The British chip startup that raised $710 million to take on Nvidia — and learned the hardest lesson in tech: being technically better doesn't matter if the other guy owns the ecosystem. Graphcore's IPUs are genuinely innovative, but Nvidia's CUDA moat and relentless execution have made the AI chip market a near-monopoly. A cautionary tale about the difference between a great chip and a great business.
Founded
2016
HQ
Bristol, UK
Total Raised
$710M
Founder
Nigel Toon, Simon Knowles
Status
Private — facing significant challenges. Reported to be exploring strategic options including a potential sale. Revenue has been well below what $710M in funding would suggest.
Website
www.graphcore.aiTHE ORIGIN STORY
Nigel Toon (CEO) and Simon Knowles (CTO) had decades of experience in semiconductor design — both had previously co-founded ICERA, a mobile chip company acquired by Nvidia. In 2016, they saw that AI computation was hitting the limits of GPU architecture and believed a clean-sheet chip design for AI could win.
They raised massive funding based on this thesis — $710 million total — and built the Colossus IPU, which they claimed was optimized for the irregular computation patterns that characterize machine learning.
WHAT THEY ACTUALLY DO
Designs and manufactures Intelligence Processing Units (IPUs) — specialized AI chips designed for machine learning workloads. Unlike Nvidia's GPUs (which were adapted from graphics), Graphcore's IPUs were purpose-built from scratch for AI computation.
Revenue comes from chip sales and cloud partnerships where customers can access IPU compute. The bet: purpose-built AI chips will eventually outperform adapted GPUs for specific AI workloads.
THE PRODUCTS
Colossus MK2 IPU (Intelligence Processing Unit), BOW IPU (latest generation with wafer-on-wafer technology), IPU-POD systems (scalable multi-IPU systems for data centers), Poplar SDK (software development kit for IPU programming), and IPU Cloud (cloud-based access to IPU compute through partnerships with Microsoft and others).
HOW THEY GREW
Differentiation through specialization. Graphcore has positioned IPUs for specific workloads where they outperform GPUs — particularly graph neural networks, recommendation systems, and sparse computation.
Cloud partnerships with Microsoft Azure and others give developers access without buying hardware. The company has also explored AI applications in healthcare and scientific computing where its architecture has genuine advantages.
THE HARD PART
Nvidia. Just...
Nvidia. Graphcore raised $710 million to compete with a company that has a $3 trillion market cap, a developer ecosystem of millions, and CUDA — the most entrenched software platform in AI.
Every time Graphcore convinced a potential customer that IPUs were technically superior for certain workloads, Nvidia would cut prices or release a new GPU that closed the gap. The AI chip market became a winner-take-most dynamic, and Nvidia was winning.
MONEY TRAIL
Series A
2016 · Led by Atomico, Bosch, Samsung
$30M raised
Series C
2018 · Led by BMW iVentures, Dell, Microsoft
$200M raised
Series E
2020 · Led by Ontario Teachers, Baillie Gifford, Fidelity
$222M raised
WHO BACKED THEM
Sequoia Capital, Atomico, Baillie Gifford, BMW iVentures, Bosch, Dell Technologies, Microsoft, Samsung, and Draper Esprit have backed Graphcore.
Related Profiles
Head-to-Head
Compare Graphcore vs another company.