Introduction
In a strategic move aimed at enhancing its technical capabilities and reducing absolute reliance on specific providers, reports recently published by the specialized news outlet The Information last Saturday indicated that Anthropic, one of the leading laboratories in artificial intelligence research, is currently holding advanced and serious discussions to purchase computer inference chips from Fractile. The latter is a promising startup headquartered in London, specializing in the development of hardware specifically designed to run massive artificial intelligence models with higher efficiency and speed than usual. These ongoing talks represent the latest strategic move by the fast-growing AI lab to expand its chip supply scope and offerings, especially as unprecedented and increasing demand for the company’s popular chatbot, Claude, places immense and clear pressure on its currently available total server capacity.
- Growing appetite for computing and skyrocketing growth rates
- Why choose Fractile’s technology?
- Broader chip source diversification in the tech market
- Frequently asked questions
Growing appetite for computing and skyrocketing growth rates
These important discussions and negotiations come amidst a period of exceptional and remarkable growth recently witnessed by Anthropic. The company explicitly revealed in early April that its annualized revenue rate has crossed the $30 billion barrier, a stunning, skyrocketing increase compared to only about $9 billion at the end of 2025. TechCrunch reported that the company’s actual current revenue rate may now be much closer to $40 billion. Compared to its market competitors, OpenAI confirmed monthly revenues of about $2 billion, equivalent to an annualized rate of $24 billion, placing Anthropic firmly in the lead and ahead of its primary and direct competitor in terms of total revenue, despite its consumer and retail user base being much smaller compared to OpenAI’s.
This explosive and steady trajectory in growth and profits has undoubtedly intensified the company’s urgent and ongoing need to secure massive computing power to run its servers. In this regard, Anthropic announced in early April a long-term strategic partnership with both Google and Broadcom aimed at acquiring several gigawatts of next-generation Tensor Processing Unit (TPU) capacity, which is expected to go live and become available online starting in 2027. Reuters also reported in April that Anthropic is strongly and seriously exploring the possibility of designing and manufacturing its own custom chips, although this ambitious effort is still in its very early stages and a dedicated team has not yet been formed or assembled to accomplish it.
Why choose Fractile’s technology?
Fractile, newly founded in 2022, emerged from stealth mode after securing $15 million in seed funding. More importantly, it introduced an innovative chip design architecture that breaks away entirely from the traditional GPU designs sold and dominated by Nvidia. The startup’s innovative technical approach is based on the concept of storing data required for computational operations directly alongside the transistors executing the actual work. This radical design aims to mitigate and solve the common memory bottleneck problem that severely and frustratingly limits the speed and performance of traditional AI hardware. Fractile has claimed with some confidence that its extensive simulations and testing clearly show this unique design can run large language models at astronomical speeds up to 100 times faster than current GPUs available in the market, and at a cost not exceeding one-tenth of the current cost.
Last February, the UK government announced major support for the tech sector, reporting that Fractile will invest £100 million over three consecutive years to significantly expand its operations in the country, including establishing a new, advanced hardware engineering and development center in Bristol. By late March, the Financial Times reported that the company was in serious talks to raise additional funding of over $200 million at a valuation of $1 billion, with potential participation from major influential backers such as the NATO Innovation Fund and Accel among the list of potential investors.
Broader chip source diversification in the tech market
To date, no precise details regarding the purchase price, contract terms, or actual timeline for any potential deal between Anthropic and Fractile have been officially disclosed. Furthermore, Fractile’s advanced and exciting chips have not yet entered commercial or mass production, with their official market launch expected by 2027. Nevertheless, these talks reflect a clear commitment and broader trend among leading global AI companies, including tech giants like Amazon, Meta, and OpenAI, to seriously seek strong and effective alternatives to Nvidia’s currently dominant GPUs by relying directly on custom or specialized silicon chips. Meanwhile, Anthropic is actively and aggressively seeking a massive potential funding round of $50 billion, targeting a valuation of $900 billion, a staggering valuation that would surpass OpenAI’s $852 billion valuation recorded earlier this year.
Frequently asked questions
Question: What technology is Anthropic seeking to buy from Fractile?
Answer: Anthropic is seeking to buy innovative, advanced inference chips specifically designed to accelerate AI model processing and reduce energy consumption.
Question: What distinguishes the chips developed by Fractile from others?
Answer: The chips feature a unique design that integrates data storage with processing transistors to avoid memory lag, increasing speed by 100 times and lowering costs by 10 times compared to the current status quo.
Question: Why is Anthropic interested in expanding and updating its computing supplies now?
Answer: Due to the company’s rapid and massive growth, with revenues exceeding $30 billion, alongside intense pressure on its servers to run and meet the demands of the Claude AI chatbot.
Question: When are Fractile’s chips expected to become available for commercial use?
Answer: The development phase is expected to conclude, with the chips entering production and commercial launch in 2027.
