Apple’s artificial intelligence roadmap just took another twist, according to new reporting from Bloomberg’s Mark Gurman. The veteran journalist said Apple “runs on” Anthropic models for many internal tools, even after walking away from a public Claude integration. Consequently, analysts are asking why Apple depends on an outside supplier while marketing privacy and vertical integration. Meanwhile, the company has already inked a headline partnership with Google Gemini for the next generation of Apple Siri. This article examines the timeline, the money, and the strategic trade-offs shaping Apple’s current AI stack.

Inside Apple’s Quiet Partnership

Public attention fixated on the January 2026 Apple–Google announcement. However, Gurman’s podcast remarks offered a rare glimpse behind the curtain. He claimed Apple engineers rely on bespoke Anthropic Claude instances hosted on Apple servers. Therefore, daily prototype testing, code generation, and design reviews reportedly flow through those models. In contrast, the official Siri build now targets Google’s much larger Gemini foundation. Such dual sourcing signals a tactical, not ideological, approach to artificial intelligence.

These revelations underscore Apple’s pragmatic streak. Nevertheless, they also raise integration complexities. The stage is set for a deeper look at the failed Claude deal.

Anthropic Claude Deal Fallout

Negotiations between Apple and Anthropic began in mid-2025, according to CNBC summaries of Bloomberg reports. Furthermore, internal project codenames like “Glenwood” surfaced, hinting at an aggressive Siri overhaul. Subsequently, talks collapsed when Anthropic allegedly demanded “several billion dollars a year” with automatic price escalators. Gurman said the rising fee schedule proved untenable for Apple’s finance team. Moreover, the startup’s insistence on maintaining model control conflicted with Apple’s privacy architecture.

The impasse forced Apple leadership to explore alternatives quickly. Consequently, Google Gemini emerged as the stopgap. These events illustrate the high stakes of LLM licensing. However, cost alone does not tell the whole story, as the next section shows.

Why Gemini Won Out

Google reportedly offered a custom 1.2-trillion-parameter Gemini model. In contrast, Apple’s own cloud model hovers near 150 billion parameters. Therefore, the performance delta favored Gemini for consumer-facing voice assistant duties. Additionally, Google accepted Apple’s Private Cloud Compute safeguards, pledging no user data would leave Apple-controlled servers. Meanwhile, the rumored annual fee sits near US$1 billion—large, yet significantly below Anthropic’s ask.

Apple executives weighed capability, timeline, and brand risk. Ultimately, Gemini’s scale promised immediate feature parity with rival services like ChatGPT. Consequently, the partnership became public on 12 January 2026. These factors explain management’s decision, yet they do not erase Anthropic from Apple’s campus, as the following section explains.

Anthropic Claude Use Explained

Gurman’s quote, “Apple runs on Anthropic,” emphasizes day-to-day reliance rather than consumer deployment. Development teams reportedly query custom Claude models for code refactoring, UI copy suggestions, and security review automation. Moreover, Apple hosts those models internally to satisfy stringent privacy rules. Consequently, the company avoids sending proprietary source code to an external cloud.

Additionally, Anthropic specializes in fine-tuned enterprise workflows. Therefore, Apple gains tailored capabilities absent from generic public models. Nevertheless, the arrangement remains unofficial, and neither firm confirms specific terms. The next section quantifies the financial dynamics involved.

Cost Pressures And Strategy

Reported Pricing Figures Today

Publicly reported numbers indicate the challenge:

• Anthropic’s requested fee: “several billion dollars” annually, doubling year over year.
• Google Gemini license: roughly US$1 billion per year, Bloomberg says.
• Apple’s device footprint: more than 2.5 billion active units, creating vast inference loads.

Consequently, each extra cent per request translates into massive recurring cost. Furthermore, Apple still invests heavily in building its own trillion-parameter model. Balancing short-term capability with long-term independence drives the hybrid strategy. In contrast, a single-vendor path would create both pricing and antitrust exposure.

These cost realities steer Apple’s planning horizon. However, privacy considerations carry equal weight, as discussed next.

Privacy And Competitive Risks

Regulator And User Concerns

Apple markets privacy as a core brand pillar. Therefore, outsourcing cognition to a rival invites scrutiny. Nevertheless, Private Cloud Compute aims to retain user data within Apple’s enclave. Google echoed that claim in the joint statement. Moreover, regulators may still probe whether concentration in large model providers threatens competition.

Additionally, relying on external models can blur accountability for hallucinations or bias. In contrast, full in-house control would centralize responsibility. Consumers will judge Apple Siri performance and trust accordingly. Consequently, every misstep could amplify calls for transparency.

These privacy and competition factors intensify pressure on Apple’s internal teams. The following section explores their roadmap.

Future Of Apple Models

Engineering groups led by Craig Federighi and Mike Rockwell reportedly target a homegrown trillion-parameter model within two years. Meanwhile, incremental Siri upgrades will ship via Gemini. Additionally, internal Anthropic support may continue, depending on renegotiated pricing. Professionals seeking to contribute to such efforts can enhance skills through the AI Engineer™ certification.

Furthermore, Apple is expected to expand on-device inference using future M-series chips. Consequently, dependence on external compute could shrink over time. Nevertheless, sustained investment and talent retention remain vital. The company’s next Worldwide Developers Conference may reveal further milestones.

These plans suggest an eventual pivot back to self-reliance. However, market dynamics can shift quickly, keeping all options open.

Apple’s AI journey now features parallel tracks: a public alliance with Google and quiet internal reliance on Anthropic. Each path carries unique benefits, costs, and risks. Ultimately, execution will determine whether Apple Siri regains its competitive edge while upholding user trust. Interested readers should monitor forthcoming product releases and certifications to stay ahead of the curve.