Anthropic, the artificial intelligence startup valued at $965 billion, is in talks with Samsung Electronics to manufacture its own AI chips using Samsung's 2-nanometer process, The Information reported on July 2. The move would let the company design custom silicon and reduce its reliance on NVIDIA's GPUs, a shift that has gained traction across the tech industry.
Anthropic hired Clive Chan last month to lead the in-house chip effort. Chan previously worked on custom chip development at OpenAI. The company is also in discussions with multiple chip design firms. Once the detailed design is finalized, Anthropic is considering Samsung's 2-nanometer foundry process and advanced packaging - the same technology Samsung used to build Tesla's AI chips.
Samsung's foundry gains a possible edge
Securing the Anthropic order would give Samsung's foundry business a second high-profile win after the Tesla deal. Industry observers see the potential collaboration as a way for Samsung to build momentum against TSMC in the advanced chip manufacturing race.
The industry-wide shift from NVIDIA GPUs
Developing custom AI accelerators has become a strategic priority for large tech firms. Google has long used its own Tensor Processing Units (TPUs). OpenAI recently unveiled its first custom inference chip, "Jalapeno," built with Broadcom. Anthropic's chip plans follow the same logic.
Anthropic, however, avoided committing to a timeline. A spokesperson said, "NVIDIA's GPUs, Google's TPUs, and Amazon Web Services' (AWS) 'Trainium' chips will continue to play a central role in Anthropic's computational resources."
Strategic ties with memory chipmakers
In May, Anthropic closed a $65 billion investment round that included Samsung Electronics, SK Hynix, and Micron as infrastructure partners. At the time, the company said, "These technologies play a key role in supplying memory, storage devices, and logic chips worldwide." The partnership sparked speculation in South Korea that Anthropic might tap Samsung's foundry, since Samsung is the only one of the three memory suppliers that also operates a chip manufacturing business.
The Information first reported the chip development talks. The Information did not specify a timeline for the design or production process.
Why this matters for IT and development
Custom AI silicon changes the compute landscape that developers and infrastructure teams navigate. If Anthropic ships its own chip, inference and training workloads may need to target multiple accelerator families - NVIDIA CUDA, Google TPUs, AWS Trainium, and now a new architecture from one of the largest private AI labs. For IT architects, this means provisioning decisions will require deeper evaluation of chip-specific performance and cost profiles. For developers, optimizing models for custom hardware could become a routine part of deployment pipelines, not just a niche concern in hyperscale environments.
Your membership also unlocks:
