Anthropic is in early talks with Samsung about co-developing a custom AI chip, The Information reported on July 2, 2026. The company has not yet decided what the chip would be used for, how it would be architected, or how powerful it would be. The talks began shortly after OpenAI disclosed its first custom inference processor, called Jalapeno, developed in partnership with Broadcom.
Key takeaways
- Anthropic is talking with Samsung about a custom AI chip - per The Information, July 2, 2026
- Intended use, architecture, and specifications have not been determined yet
- OpenAI announced its Jalapeno inference chip with Broadcom one week earlier, on June 24, 2026
- Anthropic emphasized its diversified hardware stack remains central: Google, Amazon, NVIDIA
- Samsung already manufactures chips for NVIDIA and is developing an AI chip factory in South Korea with them
Early-stage talks
Anthropic told TechCrunch that a diversified hardware stack - covering Google, Amazon, and NVIDIA - remains central to its compute strategy. On the Samsung question specifically, the company said it had nothing further to add. According to The Information, the talks are at a very early stage: Anthropic has not decided what the chip would be used for, how it would fit into a server, or how powerful it would be.
Samsung fits the role of a natural partner. It already manufactures chips that NVIDIA requires for training and running AI models, uses NVIDIA software in its fabrication processes, and is co-building an AI chip factory in South Korea with NVIDIA. Samsung has also held preliminary discussions with Google on TPU chip production.
Context: Jalapeno and the push for custom chips
On June 24, 2026, OpenAI and Broadcom Inc. announced Jalapeno - a custom LLM inference processor. OpenAI claims the chip delivers better performance-per-watt than competing solutions. Developing proprietary AI chips has become a common priority for labs seeking to reduce dependence on NVIDIA's dominant position.
Both Amazon (Trainium) and Google (TPU) already offer custom-built chips for AI workloads as part of their cloud offerings. NVIDIA remains the undisputed market leader, driving other companies to explore alternatives or build their own hardware.
| Company | Chip | Partner | Status |
|---|---|---|---|
| OpenAI | Jalapeño | Broadcom | Announced June 24, 2026 |
| Amazon | Trainium | in-house | In use (AWS cloud) |
| TPU | in-house | In use (cloud) | |
| Anthropic | undecided | Samsung | Early talks |
Why it matters
For Anthropic, a custom chip is not just a cost question. AI companies increasingly face compute shortages that constrain model scaling and drive up operating costs. A proprietary chip design offers the potential to optimize silicon for a specific model architecture - for instance, a wide-context transformer with many layers - and reduces single-vendor concentration risk.
Caution is warranted, however. The talks are at an early stage, Anthropic has not confirmed any specifics, and the path from chip vendor negotiations to production silicon typically takes several years and hundreds of millions of dollars. OpenAI's Jalapeno chip had reportedly been in development since at least 2023.
What's next
- Anthropic has not disclosed any timeline or next steps - the status of the Samsung talks remains unconfirmed by either party
- Any Anthropic/Samsung chip would compete with OpenAI's Jalapeno and Amazon's Trainium line - but only if the project reaches production
- Samsung has confirmed strategic AI chip ambitions (NVIDIA factory, Google TPU talks) - an Anthropic partnership would align with that direction
