What is Microsoft’s new AI supercomputer in Loughton, Essex?

The creation of a powerful new supercomputer is set to put Essex at the heart of the UK’s artificial intelligence (AI) revolution.

Work had already been ongoing to build a £2bn data centre in Loughton and open it by 2026.

But now, Microsoft announced it would use some of a $30bn (£22bn) investment into UK technology to fund a supercomputer at the site.

But what exactly is a supercomputer – and what could it power?

The Loughton site is due to become the UK’s largest AI data centre, creating up to 250 permanent jobs when it opens in 2026.

It is hoped the four-acre (1.6ha) facility will benefit services such as healthcare and defence.

News that Microsoft is now backing the Loughton campus dramatically upscales what its output could be.

Nscale pledged £2bn to create the centre, but it will now have a share of the £22bn pot brought by the tech giant.

Nscale, Microsoft and NVIDIA, which will be building the AI chips for the centre, now say they can build what they have dubbed a supercomputer in Loughton.

This will see thousands of devices working together 24/7 to form one large computing brain.

Nscale says the data centre will deliver 50MW of AI capacity, but is designed to grow. In the future, it could be upgraded to handle up to 90MW of power, nearly doubling its capacity.

“For perspective, even the world’s fastest supercomputer in the US draws on the order of around 30 MW at peak, which is considered huge,” says Dominic Davies, chief of AI legal-tech firm Lightbringer.

“So 90 MW is nearly three times that level, putting Loughton’s facility in the global top tier for AI infrastructure.”

These high-performance AI systems will train models like ChatGPT and power real-time services such as voice assistants.

At the start, the facility will contain 23,040 graphics processing units (GPUs) built by NVIDIA.

These are potent chips used to train and run AI models – the more GPUs, the more AI work the site can do.

“It will make the Essex supercomputer one of the most powerful AI computing centres ever built,” says Mr Davies.

Some people champion the idea of using AI for public services, arguing it improves efficiency by automating tasks and using data to make almost instant decisions.

However, others say decisions made by the technology lack transparency and accountability.

There are also logistical considerations. These AI centres need huge amounts of electricity, both to run the GPUs and keep them cool.

Ensuring they do not overheat is essential to maintain performance and avoid damage.

But that is tricky in a country where energy prices are higher than usual and there is pressure to use renewable sources.

Google’s greenhouse gas emissions in 2023 were 48% higher than in 2019, something it put down to the increasing levels of energy needed by its AI centres.

Speaking to the BBC previously, AI expert Dr Haider Raza said “awful” electricity costs could also stunt growth.

“We have to make data centres more efficient. This point is very, very important,” the University of Essex lecturer added.

The government said it was “exploring bold, clean energy solutions” to meet its AI ambitions.

AI facilities also need to be in places with strong internet connectivity so data transfer is efficient.

There is also the challenge of making sure these sites, which often process sensitive data, have robust cybersecurity in place.

On Tuesday, Prime Minister Sir Keir Starmer said the UK was on its way to becoming a world leader in AI.

“Meaning more jobs and investment, more money in people’s pockets, and transformed public services,” he added.