Anthropic has begun early work on its own AI chip and held talks with Samsung as a possible manufacturer. The project is still in its infancy, with no design, target workload, or performance goal locked in. The move fits a wider push by AI labs to loosen their grip on Nvidia hardware.

Key Takeaways

• Anthropic is in early talks with Samsung to build a custom AI chip
• No design, target workload, or performance specs are decided yet
• Samsung’s 2-nanometer process and in-house memory make it a fit
• The move follows OpenAI’s Broadcom-built custom inference chip
• Anthropic says Amazon, Google, and Nvidia chips stay central

What Anthropic and Samsung Are Discussing

The report came from The Information on July 2, 2026, which said Anthropic has held talks with Samsung Electronics as a manufacturing partner for a custom AI chip.

The plans are very early. According to Bloomberg’s write-up of the report, Anthropic is still figuring out what the processor should do, how powerful it should be, and how it would slot into a server. The company could still walk away from the effort entirely.

Anthropic played down the news. When asked for comment, it told TechCrunch that a diversified hardware stack with chips from Google, Amazon, and Nvidia stays central to its compute strategy, and added it had nothing further to share on the Samsung talks. Samsung declined to comment.

This isn’t a bolt from the blue. Reuters reported back in April that Anthropic was weighing whether to build its own chips as Claude’s compute demands outran available supply. At the time, there was no dedicated team and no chosen design.

What changed is action. Talking to an actual manufacturer is a concrete step, not a brainstorm. It signals Anthropic is testing the water with a real partner rather than just running the numbers internally.

Why Samsung Is the Talking Partner

Samsung checks a lot of boxes for a company that wants a serious chip-making partner. It already sits deep inside the AI supply chain as a major manufacturer for Nvidia, producing the silicon that powers AI training and inference.

The technical fit is the draw. Reporting notes Anthropic is eyeing Samsung’s 2-nanometer SF2P process, a node built for data-center chips that uses a gate-all-around transistor design to cut power leakage. Samsung also makes the high-bandwidth memory that many AI accelerators depend on, so a single partner could supply both the logic and the memory.

There’s a money angle too. Samsung, SK Hynix, and Micron all took part in Anthropic’s $65 billion funding round in May, which gives the two sides an existing relationship to build on.

Samsung’s foundry ambitions are heating up on several fronts. Google is separately weighing whether to use Samsung for part of a future tensor processing unit. And Samsung Group and SK Group recently committed a combined $518 billion to build four memory-chip plants in South Korea. The scale of capital pouring into the Korean chip industry is hard to overstate.

The Push to Escape Nvidia

Anthropic isn’t acting in a vacuum. Its timing lines up almost exactly with a move by its biggest rival.

Last week, OpenAI unveiled its first custom chip, a Broadcom-built inference processor named Jalapeño, designed to run large language models more efficiently and lean less on Nvidia. Amazon, Google, Meta, and Microsoft have all built their own silicon to trim their reliance on outside suppliers.

The reasons come down to two things:

• Cost and speed. A chip designed for one company’s exact needs can run faster and cheaper than a general-purpose one.
• Freedom from a single supplier. Leaning on one vendor is risky when demand keeps climbing. Owning the hardware means more control.

Yet Nvidia hasn’t given up an inch. Even amid the custom-chip arms race, The Information’s estimates put Nvidia’s AI chip market share at 74%, higher than before the inference-chip scramble began.

The wins ripple through the market. If Samsung landed a marquee client like Anthropic, it would pose a direct threat to TSMC’s foundry dominance. When the news broke, shares of Roblox and Unity moved, and chip names across the board reacted.

What It Means for Anthropic’s Compute Strategy

A custom chip wouldn’t replace Anthropic’s existing deals. It would add another layer of control on top of a deliberately mixed hardware stack.

Anthropic is, by its CFO’s account, the only frontier lab running its models across all three major chip families: Amazon’s Trainium, Google’s TPUs, and Nvidia’s GPUs. That flexibility took years of investment in compilers and orchestration to build.

The commitments are already massive. Anthropic signed a deal with Google and Broadcom in April for roughly 3.5 gigawatts of TPU compute starting in 2027, alongside a separate Amazon Trainium agreement, pushing total compute pledges past $100 billion.

The economics make custom silicon more tempting by the day. Anthropic’s annualized revenue run rate crossed $30 billion in early 2026, more than tripling from about $9 billion at the end of 2025. As the company confirmed in April:

Our run-rate revenue has now surpassed $30 billion, up from approximately $9 billion at the end of 2025.

The hiring tells its own story. Anthropic recently brought on Clive Chan, an engineer who helped build OpenAI’s custom chip program, a sign the project has moved past pure exploration.

Whether Samsung or someone else ends up making a chip for Anthropic is still an open question. The company may never ship one at all. But the direction across the industry, away from total dependence on Nvidia, is now unmistakable.

Digital Trendings is your trusted source for AI news and updates, stay tuned for more.