foundation Models). Translation: Google was cheaper and more powerful. Anthropic was out.

The Grand Irony: Everyone Is a Competitor and Partner Simultaneously

This is where the story becomes a web of ironies that deserves a relationship chart. Let's break it down.

Anthropic, the company whose products Apple uses internally, is primarily funded by:

• Amazon: $8 billion invested, less than 33% ownership
• Google: $3 billion+ invested, ~14% ownership
• Microsoft: Up to $5 billion, Claude available on Azure
• Nvidia: Up to $10 billion in hardware partnership

So Apple internally depends on the product of a company that is nearly 50% owned by Amazon and Google, its two biggest competitors in cloud and services. And for its consumer product (Siri), it pays directly to Google, which is an investor in the provider it rejected.

Think of it like Ford rejecting engines from a startup that General Motors owns 14% of, only to buy engines directly from General Motors. Meanwhile, in its factory, Ford uses tools from that same startup it rejected. This level of corporate entanglement only exists in Silicon Valley.

Apple's AI Relationship Map in 2026

Anthropic: Greed or Calculated Strategy?

Several outlets, including AppleInsider, headlined that "Anthropic got greedy" and that's why they lost the Siri contract. But the reality is more nuanced.

Anthropic has an annualized revenue run rate exceeding $9 billion, with projections of $20-26 billion for 2026. Its valuation has gone from $61.5 billion in March 2025 to $350 billion in January 2026. It just closed a $20 billion round with GIC and Coatue.

From Anthropic's perspective, asking for premium prices makes sense. Their training and infrastructure costs are astronomical, and signing a contract below cost would be financially suicidal, even if the customer is Apple. Besides, Claude Code alone already generates over $500 million in annualized revenue, proving there's a market willing to pay their prices.

But the outcome speaks for itself: Anthropic lost the biggest consumer AI contract in history. The 2+ billion active Apple devices worldwide will run Gemini, not Claude. That's a distribution opportunity that doesn't come around twice.

What This Means for Siri and for You

If you've already read our analysis of the new Siri with Gemini, you know the launch is planned for iOS 26.4 in the second half of February. But this revelation adds important context:

1. Siri Wasn't Plan A with Gemini

Apple wanted Claude. The fact that they ended up with Gemini isn't because Google was their first choice, but because Anthropic priced themselves out. This means Apple considers Claude technically competent (they use it internally), but chose Gemini for financial reasons.

2. Apple Is Building Its Own Model

Apple's long-term strategy is clear: use Gemini as a bridge while developing its own 1 trillion parameter model (project "Ferret-3"). Tim Cook doesn't want to depend on anyone long-term. Gemini is temporary; AI independence is the endgame.

3. Privacy Remains the Central Argument

Gemini will run on Apple Private Cloud Compute, isolated from Google's servers. Data won't be shared with Google. Apple maintains its privacy narrative even when outsourcing the intelligence. It's an elegant compromise, though it remains to be seen whether regulators and users accept it without question.

4. OpenAI Is Still in the Picture

The Google deal isn't exclusive. Apple maintains its ChatGPT integration as an alternative option in Siri. This means in 2026, your iPhone could use three different AI models depending on the task: Apple's own model for simple tasks, Gemini for the new conversational Siri, and ChatGPT as an explicit user choice.

The Big Tech AI War Context

This story doesn't happen in a vacuum. The week of January 27-31, 2026 has been one of the most intense in the Big Tech AI war:

• Microsoft reported solid results but shares fell because Azure slightly decelerated. Massive AI spending is raising Wall Street doubts.
• Anthropic closed a $20 billion round, doubling its initial target. Its valuation reached $350 billion.
• Google is showcasing Gemini 3 as the most capable model on the market, with the Apple contract as validation.
• Apple revealed its AI strategy is a mosaic of dependencies: Anthropic inside, Google outside, OpenAI as an option, and a proprietary model in development.

What most guides won't tell you is that this fragmentation actually benefits users. When Apple depends on multiple providers, none has monopoly power over your iPhone's AI experience. If Gemini fails, there's Claude inside and ChatGPT as backup. It's a diversification strategy that, intentionally or not, protects consumers.

What This Reveals About Siri's Historic Failure

Gurman was particularly harsh on one point: Apple "completely screwed up" its AI strategy. The hiring of John Giannandrea from Google in 2018 was called "Tim Cook's biggest mistake." Giannandrea prioritized privacy over capability, and the result was a Siri that for years was inferior to Alexa, Google Assistant, and ChatGPT.

The fact that Apple now needs to pay $1 billion per year to Google for a competent Siri is the ultimate proof of that failure. No other company of its size and resources has had to outsource the intelligence of its flagship product like this.

But it also shows pragmatism. Tim Cook could have insisted with the proprietary model for years longer, as he did with Maps (which took a decade to become competent). Instead, he chose to pay for the best available technology while developing his own. It's corporate humility, something rare at Apple, and probably the right call.

Frequently Asked Questions

Did Apple invest money directly in Anthropic?

No. Unlike Amazon ($8B), Google ($3B), Microsoft ($5B), and Nvidia ($10B), Apple doesn't appear as an equity investor in Anthropic. The relationship is strictly a technology partnership: Apple pays to use Claude but doesn't own shares or stake in the company. It's a decision that reflects Apple's philosophy of maintaining independence from its suppliers.

Will Siri ever use Claude?

There are no public plans for that. The Google Gemini contract covers Siri's near future (iOS 26.4 and iOS 27). Apple keeps Claude exclusively for internal use (Xcode, development tools). Long-term, Apple is working on its own model (Ferret-3) to eventually replace Gemini. Claude in Siri seems ruled out unless commercial conditions change drastically.

Does this affect developers using Xcode with Claude?

Not negatively. In fact, it reinforces it. The fact that Apple depends internally on Claude for its own tools means it has a direct incentive to maintain and improve Claude's integration in Xcode. If Apple only used Claude as an optional extra, it could easily drop it. By depending on it internally, continuity is more guaranteed.

How much does Apple pay Anthropic for internal Claude use?

Gurman didn't reveal specific figures for the internal deal. What we know is that Anthropic asked ~$1.5 billion/year to power Siri. The deal for internal use and Xcode is presumably a fraction of that figure, since it doesn't include mass consumer distribution. Analysts estimate it could be in the $100-300 million annual range.

Does Google know Apple uses Claude internally?

There's no indication of exclusivity in the Apple-Google agreement. Google invested $3 billion in Anthropic and owns 14% of the company, so it has visibility into Anthropic's business growth. Plus, Gurman himself made the revelation publicly, so at this point everyone knows. The Apple-Anthropic relationship doesn't violate any known terms of the Apple-Google Siri deal.

Conclusion: AI Is a Market of Unlikely Alliances

Gurman's revelation exposes an uncomfortable truth for those who see the AI war as a clear competition between sides: there are no sides. Apple uses Claude inside and Gemini outside. Google invests in Anthropic while competing against them. Amazon is Anthropic's largest investor while developing its own models. Microsoft invests in both Anthropic and OpenAI simultaneously.

AI in 2026 is not a war between companies. It's an ecosystem where everyone depends on everyone, and the question isn't who wins, but who manages their dependencies best. Apple, with its strategy of internal Claude + external Gemini + proprietary model in development, seems to have found a pragmatic balance. It's not elegant, but it works.

And you, as a user, come out ahead. Because when Apple doesn't put all its AI eggs in one basket, your iPhone gets more options, more resilience, and more competition pushing prices down. That, in the end, is what matters.