Anthropic is moving more aggressively toward its own AI chip strategy, and Samsung Electronics has emerged as a possible manufacturing partner. The talks reportedly center on producing a custom AI chip using Samsung’s 2nm process, one of the most advanced fabrication nodes in semiconductors.
The move reflects a broader shift in the AI industry, where major players are working to reduce reliance on Nvidia GPUs. For companies building large-scale AI systems, custom hardware can lower operating costs, improve efficiency, and better match the performance needs of their models.
Samsung’s 2nm process is the main attraction
Samsung’s 2nm manufacturing capability offers higher transistor density, which can improve both power efficiency and computing performance. That combination is especially important for AI workloads that demand massive throughput with tight energy constraints.
The company also has advanced packaging expertise that can place processors closer to memory chips. By reducing data-transfer bottlenecks, that approach can speed up communication between components and improve overall system responsiveness.
| Focus Area | Why It Matters | Status |
|---|---|---|
| 2nm fabrication | Higher density and better efficiency | Under discussion |
| Advanced packaging | Closer processor-memory placement | Under review |
| Server integration | Adapting the chip to AI infrastructure needs | Not yet finalized |
Design work is still in the early stage
According to The Information, which was cited by TechCrunch, Anthropic has already started discussions with Samsung’s foundry division. However, the project has not yet reached detailed design, testing, or manufacturing.
Anthropic has also not decided what the chip will do, how it will be integrated into servers, or what performance target it should hit. The company is still evaluating technical options and remains in talks with several other chip-design firms.
That includes possible chip use from Microsoft and Fractile, the UK-based startup. The fact that multiple options remain open suggests Anthropic is still shaping its long-term AI infrastructure plan.
New hiring signals internal ambition
Anthropic’s hardware push is also being reinforced by new recruitment. The company hired Clive Chan, one of the early members of OpenAI’s chip team, adding experience from a group that helped build an internal hardware effort from the ground up.
Chan announced the move to Anthropic on X on 6 June. His arrival is likely to strengthen the company’s ability to build a more mature AI hardware team.
That matters because chip development requires expertise across semiconductor design, systems integration, and AI workload analysis. Chan’s background gives Anthropic an experienced hand from a closely related effort at OpenAI.
Funding, partners, and a bigger strategic picture
Anthropic’s hardware ambitions are backed by major capital. On 28 May, the company closed a Series H round worth US$ 65 billion, led jointly by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital.
Samsung Electronics, SK Hynix, and Micron joined as strategic infrastructure partners in that round. After the funding announcement, Anthropic’s valuation reached around US$ 965 billion, surpassing the estimated US$ 852 billion valuation of OpenAI.
Anthropic also said in that announcement that the world’s three largest memory makers play an important role in supplying logic chips for its infrastructure. Samsung is the only one of those partners that also runs a foundry business, which is why the market is watching the possibility of a custom chip deal so closely.
Samsung could gain a stronger foundry position
If production talks lead to an agreement, Samsung would add another high-profile client to its foundry roster. The company already works with major customers such as Tesla and Apple, while Google is also said to be considering Samsung for some of its next-generation TPU chips.
That opportunity matters because TSMC in Taiwan still dominates the AI foundry market. A successful custom chip project could help Samsung strengthen its position in one of the fastest-growing segments of semiconductors.
But the challenge remains significant. The Information noted that the performance of Samsung’s 2nm process will be crucial, especially after the company previously faced yield issues on some manufacturing nodes.
Industry observers also say Samsung Foundry is now more selective about the projects it accepts. The company is said to be prioritizing the most promising customers and programs rather than taking every order.
AI chip competition is widening
Anthropic’s move comes as more AI companies build their own silicon to reduce dependence on Nvidia. Google has long relied on its internal TPU line, and Amazon Web Services has been developing Trainium for AI training workloads.
OpenAI recently introduced its first inference chip, Jalapeño, built with Broadcom. The chip was announced on 24 June and is targeted for use this year, with OpenAI saying it offers stronger energy efficiency and better performance per watt than several competing chips.
Broadcom CEO Hock Tan said Jalapeño is only the beginning of a multi-generation AI chip roadmap. Even so, Nvidia remains the dominant force in AI accelerators, holding about 74% of the market, which explains why the push for alternatives keeps accelerating.
Anthropic has also said its internal chip plan is not meant to replace every hardware platform it uses today. The company expects Nvidia GPUs, Google TPUs, and AWS Trainium chips to remain part of its future compute stack.
