Key Takeaways
- Anthropic is exploring a custom chip partnership with Samsung, though purpose and specs are not yet decided
- The move follows OpenAI's announcement of its own custom inference chip 'Jalapeño' with Broadcom
- Anthropic confirmed it will continue using a diversified hardware stack including Google, Amazon, and Nvidia chips
Anthropic is in early talks with Samsung about developing a custom AI chip, according to The Information. The discussions remain preliminary. Anthropic hasn't decided what the chip would do, where it would sit in a server, or how powerful it needs to be.
When TechCrunch asked for comment, Anthropic said its compute strategy will continue to rely on a diversified hardware stack, including chips from Google, Amazon, and Nvidia. On Samsung specifically, the company declined to elaborate.
Why is Anthropic exploring custom silicon?
Reuters reported back in April that Anthropic was considering its own AI chips as a hedge against persistent shortages. Nvidia controls over 90% of the AI training chip market, and demand continues to outpace supply. Building custom silicon gives AI companies two things: hardware optimized for their specific workloads, and less dependence on a single supplier that everyone else is also fighting to buy from.
Anthropic isn't alone in this calculation. OpenAI announced last week that it's partnering with Broadcom on a custom inference processor called "Jalapeño." OpenAI claims the chip delivers better performance-per-watt than competitors. Google has its TPUs. Amazon has Trainium and Inferentia. The AI infrastructure race is no longer just about models. It's about owning the stack.
Why Samsung?
Samsung operates the world's second-largest chip foundry after TSMC. The company already manufactures chips for Nvidia and uses Nvidia's software in its fabrication process. Samsung and Nvidia are building an AI chip factory together in South Korea. Samsung has also held discussions with Google about chip manufacturing.
For Anthropic, Samsung offers scale and existing AI chip expertise without going through TSMC, which is already stretched thin by demand from Apple, Nvidia, AMD, and others. A Samsung partnership could also provide geographic diversification, relevant given ongoing export restrictions and geopolitical uncertainty around Taiwan.
What's actually decided?
Not much. The Information's report emphasizes that Anthropic hasn't settled on the chip's purpose. Training chips require different architectures than inference chips. The power envelope, memory bandwidth, and interconnect design all depend on the use case. Until Anthropic answers these questions, a Samsung partnership remains exploratory.
Anthropic has raised over $6 billion from investors including Amazon ($4 billion) and Google ($2 billion). Both Amazon and Google already offer custom AI chips through their cloud platforms, which Anthropic uses. Any Anthropic-Samsung chip would need to fit into, rather than disrupt, those existing relationships.
The competitive pressure from OpenAI
OpenAI's Jalapeño announcement last week adds urgency. If OpenAI can reduce inference costs through custom silicon, it gains margin that can fund more research or undercut Anthropic on pricing. Custom chips also create differentiation that's hard to copy. Anyone can buy Nvidia GPUs. A chip designed for your specific architecture is a moat.
Anthropic's response, even an early-stage one, signals it won't cede this ground without a fight.
Logicity's Take
The real story isn't the Samsung talks themselves. It's that Anthropic is now publicly acknowledging custom silicon as part of its roadmap. Anthropic's $4 billion Amazon investment means it's deeply tied to AWS infrastructure, including Amazon's Trainium chips. A Samsung partnership would represent a hedge against that dependency. For enterprise buyers evaluating AI vendors, the infrastructure question matters: a company that controls its own chips can make longer-term pricing commitments. Watch whether Anthropic goes the training or inference route. Inference chips are cheaper and faster to iterate. Training chips signal a company planning to stay independent at the frontier.
Frequently Asked Questions
Is Anthropic building its own AI chip?
Anthropic is in early discussions with Samsung about a custom chip, but hasn't decided on specifications or purpose yet. The company says it will continue using chips from Google, Amazon, and Nvidia.
Why are AI companies making custom chips?
Custom chips let AI companies optimize hardware for their specific workloads and reduce dependence on Nvidia, which dominates the market. They can also lower inference costs and create competitive differentiation.
How does this compare to OpenAI's chip plans?
OpenAI announced its custom inference chip 'Jalapeño' with Broadcom last week. Anthropic's Samsung discussions are earlier stage and haven't specified whether the chip would be for training or inference.
Who manufactures AI chips for Anthropic currently?
Anthropic uses a mix of hardware from cloud partners. Amazon provides Trainium chips through AWS, Google offers TPUs, and Nvidia GPUs remain part of the stack.
How Claude performs against real-world financial benchmarks
Need Help Implementing This?
Evaluating AI infrastructure for your organization? Logicity helps technical teams cut through vendor claims and assess real performance tradeoffs. Reach out at hello@logicity.in.
Source: TechCrunch / Lucas Ropek
Huma Shazia
Senior AI & Tech Writer
Produced with AI assistance and reviewed by the Logicity editorial team. Learn more in our Editorial Policy.
