Anthropic in Early Talks to Add Fractile as Fourth Chip Supplier as Inference Costs Mount

Key facts

• Anthropic has held early talks with Fractile about purchasing inference accelerators.
• Fractile, founded in 2022 by Oxford PhD Walter Goodwin, raised $15 million in seed funding.
• Fractile is in talks to raise $200 million at a valuation exceeding $1 billion.
• Anthropic's annualized revenue run rate passed $30 billion in March, up from $9 billion at end of 2025.
• Anthropic uses chips from Nvidia, Amazon (Trainium), and Google (TPUs), with 3.5GW of TPU capacity from 2027.
• Fractile's chips are not expected to reach commercial readiness until around 2027.
• Nvidia acquired Groq for $20 billion in December and launched its own inference accelerator.

Anthropic Courts London Startup Fractile for Inference Chips

Anthropic has held early-stage discussions with Fractile, a London-based chip startup, about purchasing its inference accelerators, according to people familiar with the matter. The talks would add Fractile as a fourth source of AI server silicon for the Claude developer, which already relies on chips from Nvidia, Google, and Amazon. Fractile's chips are not expected to reach commercial readiness until around 2027, placing any deployment well outside Anthropic's near-term procurement plans. The timeline roughly coincides with Anthropic's Google-Broadcom TPU partnership, which will provide 3.5GW of compute capacity from 2027 through 2031.

Fractile's SRAM Architecture Promises Speed and Cost Gains

Founded in 2022 by Oxford PhD Walter Goodwin, Fractile is developing an inference chip that co-locates memory and compute on the same die using SRAM rather than shuttling data to separate DRAM chips. This data movement between GPU and off-chip DRAM is a major bottleneck in running large AI models at speed. Goodwin told Fortune in July 2024 that Fractile's design stores data needed for computations directly next to the transistors that perform the arithmetic. Based on simulations, Goodwin said Fractile could run a large language model 100 times faster and 10 times cheaper than Nvidia's GPUs, though the company had not yet manufactured test chips.

Anthropic's Diversified Chip Strategy Amid Surging Demand

Anthropic has deliberately avoided dependence on any single chip vendor. It runs Claude on Nvidia GPUs, Amazon's Trainium processors through Project Rainier, and Google's TPUs under a deal announced in October that provided over 1GW of compute capacity. In early April, that expanded to 3.5GW of TPU capacity from 2027 through 2031. The interest in Fractile coincides with surging demand on Anthropic's existing infrastructure. The company's annualized revenue run rate passed $30 billion in March, up from around $9 billion at the end of 2025. However, inference costs have been a drag on gross margins, and unlike OpenAI and xAI, which are building or expanding their own massive data center footprints, Anthropic has opted to rent capacity from multiple providers and negotiate leverage through diversified chip supply.

Fractile's Funding and Competitive Landscape

Fractile raised $15 million in seed funding, co-led by Kindred Capital, the NATO Innovation Fund, and Oxford Science Enterprises. The startup is now in talks to raise $200 million at a valuation exceeding $1 billion, with Founders Fund, 8VC, and Accel among potential investors. The Fractile team reportedly includes engineers from Graphcore, Nvidia, and Imagination Technologies, and the company is building its own software stack alongside the hardware. Fractile is one of several inference-focused startups pursuing SRAM-based or near-memory architectures, including Groq and Cerebras. Nvidia struck a $20 billion acquisition deal with Groq in December and subsequently launched its own dedicated inference accelerator, Groq 3 LPX, acknowledging the growing commercial pressure to optimize cost-per-token at scale.

Financial Sustainability Questions Loom

Anthropic's detailed internal financials remain confidential as a private LLC, fueling speculation about its cash burn and profitability. despite the surging revenue run rate, neither Anthropic nor OpenAI is profitable, and inference costs remain a significant burden. The company's ability to fund commitments to multiple chip partners while maintaining growth remains an open question. Anthropic's strategy of renting capacity rather than building its own data centers may provide flexibility, but it also exposes the company to rising infrastructure costs as demand for Claude continues to climb.

Outlook: A Bet on Future Chip Technology

The early talks with Fractile represent a long-term bet on a technology that may not be ready for years. If Fractile's SRAM-based chips deliver on their promised speed and cost improvements, they could help Anthropic reduce inference costs and improve margins. However, the startup faces significant technical and manufacturing hurdles before commercial deployment. For now, Anthropic's diversified chip strategy positions it to leverage multiple suppliers, but the success of that strategy hinges on the timely delivery of new silicon and the company's ability to manage its financial runway.

The bottom line

• Anthropic is in early talks to add Fractile as a fourth chip supplier, alongside Nvidia, Google, and Amazon.
• Fractile's SRAM-based inference chips promise 100x speed and 10x cost improvements over Nvidia GPUs, but won't be ready until 2027.
• Anthropic's annualized revenue run rate reached $30 billion in March, but inference costs pressure gross margins.
• Fractile has raised $15 million in seed funding and is seeking $200 million at a $1 billion+ valuation.
• Nvidia's $20 billion acquisition of Groq signals industry focus on inference efficiency.
• Anthropic's strategy of renting capacity from multiple providers aims to leverage diversified chip supply and negotiate better terms.