TL;DR
· Anthropic is reportedly exploring self-developed AI server chips, but design, tape-out, or mass production plans have not yet been confirmed.
· OpenAI has already disclosed the Jalapeño inference chip and begun testing, with plans to deploy by the end of 2026.
· Samsung may benefit from the trend of outsourcing AI chips, but Anthropic still relies on AWS, Google, and NVIDIA computing power in the short term.

Discussion around Anthropic developing its own AI server chips is heating up, but this is not yet a confirmed chip order line. The core concern for outsiders is that the inference costs behind Claude, GPU supply, data center power, and rack capacity are becoming hard constraints for large model companies. OpenAI has already disclosed its Jalapeño inference chip developed with Broadcom, and Anthropic is reportedly evaluating specialized chips better suited for its own models. However, based on publicly available information, whether Samsung is involved in manufacturing or whether the project has entered formal design has not been confirmed.

Anthropic Still in Early Exploration, Not on the Eve of Mass Production

The direction Anthropic is reportedly exploring is a server chip more suitable for its own AI model operations. Compared to general-purpose GPUs, a custom chip, if successfully designed, could reduce costs for specific inference tasks, improve energy efficiency, and reduce reliance on external chip supply.

The difficulty of such chips is not just about single-chip performance. Large model companies need to simultaneously handle computing speed, memory bandwidth, interconnect networks, power consumption, thermal dissipation, and cluster stability. The real challenge is getting thousands of chips to work stably together in a data center and continuously serve training or inference tasks.

A more accurate description at this stage is that Anthropic is still in the early evaluation and definition phase. Which specific AI tasks the chip will primarily handle, how performance and power consumption targets will be set, how server and cluster integration will work, and whether an external chip design company needs to be involved—none of these questions have clear public answers yet.

The company's external messaging also remains cautious. In April of this year, Anthropic announced an expansion of its partnership with Amazon, committing over $100 billion to AWS technology over the next decade, locking in up to 5GW of capacity, and stating that it has already used over 1 million Trainium2 chips to train and serve Claude. Anthropic also emphasized a multi-hardware strategy, but AWS remains its primary training and cloud service provider.

This means that even if the self-developed chip exploration continues, it will be difficult to replace existing suppliers in the short term. AWS Trainium, Google TPU, and NVIDIA GPUs remain important components of Anthropic's scaled computing power system.

OpenAI Takes the Lead, Inference Cost Pressure is More Direct

One important backdrop to Anthropic being placed in the discussion of self-developed chips is that OpenAI has already set a precedent.

Broadcom's official announcement shows that OpenAI and Broadcom released Jalapeño on June 24, 2026, positioning it as an accelerator for large language model inference, also referred to as an Intelligence Processor. OpenAI and Broadcom stated that this chip went from initial design to manufacturing tape-out in about 9 months, with engineering samples already running in the lab, and plans to begin deployment by the end of 2026.

A distinction needs to be made here between two stages. Jalapeño has been released and entered testing, but that does not mean it is already in large-scale commercial use. It represents that leading model companies are starting to bring inference costs under deeper hardware control, not that GPU demand is about to be replaced.

Inference is the computing process by which a model generates answers after a user asks a question in ChatGPT, Claude, or other products. Compared to training, inference occurs more frequently, and as user scale grows, cost pressure will continue to rise. For large model companies, even a small percentage reduction in single inference costs can translate into significant savings when applied across massive request volumes and long-term data center expenditures.

Anthropic's pace is clearly earlier. It has not released chip specifications, nor has it disclosed performance metrics, partner lists, or mass production timelines. OpenAI's progress simply shows the market a direction: the largest model companies are no longer just buying GPUs; they are also trying to bring part of their computing power infrastructure under their own control.

Samsung's Imagination Sparks Heat Up, But Orders Are Not Finalized

Samsung is attracting market attention because it has advanced manufacturing capabilities and is also vying for more AI chip foundry opportunities. After news emerged about Anthropic's funding and infrastructure partnerships, outsiders naturally linked Samsung to potential AI accelerator manufacturing opportunities.

But this needs to be viewed with caution. What can be confirmed from public information is that Samsung, SK Hynix, Micron, and other companies have appeared in discussions about Anthropic's infrastructure partners. Micron announced on June 22, 2026, a strategic agreement with Anthropic that includes memory and storage AI architecture design, supply agreements, internal adoption of Claude at Micron, and a strategic investment in Anthropic's Series H.

These cooperation signals cannot be directly equated to Samsung having already secured a chip order from Anthropic. Claims that Anthropic has contacted Samsung regarding manufacturing collaboration are not well supported by publicly verifiable information. A safer assessment is that if Anthropic's self-developed chip project advances to the manufacturing stage, Samsung could become a potential participant of market interest, but it cannot yet be written as a confirmed deal.

For chip projects, from early evaluation to final mass production, there are still stages such as architecture definition, design verification, manufacturing process selection, packaging testing, and supply chain coordination. As long as the chip design is not finalized, the foundry role is also hard to truly settle.

Hiring Moves Increase Credibility, But the Roadmap Remains Uncertain

Talent moves have drawn more attention to Anthropic's hardware clues. According to reports, Clive Chan, an early member of OpenAI's custom chip team, has joined Anthropic. Public records show he was involved in the early buildup of OpenAI's chip team and also has experience with Tesla Dojo. Anthropic has also been ramping up recruitment for chip engineers recently.

This suggests the company is at least preparing for hardware capabilities. For a model company, a hardware team that understands models, inference workloads, and data center systems can help determine which tasks are suitable for custom chips and which should still rely on GPUs, TPUs, or cloud provider chips.

But talent additions and hiring expansion are still just signals of early investment. Whether the project can continue depends on whether the chip can achieve sufficient advantage in cost, performance, energy consumption, and deployment complexity. If a custom chip only shows efficiency gains on paper but cannot operate stably at scale, or if manufacturing and software adaptation costs are too high, the company may still primarily rely on external chips.

This is also why NVIDIA is not easily replaceable in the short term. NVIDIA GPUs remain the mainstay for AI training and inference, with a mature software ecosystem, and data center customers have already built extensive systems around its platform. Custom chips are more likely to offload part of the workload in specific inference scenarios rather than fully replace GPUs.

For investors, the real impact of the discussion around Anthropic's self-developed chips is more like supply chain maneuvering in the short term. Leading model companies want more options for computing power, and cloud providers, Broadcom, Samsung, TSMC, memory manufacturers, and advanced packaging supply chains could all benefit from this trend. However, in Anthropic's case, the clear facts are still limited: self-developed exploration is still early, Samsung's role is unconfirmed, and Claude's scaled computing power still relies on AWS, Google, and NVIDIA.

Click to learn about job openings at BlockBeats

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Chat Group: https://t.me/BlockBeats_App

Twitter Official Account: https://twitter.com/BlockBeatsAsia