The Rt Hon Rishi Sunak MP
Prime Minister of the United Kingdom
Dear Prime Minister,
In your Spring Budget Statement, you committed £900m of funding to support the delivery of a public Exascale and AI compute capability for the UK, as recommended by the Independent Review of The Future of Compute.
The Government’s backing for this initiative is most welcome, given its potential to accelerate scientific research and other contributors to economic growth and the public good. As the report observes, if our country does not make progress in this area: “the UK’s position will continue to decline as other countries press ahead with plans to bolster their compute infrastructure.”
Having welcomed your broad support for UK Exascale and AI compute, we are calling for a similar commitment to supporting those UK homegrown technology companies that are well placed to help deliver such a capability.
Amongst these is Graphcore, the Bristol-based company that I lead, and which designs and manufactures specialist AI compute systems, powered by our Intelligence Processing Unit (IPU) AI accelerator, one of the most advanced semiconductors ever made.
Our IPUs are already being used at numerous publicly funded compute facilities overseas. Graphcore systems are deployed across the US Government’s network of research laboratories and supercomputer facilities, in Germany’s National Supercomputing Centre, the Karlsruhe Institute of Technology and at Korea’s Electronics and Telecommunications Research Centre, as well as many other publicly funded facilities, enterprise companies and startups globally.
However, we have not yet seen a signal that the British Government intends to include British made technology in this country’s planned AI compute capability. By backing UK suppliers, the government not only develops a valuable AI compute ecosystem, but also nurtures future UK-based, highly skilled teams of engineers with the knowledge and confidence to build the next generation of digital hardware. This was the case with ARM, which has indirectly been responsible for a series of digital hardware startups in the UK, as ARM-trained engineers founded other businesses.
We are concerned that unless a significant portion of the budget is explicitly earmarked for UK-based suppliers, this funding commitment will quickly be consumed by digital giants like US-based chipmaker Nvidia.
Too often we have seen British-made innovation leading the world, only to be edged-out or bought-out by overseas rivals. By directing at least some of the UK taxpayer’s investment towards world class UK businesses, the Government has the power to grow our knowledge economy and ensure that companies like Graphcore can provide jobs for semiconductor engineers, programmers, AI engineers and other highly skilled professionals. It would mean that graduates in these fields can consider roles in Bristol, Cambridge, London and other parts of the country, instead of San Francisco and Santa Clara.
We must also recognize that with their advantage of dominant market share, companies like Nvidia have been supplying their GPUs at low cost to incentivise their use amongst UK researchers in a way which shapes the habits of AI practitioners and researchers and excludes other hardware suppliers. This lack of competition does not benefit the end-user. In the few short years that GPU processors have been employed for AI computing, popular tools and techniques have inclined towards their strengths and away from their weaknesses.
Forward-looking research institutions like the US Government’s Pacific Northwest National Lab, Sandia Labs and Argonne Labs have recognised that we need to adjust this approach and pursue the potential of made-for-AI processors to open-up new avenues of AI research beyond the capabilities of GPUs. These institutions have promoted diversity of supply in AI processors, investing accordingly and, ensuring that hardware such as Graphcore’s IPUs make up part of their compute offering. The UK’s future compute infrastructure needs to be similarly diverse; empowering researchers by giving them access to new types of compute rather than limiting them to legacy architectures.
With these points in mind, we at Graphcore would like to make the following suggestions, as you consider how to deliver the UK’s next generation compute infrastructure:
- Specify that a large percentage of the compute system budget be directed to home-grown UK technology companies.
- Maintain technological diversity within the UK Exascale and AI compute system, ensuring that, in addition to CPUs and GPUs, our UK compute infrastructure includes home-grown AI-specific accelerators.
- Resist the lure of large foreign technology companies who are trying to edge-out our UK companies.
Finally, we appreciate that the government wishes to move quickly on its compute commitment, especially for AI, to ensure that systems are in place and being used by UK innovators as soon as possible. Graphcore has offered to reserve – for this project – up to 3,000 UK-developed IPU AI processors, located in a South Wales based datacenter that can serve the whole country. These systems are installed and ready to go.
We are keen to support the UK scientific and AI computing efforts and believe that by investing in homegrown AI hardware the UK can build a complete ecosystem that will ensure the UK fulfils on its science and technology superpower ambitions.
CEO & Co-Founder, Graphcore
Letter Cc'd to:
Rt Hon Jeremy Hunt MP, Chancellor of the Exchequer
Rt Hon Michelle Donelan MP, Secretary of State for Science, Innovation and Technology