Trump is taking AI seriously – why aren’t we?

AI is an expensive enterprise, one that is as much about raw materials and energy as it is clever algorithms. You need to get hold of chips, stick them in a data centre somewhere and ask them to multiply matrices until the sand starts to think. This is the basic reality at the heart of any attempt for middle powers like the UK to build independent AI capacity. You need cash; you need energy; you need know-how – in that order. 

Despite some impressive recent models from Chinese labs, the United States currently leads AI development. That’s because it has deep capital markets willing to spend big on chips to train and serve models, which allows individual companies to spend more on compute than the whole of the UK. The US also has Nvidia, which builds the chips that keep the AI show on the road.

In the run up to yesterday’s ‘AI Action Day’ in America, which saw President Trump introduce measures to help US firms build and power compute clusters, OpenAI said it was developing 4.5 gigawatts of additional data centre capacity domestically. According to the group, that brings the total to over 5GW of Stargate AI data centre capacity under development.

Let’s compare this to a similar announcement in the UK. The Government talked up a new Memorandum of Understanding (MoU) with OpenAI, in which the Department for Science, Innovation and Technology (DSIT) alluded to ‘driving investment in infrastructure’. Of course, the MoU itself actually doesn’t mention a single pound worth of investment from the firm; instead, it simply says ‘OpenAI will work with DSIT to identify potential routes to deliver the infrastructure priorities laid out in the AI Opportunities Action Plan’.

Okay, well what about the AI Opportunities Action Plan? The Government made some noise when it accepted its recommendation to increase compute by a factor of 20 over the next five years. That sounds like a lot, but a quick back of the envelope calculation suggests that would still leave the UK with well under 4% of the total raw horsepower the American public and private sector already has on the books. Even the Compute Roadmap, which looks impressive on first blush, talks up total investment that is about half of what Microsoft plans to spend independently in 2025.

The same document calls for an additional 6GW of UK power generation capacity by 2030 to feed the models – a figure roughly in the same ballpark as that touted in OpenAI’s new plan for its American clusters. Even if the UK manages to deliver on these goals, I struggle to see why anyone would build compute in the country given the eye-watering price of energy. Maybe you can use some behind the meter solutions to bring down costs, but it is simply not possible to use these methods alone to deliver the full 6GW uplift. 

Westminster’s basic problem is that it doesn’t have the core ingredients needed for building a competitive AI industry. We have plenty of the best AI researchers in the world, but what we don’t have are deep capital markets or cheap energy to build a competitive frontier model-maker. France has Mistral, Japan has Sakana and even Singapore has Sapient Intelligence. These firms aren’t at the bleeding edge to be sure, but they are closer than any UK outfit. In practice, that means whatever capacity we do build will be used for running either proprietary or open source models (both of which are likely to be American). Some people question how much this matters. They say open models might catch the frontier sooner than we think, and from a sovereignty perspective you may only need a handful of critical uses like defence or healthcare.

The problem with this line of thinking is that it assumes a world where AI is powerful enough to automate the most essential areas of an economy, but not powerful enough to do much more than that. At the point at which compute can be used to automate the stuff that matters, it can be used to automate everything else. That means we should expect the countries that have the greatest reserves of compute to be those that get rich in the age of Turing.

You can sell your compute abroad or you can use it at home, in the same way that oil-producing countries do with black gold. Even if you can automate today’s economy with ease, you’ll quickly find ways to put the surplus to work. That’s because such a scenario is premised on the arrival of extremely powerful models that can be used for anything you can imagine (and plenty of things you can’t). 

This might all sound rather sci-fi, but this future is actually more conservative than the one currently being forecasted by the Secretary of State for Technology. On a recent podcast, Peter Kyle said: ‘I think by the end of this parliament we’re going to be knocking on artificial general intelligence.’ That puts the arrival of a system that would turn compute into the world’s most valuable currency about four years away. 

Of course, the Government isn’t acting like AGI is four years away. If it was, it would be looking to increase compute by 100x from the current floor. It would pull out all the stops to make energy cheaper. And it would offer US firms tax breaks to build compute capacity at home. This would all be small beer, but it might take us from ‘bad’ to ‘somewhat bad’ for the UK’s size relative to Uncle Sam.

 – the best pieces from CapX and across the web.