The AI chip race just gained another heavyweight. Anthropic is exploring the development of a custom AI chip and is in talks with Samsung Electronics. The maker of Claude is following the same playbook as OpenAI, Google, Amazon, and Microsoft: reduce reliance on Nvidia and take tighter control of its AI infrastructure.
According to The Information, citing multiple sources, the project is still in a very early stage. Anthropic hasn’t chosen a final design and is evaluating what tasks the chip should handle, how much compute it needs, and how the hardware would slot into servers. TechCrunch confirmed Anthropic didn’t deny the talks with Samsung but stressed no formal partnership has been announced.
Why build a custom AI chip?
More AI companies want to loosen their dependence on Nvidia. The GPU giant dominates AI processors, but surging demand has led to years of supply constraints, high prices, and limited availability.
A homegrown chip offers several strategic upsides:
- lower cost per AI request;
- better energy efficiency;
- more control across the full AI stack;
- less reliance on external suppliers;
- optimization for in-house models like Claude.
For AI vendors, running models—known as inference—is becoming the biggest priority. As hundreds of millions use AI assistants daily, inference now represents the largest cost line for generative AI providers.
Samsung eyes a pivotal win
If Anthropic ultimately chooses Samsung, it would mark a significant victory for the South Korean chipmaker.
Today’s cutting-edge AI chip market is dominated by Taiwan’s TSMC, which manufactures nearly all top-tier AI processors for Nvidia, AMD, and Apple. Samsung has been pushing to claw back share with its latest 2-nanometer process and is investing billions in its foundry division.
The Information reports Samsung is actively pitching its 2nm technology to major AI players. Google is also exploring whether future TPU processors could be partially produced by Samsung.
Samsung also invested in Anthropic earlier this year during a major funding round, creating an existing business tie between the two companies.
Following OpenAI’s lead
The timing stands out. Just days ago, OpenAI unveiled its first custom AI chip, Jalapeño, developed with Broadcom.
That processor is purpose-built for inference—running AI models like ChatGPT. OpenAI claims significantly better performance per watt than current solutions and frames it as the start of a multi-year chip strategy. Reuters reported the move also aims to reduce dependence on Nvidia.
Anthropic now appears to be taking the same path.
Like OpenAI, Anthropic operates fast-growing AI services that demand massive compute. Claude is available via Amazon Web Services, Google Cloud, and a range of enterprise platforms.
Custom silicon is becoming table stakes
Anthropic isn’t alone. Nearly every major AI player is building specialized AI processors.
Among them:
- Google has spent years developing its Tensor Processing Units (TPUs);
- Amazon builds Trainium and Inferentia chips;
- Microsoft is investing in the Maia AI chip;
- Meta is working on multiple generations of in-house AI silicon;
- OpenAI recently introduced Jalapeño.
As a result, competition is shifting from just AI models to the entire stack behind them.
Where companies once battled over better language models, the next phase is about who can deliver the cheapest, fastest, most energy-efficient AI infrastructure.
Still early days
Even if the talks with Samsung are serious, it’s important to note Anthropic hasn’t made a final call.
Reports indicate the company is still assessing:
- which architecture the chip should use;
- which chip design partners to involve;
- which manufacturing partner to select;
- when a first prototype could be ready.
Designing an advanced AI chip typically takes years. Even if Samsung manufactures it, the processor likely won’t hit data centers for some time.
Anthropic told TechCrunch that, for now, its AI infrastructure will continue to run on a mix of hardware from Nvidia, Google, and Amazon. The company declined to comment further on the Samsung talks.
Why it matters
A potential Anthropic–Samsung deal underscores how fast AI is evolving. The contest is moving beyond software to full-stack control of hardware, manufacturing, and data centers.
For Samsung, a win here would be a key step in clawing back ground from TSMC. For Anthropic, a custom chip means tighter control over Claude’s cost, performance, and scale.
It also highlights a broader shift: the biggest AI companies are increasingly building end-to-end platforms—from language model to processor.
Latest comments
Loading