--- title: "The Dyson spirit is gripping Britain’s microchip industry" type: "News" locale: "en" url: "https://longbridge.com/en/news/286752688.md" description: "Britain's semiconductor industry is experiencing a surge in innovation, with significant investments in AI chip startups like Fractile, which aims to revolutionize chip design using analogue technology. This approach could drastically reduce costs and improve efficiency compared to traditional digital chips. Other startups, such as Olix and Pragmatic, are also exploring radical solutions to enhance computing capabilities. Despite these advancements, UK firms face challenges in securing funding compared to their US counterparts." datetime: "2026-05-18T10:09:14.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/286752688.md) - [en](https://longbridge.com/en/news/286752688.md) - [zh-HK](https://longbridge.com/zh-HK/news/286752688.md) --- # The Dyson spirit is gripping Britain’s microchip industry Why do what someone else is doing but only slightly better? While incremental improvement has been good enough for China, Britain’s semiconductor industry is showing the merit of making big, dramatic breaks with convention. With substantial new capital investments, our chips sector is finally enjoying its hour in the sun. In each case, it’s because scientists and engineers made some quite daring decisions. Our technological audacity should terrify the central planners in Beijing. Recently, British AI chip startup Fractile announced a $220m (£165m) funding round on top of the $100m it received three months ago. The company’s founder, Walter Goodwin, is gambling on replacing one conventional part of a computer chip with an analogue replacement. Goodwin thinks his chip can perform a key AI job 100 times faster than Nvidia’s hot and expensive graphics processing units and come in at a tenth of the cost. Nato’s venture capital fund is among Fractile’s investors. But wait – analogue? Yes, you read that correctly. Computing doesn’t have to be digital and as recently as the 1970s, research facilities at Lockheed, Nasa and Westinghouse used analogue computers because they performed some kinds of calculations much more efficiently, such as fluid dynamics. I once visited a distinguished former IBM researcher at his home in Amsterdam where he was restoring an analogue computer in his living room. It resembled a giant, steampunk cathedral organ. In his opinion, some very challenging calculations – such as those required for interplanetary space travel – would not be possible without analogue computing. So, I joked, the little green men in flying saucers would be using analogue computers? “Of course,” came the deadpan reply. Fractile is addressing an urgent problem. By the end of the year, 80pc of the cost of running an AI model will not be the upfront training – as it was a year ago – but the number crunching that it does as you engage with it, called “test time training” or inference training. A bottleneck is created in this system by relying on a very small unit on the digital chip. It fetches a number from its memory, performs a multiplication then adds it to a running total. This is being performed not billions but trillions of times a second in the largest AI models. Once the arithmetic is complete, the AI will come up with an answer to your question based on probabilities. “Fetching each number from memory is the bottleneck. It’s very inefficient and takes power and time,” says Andy Sellars, of Southampton University. This journey does not need to be made with Fractile’s analogue matrix replacement. Instead, it allows the computation to take place in the “memory” itself, simplifying the job greatly. “A floating-point multiply operation involves 60,000 transistors,” says Rupert Baines, a serial chief and entrepreneur who co-authored the national semiconductor strategy with Sellars. “In the analogue world, you can do it with 30.” That’s the kind of dramatic difference an analogue component can make, if Fractile can pull this off. Fractile needed “to radically reinvent the hardware that we run our frontier AI models on”, Goodwin wrote last week. Olix is another British start-up trying to solve the inference bottleneck. It’s using photonics, or computing using light waves, to do so. Others have tried using photonics in the past, acknowledges James Dacombe, the Olix founder, in his Compute Manifesto. But in his opinion, they didn’t go far enough. His company won $250m of investment in February. Yet another example of radical innovation can be found at semiconductor company Pragmatic. Its flexible-film chips can be printed on paper and we can expect to see orders in the many millions as tags appear on food packaging. Pragmatic is manufacturing its parts, too, opening the biggest silicon foundry in the UK two years ago. In effect, these innovators are broadly comparable to how Sir James Dyson approaches a problem. Rather than incremental improvements, the entrepreneur re-engineers products from the ground up. His cyclonic vacuum cleaner didn’t seek to offer a better dust bag but removed the dust bag altogether. Our semiconductor industry is now firing on all cylinders. Such quantum leaps in deep tech don’t appear overnight, however, and are only possible because of the foundational scientific research coming out of our universities. Britain’s lead in compound semiconductors – which are specialist components that transmit power and light and are widely used in smartphones and electric cars – was a 20-year national research strategy. It was a bet that alternatives to silicon could be useful and create new product categories. British pragmatism then turns ideas into real products. Nevertheless, our firms envy the generosity and risk-taking of American venture capitalists. One industry veteran despairs: “UK and European semiconductor companies have to make do with one tenth of money available in US funding rounds. Maybe we are two times smarter but are we 10 times smarter than the Americans?” Alas, our money men are not aligned with the national interest and don’t invest big for the long term. In fact, they can’t wait to flog off our most successful companies fast enough. “Look at DeepMind,” says Baines. “Sold for $500m \[to Google in 2014\] when it should have been worth $500bn and the most valuable company in Europe.” Perhaps the day when we retain our world champions is the day we can say Britain is back. ### Related Stocks - [SMH.UK](https://longbridge.com/en/quote/SMH.UK.md) - [NVDA.US](https://longbridge.com/en/quote/NVDA.US.md) - [LMT.US](https://longbridge.com/en/quote/LMT.US.md) - [IBM.US](https://longbridge.com/en/quote/IBM.US.md) - [NVD.DE](https://longbridge.com/en/quote/NVD.DE.md) ## Related News & Research - [EXCLUSIVE-At Samsung, the global AI boom spurred a looming strike and deep divisions](https://longbridge.com/en/news/286544840.md) - [04:29 ETScioSense launches UFC23 ultrasonic flow converter for high-precision, ultra-low-power smart metering](https://longbridge.com/en/news/286383429.md) - [Thanks for the memory: Why AI-chip stocks of all types are going supernova](https://longbridge.com/en/news/286074182.md) - [Cerebras Systems stock soars 68% in blockbuster IPO: What investors should know](https://longbridge.com/en/news/286493984.md) - [Analysts see Nvidia poised for record Q1 and possible guidance boost](https://longbridge.com/en/news/286654507.md)