Anthropic создаёт команду для разработки собственных AI-чипов под модели Claude

On August 5, 2026, Anthropic announced the formation of a dedicated engineering team to develop its own AI chips. According to the company's statement, the maker of Claude will co-design hardware and models together — so that the technology runs faster and consumes fewer computational resources.

Why Anthropic Needs Its Own Silicon

Anthropic is betting on co-design — jointly developing hardware and algorithms from scratch, rather than adapting off-the-shelf GPUs to the model's needs. Standard NVIDIA chips were originally designed for a broad class of computations: scientific calculations, 3D graphics, and data processing. Transformer models like Claude represent a highly specialized workload — continuous matrix operations over enormous tensors with specific memory bandwidth requirements. For such a workload, a general-purpose GPU is structurally redundant: it contains blocks that are not engaged when running a language model but still consume power and add latency. The efficiency gap grows as model size increases: the more parameters Claude has, the more pronounced the advantage of a chip tailored to that specific architecture.

Custom silicon optimized for the mathematics of specific models reduces the cost of inference — a key operational expense for a company handling millions of requests daily. Over the long term, optimized hardware directly affects margins and API accessibility for enterprise clients.

In the AI industry, this path has already been taken by the largest players: Google has used its own TPUs since 2016, Amazon deployed Trainium for training and Inferentia for inference within AWS, and Meta is developing MTIA for internal AI infrastructure. For Anthropic, which operates through AWS and Google Cloud, its own chip also means reduced dependence on external suppliers at a critical link in the production chain.

What Co-Design Is and How It Works

According to Anthropic's announcement, the company will design hardware and models within a single team — the chip architecture is shaped around Claude's specific computational patterns, while algorithms are adapted to the capabilities of the hardware being developed iteratively, not sequentially.

"We will co-design the hardware and models so that our technology runs faster and more efficiently," reads

Anthropic's official announcement about hiring a chip team.

This approach requires specialists with profiles atypical for an AI lab:

• Chip architecture — designing the computational blocks of an accelerator
• VLSI design — creating physical integrated circuit layouts
• AI hardware acceleration — hardware optimization for neural network tasks
• Hardware-software co-design — aligning hardware and the software stack

Specialists with such skills were previously concentrated at companies like Intel and Qualcomm, as well as AI startups — Groq (accelerators for LLM inference) and Tenstorrent. The appearance of such positions at Anthropic is a signal of a serious long-term bet on the hardware vertical.

What This Means

Anthropic is moving toward vertical integration — from an AI lab to a company controlling the stack from silicon to the user interface. Building a chip team is a long-term project: the first commercial results will not appear for several years at the earliest. But for a company focused on the enterprise market, inference unit economics are critical: even the smallest reduction in computation cost at Claude's scale translates into millions of dollars in savings. Forming a chip team is not yet a product, but a strategic signal that Anthropic views hardware control as a long-term competitive advantage.

*Meta has been recognized as an extremist organization and is banned in Russia.