Anthropic on August 5, 2026, publicly confirmed for the first time that it is assembling an in-house team to design custom artificial intelligence (AI) chips for its claude family of Large Language Models (LLMs). <cite index="2-1,2-6">The company confirmed the plans through a spokesperson statement to Business Insider</cite>, ending months of industry speculation that had followed earlier reporting from Reuters and The Information.
The Co-Design Strategy
<cite index="2-7">According to the spokesperson, Anthropic aims to co-design hardware and models so that Claude runs faster and more efficiently at the scale users require.</cite> <cite index="15-11">A company spokesperson confirmed the program will follow a software-hardware co-design strategy — meaning the chip and the Claude model will be developed together, each shaping the other's architecture.</cite> <cite index="2-10">A recently posted job listing refers to a "custom silicon team," with Anthropic seeking engineers with broad expertise in chip design and verification.</cite>
<cite index="5-6">Last month, *The Information* reported that Anthropic was scouting Samsung as a potential partner for building such chips.</cite> <cite index="7-3">Anthropic did not disclose when its first custom chip could be ready or whether it intends to manufacture the processors itself.</cite>
A Multi-Chip Approach, Not a Departure
Despite the new initiative, Anthropic has been explicit that the custom silicon effort does not replace its existing supplier relationships. <cite index="7-2">In a statement reported by Business Insider, the company said that the new team forms part of its broader "multi-chip" strategy and will complement — not replace — its existing use of hardware from Amazon Web Services (AWS), Google, Nvidia, and AMD.</cite>
<cite index="8-3,8-4,8-5">Anthropic says it trains and runs Claude on AWS Trainium, Google Tensor Processing Units (TPUs), and Nvidia Graphics Processing Units (GPUs), and that Amazon remains its primary cloud provider and training partner. The company and Amazon announced an agreement in April covering up to 5 gigawatts of new capacity and more than $100 billion in AWS technology commitments over ten years, with Anthropic already using more than one million Trainium2 chips to train and serve Claude.</cite> <cite index="8-8">Anthropic also announced a strategic AMD partnership in July covering up to 2 gigawatts of MI450-series GPUs, with the first gigawatt scheduled to begin deployment in the first half of 2027.</cite>
Revenue Scale Makes the Investment Plausible
<cite index="1-7">Anthropic's Annual Recurring Revenue (ARR) run-rate has passed $30 billion, up from about $9 billion at the end of 2025, with more than 1,000 business customers each spending over $1 million a year.</cite> <cite index="12-4">As model sizes continue to grow, compute costs for training and inference have become one of the largest expenses for AI labs.</cite> <cite index="6-6">According to industry sources cited by Reuters, the price tag for developing an advanced AI chip can approach half a billion dollars, driven by the need for specialized engineering talent and the high cost of achieving reliable, defect-free fabrication.</cite>
Broader Industry Trend
Anthropic's move reflects a pattern now well-established among frontier model labs. <cite index="24-10,24-11">OpenAI unveiled "Jalapeño," its first proprietary chip, an Application-Specific Integrated Circuit (ASIC) built with Broadcom and aimed specifically at inference — the work of running a trained model rather than teaching it — as a deliberate step toward a vertically integrated stack OpenAI controls from the model down to the silicon.</cite> chatgpt's parent company has been among the most prominent drivers of this structural shift.
<cite index="12-6">The decision reflects a broader industry trend: leading AI labs are increasingly moving toward vertical integration, extending their reach from model development down to the underlying hardware layer to reduce dependence on a limited number of chip suppliers.</cite> <cite index="13-3">As model scale and commercial application scenarios expand dramatically, optimizing hardware for specific model architectures and inference tasks has become a critical path to reducing costs and boosting performance.</cite>
<cite index="6-5">The company offered no indication of when its chip efforts might bear fruit, nor did it clarify whether Anthropic plans to handle manufacturing on its own.</cite> The absence of a timeline underscores the scale of the undertaking: <cite index="13-5,13-6">developing custom chips is an endeavor requiring massive investment, long cycles, and extremely high risk, where a misstep at any stage from architecture design to tape-out could result in hundreds of millions of dollars going to waste.</cite>