The UK’s billion-dollar chip bet: Nigel Toon on making Graphcore bigger than Arm

So what is it about its technology that has investors buzzing?

In recent years, the technology of artificial intelligence has moved forward rapidly, bringing driverless cars, automated drug discovery and a host of other innovations close to reality. All of them have been powered by AI algorithms, which rely on a new breed of advanced chips to work.

The current generation of Graphics Processing Units processors, which have enjoyed a boom in the past five years, are best tuned for crunching large volumes of data - but they are largely one-trick-ponies.

They struggle when faced with lots of variable, unpredictable tasks - essentially complex decision making - something that is essential to recent advances in AI such as autonomous vehicles.

That is where Graphcore’s expertise comes in. The company has excelled at building flexible, yet efficient and powerful chip designs, says Toon.

“What we are doing is allowing people to take the next steps in artificial intelligence,” Toon says. “The arithmetic needed to do [that] is very different.”

According to Toon, current processors lack the ability to perform well at speed. This need for power mixed with efficiency, he believes, is Graphcore’s sweet spot.

At 55, Toon is already a seasoned technology entrepreneur, returning for his next swing at a multi-million pound business. Born in Scotland and an engineering graduate from Edinburgh’s Heriot-Watt university, Toon cut his teeth at US tech company Altera in the 1990s, before founding Icera, along with Knowles, in 2002.

Icera, which made modem chips, went on to be snapped up by US giant Nvidia in 2011 for $370m. A number of Icera alumni (not to mention others from Arm, Nvidia and further afield) now make up the ranks at a fledgling Graphcore.

“Our history is building silicon chips,” says Knowles, Toon’s co-founder. “Nigel and I were in a pub not long after we finished selling Icera, and we thought: ‘Well, shall we do another one then?’”

The seed of Graphcore was incubated in Xmos, another Bristol company, with the help of the founder of Amadeus Capital Partners, Hermann Hauser, who some in the industry describe as the “grandfather” of semiconductors in the UK. Hauser, an early angel investor in Graphcore, is perhaps best known as a co-founder of Arm - considered the UK’s crown jewel of tech.

In 2016, the same year that Arm was being bought by Japan’s Softbank for £24bn, Graphcore launched in full. The next two years in the chip industry was characterised by deal making and consolidation.

“[The question was] who was more about who was merging with whom, rather than innovation,” says Toon

It was at this point that Graphcore decided to emerge from “stealth mode” and go public with its plans. Its first product is an “Intelligence Processing Unit”, or IPU. It is built quite differently to current processors after Graphcore redesigned the structure of the connections on the chip to make them ultra-efficient for AI tasks. Graphcore says its new approach could deliver up to performance that is 100 times better than standard chips. The product, called the “Colossus” after one of the first computers developed during the Second World War at Bletchley Park, has already been rolled out to data centres and is ready for sale. It could soon be deployed for other technologies like driverless cars.

It may face some challengers, however.

“Graphcore has been phenomenally successful at raising capital,” says Karl Freund, senior analyst at Moor Insights & Strategy. “[But] there is lots of competition - around 40 start-ups.” He also points out that rival Nvidia currently controls around 99pc of the AI chip market.

But according to Toon, the market has only just realised what it is missing in its AI technology, and their product has the potential to leap-frog what is currently being used.

“People have only now started talking about this technology, four or five years after we got started,” Toon says. “No one fully has their head around it, other than Nvidia and maybe Intel. There is an opportunity for a rapid change.

“The market is big enough that we can create a massive return for our investors, who think this company will be worth tens of billions.”

Toon seems to be a serial entrepreneur or, as he puts it, perhaps he is “congenitally unable to work for anybody else”.

“It’s why entrepreneurs do what they do. We can build this company, so we do build this company.”

However, Toon is eyeing something more than just another tech “unicorn”, or billion dollar company, in Graphcore.

With the sale of Arm, the UK lacks a real tech leader, at least in the hardware space, and has no more than a handful of listed tech companies on the London Stock Exchange.

Toon says the UK is “probably too harsh” on itself and that there is a “good amount of depth” of venture capital. But public markets remain a challenge, a hurdle Graphcore may have to face in two or three years time, unless it takes more funding. (Toon says the start-up still has $250m in cash, having spent around $60m so far).

Building a tech icon in the UK may be tricky. “If we could [list] in London we would love to do it in London,” Toon says. But, he adds, the UK market appears less enamoured by high-growth, low earnings tech ventures and less focused on the ten-year picture.

“The LSE has almost no tech companies at all. If there is not an exit here to becoming a public company, naturally they will list in the US.”

But for now - even as the start-up expands globally (with plans to triple headcount and already sporting offices in London, Taiwan, Beijing, Palo Alto and Oslo) Graphcore’s heart remains in Bristol.