Fractile
Fractile is a UK-based AI hardware startup building SRAM-based in-memory compute inference accelerators paired with a proprietary software stack, targeting frontier AI labs and hyperscalers with first commercial chips expected around 2027. The company is pre-revenue, venture-backed at approximately $1B post-money, and operates from London, Bristol, San Francisco, and Taipei.
- Company typePrivate
- Founded2022
- HeadquartersNewbury, United Kingdom
- Headcount101–250
- GTM typeB2B
- OfferingHardware or Manufacturing
What Fractile does
Fractile is a UK-based AI hardware startup, founded in 2022 by Oxford PhD Walter Goodwin, that is designing next-generation AI inference processors built on a proprietary SRAM-based in-memory compute architecture. In Fractile's design, memory and compute are physically interleaved on the same die, eliminating the data-shuttling bottleneck to off-chip DRAM that constrains conventional GPU-based inference. The company claims this architecture can deliver up to 25x faster inference at 1/10th the cost and 20x better performance per watt versus existing AI hardware, with a stated target throughput of approximately 1,200 tokens per second (versus roughly 40 t/s on current systems). Fractile takes a full-stack approach spanning transistor-level circuit design, silicon, hardware systems, and cloud inference server logic, and is building a proprietary software platform intended to compete with Nvidia's CUDA for developer mindshare.
Fractile's go-to-market targets frontier AI labs and hyperscalers through direct enterprise sales. The company has confirmed early-stage commercial discussions with Anthropic (reported May 2026) as a potential flagship customer for chips shipping around 2027. Manufacturing is outsourced to third-party foundries (with a Taipei office supporting that partner ecosystem), and Fractile is pre-revenue: first commercial chips are not expected until approximately 2027. Pricing is bespoke and negotiated privately, with no list prices published. The company sells integrated inference systems combining its custom chip with its proprietary software stack, positioning for long-context, reasoning, and agentic-coding workloads that are economically constrained on today's GPU hardware.
Fractile is venture-backed at scale, with cumulative disclosed funding of roughly $260M+ across a $15M seed (July 2024, co-led by Kindred Capital, NATO Innovation Fund, and Oxford Science Enterprises), a £5M ARIA grant (October 2024), a $22.5M follow-on (January 2026), and a $220M Series B (May 2026, co-led by Accel, Factorial Funds, and Founders Fund at approximately $1B post-money). The company operates from four offices — London (HQ), Bristol (engineering), San Francisco (US), and Taipei (manufacturing support) — and is investing £100M over three years in UK operations including a new Bristol hardware engineering facility. Headcount has grown from 14 (mid-2024) to 80 (February 2026), with 56 open roles listed on its Greenhouse job board across Silicon, Software, Hardware Test, Manufacturing Operations, and Hardware Systems as of mid-2026.
Fractile firmographics
Firmographics- Name
- Fractile
- Legal name
- Fractile
- Website
- https://fractile.ai
- Company type
- Private
- Founded year
- 2022
- Operating status
- Operating
- Headcount range
- 101–250 employees
- Short description
- Fractile is a UK-based AI hardware startup building SRAM-based in-memory compute inference accelerators paired with a proprietary software stack, targeting frontier AI labs and hyperscalers with first commercial chips expected around 2027. The company is pre-revenue, venture-backed at approximately $1B post-money, and operates from London, Bristol, San Francisco, and Taipei.
- Ownership category
- akta.pro rank
Fractile industry classification
Industry- NAICS
- Computer Systems Design and Related Services (5415), Computer Systems Design Services (541512)
- SIC
- Electronic Computers (3571)
- akta.pro industry
- Compute & Acceleration ICs (CPU/GPU/AI/FPGA/ASIC)AI Inference Hardware & Edge AcceleratorsHardware, Technology & Telecom/Artificial Intelligence & Machine Learning/AI Infrastructure & Compute (GPUs/ASICs, accelerators)
Keywords
Where Fractile is headquartered
LocationHeadquarters
- HQ city
- Newbury
- HQ country
- United Kingdom
- HQ region
- Europe
Offices4 records
Markets served
Fractile business model
Business model- GTM type
- B2B
- Offering type
- Hardware or Manufacturing
- Cost components
- Technology or R&D, Personnel, Infrastructure, Supply Chain, Operations, Marketing or Sales
Revenue model
- AI inference chip and system sales: Fractile is pre-revenue, building chips and systems for commercial deployment targeted around 2027. Future revenue will come from selling its inference accelerators (and integrated systems) to AI labs, hyperscalers, and other large compute buyers; chips will be sourced from third-party foundries (no in-house manufacturing indicated).
- Pre-commercial strategic supply agreements: Fractile is in early-stage discussions with AI labs such as Anthropic for potential multi-year chip supply deals when its hardware becomes commercially available in 2027, indicating future recurring, contracted hardware revenue streams anchored by flagship customers.
Pricing tiers
| Model | Billing | Price |
|---|---|---|
| Unit Pricing | Multi-year contract | Quote-based enterprise pricing (not publicly disclosed) |
Go-to-market motion2 records
Distribution channels2 records
Marketing channels5 records
Fractile product offering
Product offeringCore offering
Fractile is building next-generation AI inference processors using a custom SRAM-based in-memory compute architecture that physically interleaves memory and compute on the same die, eliminating the off-chip DRAM bottleneck of GPU-based inference. The company takes a full-stack approach spanning transistor-level circuit design through cloud inference server logic, and ships integrated inference systems combining the chip with a proprietary software stack designed to compete with Nvidia CUDA. First commercial chips are targeted for 2027, with stated targets of up to 25x faster inference at 1/10th the cost of existing GPU systems and ~1,200 tokens/sec throughput.
Product overview
Fractile offers a single vertically integrated product line — its AI inference processor and surrounding software stack — rather than a platform-plus-modules architecture. The offering comprises three tightly coupled components: (1) the Fractile AI Inference Chip, a custom SRAM-based in-memory compute accelerator that interleaves memory and compute on-die; (2) the Fractile Software Stack, a proprietary full-stack software platform spanning from cloud inference server logic to foundational AI research; and (3) Fractile Inference Systems, the integrated deployment units that combine the chip and software for hyperscaler and frontier AI lab customers. Together, these components target the AI inference bottleneck, with the company claiming up to 25x faster inference at 1/10th the cost versus conventional GPU systems, and targeting ~1,200 tokens/second throughput per user versus ~40 tokens/second today.
Differentiator
Problem solved
Functional benefit
Quantifiable outcome
- Runs frontier models up to 25x faster and at 1/10th the cost of existing systems
- +4 more outcomes
Companies that use Fractile
Customer profileNamed customers1 record
Segments3 records
Ideal customer profiles3 records
Fractile technology and API
TechnologyAPI detail
- Has API
- No
- API docs
- API detail
Core technology
AI maturity
App detail
AI capability4 records
Feature6 records
Fractile partnerships and signals
Strategic signalPartnerships
Three partnerships are on record, tiered flagship and core.
- AnthropicflagshipAnthropic is in early-stage discussions to purchase Fractile's AI inference chips when commercial hardware becomes available in 2027. Reported (May 2026) by The Information and corroborated by Reuters, Bloomberg, Tom's Hardware, and others. The talks are part of Anthropic's strategy to diversify away from Nvidia alongside existing deals for Google TPUs and AWS Trainium.
- ARIA (Advanced Research + Invention Agency)coreUK's ARIA awarded Fractile £5m (October 2024) to support its AI chip design work, signalling alignment with national strategic priorities in AI hardware.
- UK Government (Department for Science, Innovation and Technology)coreFractile's £100m UK expansion was publicly endorsed by the UK AI Minister Kanishka Narayan (February 2026) as part of the government's push to make Britain 'an AI maker, not just a taker'. Fractile is also implicitly positioned within the UK's £1.1bn AI Hardware Plan (June 2026), which earmarks £150m for domestic AI hardware.
Scale indicators11 records
Recent moves7 records
Expansion highlights7 records
Fractile competitors and assessment
Company assessmentDirect peers
- Cerebras Systems: Cerebras builds wafer-scale AI accelerators (WSE) used for both training and inference. While larger and more training-focused, Cerebras directly competes for the same frontier AI lab and hyperscaler inference workloads that Fractile targets.
- SambaNova Systems: SambaNova develops reconfigurable dataflow units (RDUs) for enterprise and government AI inference and training. It is a direct competitor in the high-throughput inference accelerator market and serves similar frontier AI lab and enterprise customers.
- Groq: Groq builds LPU (Language Processing Unit) inference accelerators using on-chip SRAM and a TSP architecture — the closest direct competitor to Fractile's in-memory compute, inference-focused approach. Both target frontier AI labs and hyperscalers with deterministic, low-latency inference silicon.
Broad incumbents
- Nvidia: Nvidia dominates AI compute with its GPU platform (H100, B100, upcoming Rubin) and CUDA software ecosystem. Fractile is explicitly positioning against Nvidia on inference cost-per-token, making Nvidia the incumbent benchmark Fractile must displace to capture share.
- Amazon (Trainium / Inferentia): AWS Trainium (training) and Inferentia (inference) are Amazon's custom AI silicon, used internally by Anthropic as part of its silicon diversification. AWS represents both a potential customer channel and a competing inference silicon alternative for frontier AI workloads.
- Google (TPU): Google's TPU is a custom AI accelerator used internally and offered via Google Cloud for both training and inference. Anthropic already uses Google TPUs as part of its silicon diversification, making TPU a competing alternative for the same frontier AI lab customers Fractile targets.
Emerging players
- Tenstorrent: Tenstorrent (led by Jim Keller) develops RISC-V-based AI accelerators for training and inference. It targets similar enterprise and hyperscaler customers with a different (CPU-style) architectural approach, representing an adjacent inference accelerator competitor.
- Lightmatter: Lightmatter builds photonic AI processors targeting inference workloads. While using a fundamentally different technology (silicon photonics vs. Fractile's electronic in-memory compute), it pursues the same memory-bandwidth-bottleneck thesis in AI data centers.
- Mythic: Mythic develops analog in-memory compute AI inference accelerators, directly addressing the same compute-near-memory architectural problem as Fractile. It is an emerging peer with a related but distinct (analog vs. digital) technical approach.
- Etched: Etched is building transformer-specific ASICs (Sohu chip) optimized for LLM inference, taking a different architectural approach (specialized ASIC vs. Fractile's in-memory compute) to the same inference throughput challenge. It competes for the same frontier AI lab design wins.
Market position
Strengths5 records
Weaknesses5 records
Competitive moat5 records
Key risks6 records
Key highlights7 records
Customer concentration
Fractile social profiles
Digital presenceFractile financial estimates
Financial estimateRevenue estimate
Valuation estimate
Fractile leadership team
Management profileNumber of profiles
Fractile funding detail
Funding detailFunding overview
Funding rounds6 records
Investors32 records
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
Fractile M&A and investment
M&A and investmentM&A
Investments
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about Fractile
What does Fractile do?
Fractile is building next-generation AI inference processors using a custom SRAM-based in-memory compute architecture that physically interleaves memory and compute on the same die, eliminating the off-chip DRAM bottleneck of GPU-based inference. The company takes a full-stack approach spanning transistor-level circuit design through cloud inference server logic, and ships integrated inference systems combining the chip with a proprietary software stack designed to compete with Nvidia CUDA. First commercial chips are targeted for 2027, with stated targets of up to 25x faster inference at 1/10th the cost of existing GPU systems and ~1,200 tokens/sec throughput.
Is Fractile a public or private company?
Fractile is a private company. It is currently operating.
When was Fractile founded?
Fractile was founded in 2022. It employs 101 to 250 people.
Where is Fractile based?
Fractile is headquartered in Newbury, United Kingdom, in the Europe region.
How does Fractile make money?
Two revenue lines are on record. AI inference chip and system sales are the primary driver. The others are pre-commercial strategic supply agreements.
Who are Fractile's main competitors?
Direct peers on record are Cerebras Systems, SambaNova Systems and Groq. Broad incumbents are Nvidia, Amazon (Trainium / Inferentia) and Google (TPU). Emerging players are Tenstorrent, Lightmatter, Mythic and Etched.
Does Fractile have an API?
No public API is recorded for Fractile.
What industry is Fractile in?
Its primary akta.pro industry code is HDAHABAC, Compute & Acceleration ICs (CPU/GPU/AI/FPGA/ASIC), with a secondary code of HDAAAAAC, AI Inference Hardware & Edge Accelerators. Its NAICS code is 5415 and its SIC code is 3571.