The British government is scrambling to find extra electricity capacity in the area where it plans to house its new artificial intelligence lab.
It is believed that the Prime Minister has become personally involved in efforts to secure additional grid capacity for a Bristol supercomputing laboratory, which is leading candidate for a £100m taxpayer-funded “AI Research Resource”.
The move comes amid concerns that Mr Sunak’s efforts to establish Britain as a hub for AI could be undermined by a lack in power and the long delays involved with accessing additional grid capacity.
The Prime Minister will host an international summit on AI in November, and he hopes that Britain will play a major role in setting global standards and rules around AI.
This effort will be aided by a powerful system that can run and evaluate AI systems.
Bristol University’s Isambard 3 supercomputer facility, which is one of only a few “Tier 2”, UK-based supercomputers, has been identified as the likely host for the project.
Currently, the site has a 5 megawatt electricity supply. This capacity will likely be maxed out by running 5,000 high-performance GPUs that power AI systems.
But today’s top models run on much more. Chinese tech giants are said to have ordered around 100,000 chips. The latest version of ChatGPT, however, is believed to have been trained on up to 25,000.
Next year, the site could double its electricity supply from 5 megawatts up to 10 megawatts. This would be enough to power around 20,000 households. However, it’s possible that additional upgrades may be required to match the site’s ambitious goals.
It is believed that the Prime Minister became personally involved in this project to try to push for upgrades.
Long delays in connecting the grid may stymie efforts. Many projects, from solar panel farms to housing developments, have complained about the long waits for grid connections.
Virtus, an operator of data centres, informed the energy regulator Ofgem, in June, that a lack electricity supplies held back investment, and put Britain’s AI aspirations at risk. Ofgem said that the high number of delays in connection meant “urgent reform” was needed.
This month, a final decision is expected on the location for the AI Research Resource. It could be operational by summer next year.
Hewlett Packard Enterprise, the Silicon Valley IT giant is working on its bid with Bristol University, and has had discussions with Nvidia AMD Intel – – the top three GPU manufacturers.
Simon McIntosh Smith, a Bristol-based professor of high performance computing, refused to comment on the potential plans for a AI Research Resource. HPE declined to comment.
A spokesperson for the Department of Science, Innovation and Technology said that “No decision has been made on where Resource will be located.” Reports to the contrary are completely false.
We are committed to a thriving computing environment that maintains the UK’s leadership in science, technology and innovation. The AI Research Resource funding is part of the PS900 million in compute investment announced in the spring budget, which will be delivered through UKRI.
A spokesperson for Ofgem said: “We recognize the challenges surrounding grid connection waiting time and are working with the industry and the government to resolve the issue.
We launched a major evaluation of this issue earlier in the year and we are now considering all responses. In the fall, we will be presenting a joint action plan with DESNZ (The Department of Energy Security and Net Zero).
It was reported last month that UK Research and Innovation and DSIT officials were preparing to purchase up to 5,000 GPUs, as countries and technology giants sought to ensure a steady supply of the highly-requested chips.