The Government’s plan to buy £100m of AI chips may be too little, too late

On the face of it, Rishi Sunak’s plan to spend £100mof taxpayer money to buy up to 5,000 graphics processing units (GPUs) – the computer chips powering the current AI revolution – sounds like a smart move.

First reported by The Telegraph, UK Research and Innovation (UKRI) is reportedly at the advanced stages of negotiating an order from AI chip makers, king among them Nvidia, which is expected to report bumper financial results this week.

The order is designed to plug a hole in the UK’s AI capabilities: a Government review this year raised concerns that there were fewer than 1,000 of the types of chips needed for high-spec AI within the country and accessible to researchers.

But saying you’ll buy chips and actually getting them are two drastically different things.

Since the release of ChatGPT in October 2022, the race for AI supremacy has become so heated that there is a global shortage of GPUs, which are the specialised chips needed to power high-level AI systems.

Tesla CEO and Twitter owner, Elon Musk, told a conference earlier this year that “GPUs are at this point considerably harder to get than drugs.” Mr Musk later said: “It seems like everyone and their dog is buying GPUs.”

And Nvidia, the seller of the crucial chips, is reportedly so swamped by demand that it’s able to pick and choose who to prioritise based on their history of ordering from the company.

One widely shared blog, believed to be written by a tech industry insider, claimed that there will be a shortage of Nvidia GPUs for those ordering hundreds or thousands – the range at which the UK Government is intending on buying into the chip sector – for at least the remainder of the year, and likely long into 2024.

The scale of the UK’s ambition also appears to be lower than AI’s boosters may hope. Planning to order 5,000 chips may sound impressive out of context, but in the grand scheme of AI chip speculation, it’s small fry.

Earlier this month China forked out $5bn to buy 100,000 GPUs of a similar stature to those the UK is targeting from Nvidia, the market’s leading seller. Saudi Arabia recently bought 3,000 chips of its own. And the latest version of ChatGPT was reputedly trained on around 25,000 GPUs.

“The Government has rightly recognised the importance of processing power but the low ambition of this move just isn’t appropriate to the scale of the challenge,” said Sam Sharps of the Tony Blair Institute for Global Change, which called for the Government to procure 30,000 GPUs in June. “Anything less looks too scant, even before we start to think about how they will be distributed and exploited,” Mr Sharps said.

His colleague at the Tony Blair Institute, Benedict Macon-Cooney, said: “This is nowhere near the scale necessary. GPUs are a critical strategic technology and if the UK has serious ambitions to be an AI power it needs to up its ambition.”

Azeem Azhar, founder of the Exponential View newsletter, believes that the investment is comparatively small, given the increasing computing demands of AI systems. However, he said that cutting the computing cloth to fit the UK’s budget may have an inadvertent benefit. “Perhaps a limited budget would create the right incentives for researchers to develop much more computer-efficient approaches to AI.”

Mr Azhar added money might be better spent leasing capacity to GPUs through commercial cloud data centres. “GPUs depreciate and get outdated very quickly,” he said.

The unnamed tech insider’s blog estimates that big tech companies may already have committed to buy around half a million GPUs, which would push the UK to the back of the line until those orders were fulfilled.

The UK may be able to skip part of the queue given it is positioning itself as the global hub for AI regulation, with a global AI safety summit taking place in the UK in November. Chipmakers will want to stay on the Government’s right side. But it’s unlikely that such a comparatively small order would jump to the front of the line.

UKRI, who are reportedly likely to place the order, did not respond to a request for comment. A Government spokesperson told The Telegraph: “We are committed to supporting a thriving environment for compute in the UK which maintains our position as a global leader across science, innovation and technology.”

Darren Jones, Labour MP and chair of the Commons Business and Trade Select Committee, has sent a letter to Michelle Donelan, Science and Technology Secretary, asking for further clarity on Government AI expenditure. “It’s important that the Government knows what it’s building and why,” Mr Jones told the i. “Based on current information, I don’t think ministers have the faintest idea what they’re doing.”

A Department for Science, Innovation and Technology spokesperson said: “We are committed to supporting a thriving compute environment which maintains the UK’s position as a leader across science, innovation and technology. The funding for the AI Research Resource is part of our £900 million compute investment, as announced in the Spring Budget and which will be delivered by UKRI.

“No decisions have been taken on who will provide hardware for the Resource, and further announcements will follow in due course.”