UK-based AI chip designer Graphcore was once seen as a potential rival to Nvidia and AMD but fell on hard times after failing to capitalise on the AI boom.
In February 2024, we wrote that Graphcore desperately needed to raise significant funding by May if it wanted to survive, and that several potential buyers were circling the troubled company. These included Arm, Japanese tech conglomerate SoftBank (which has a majority stake in Arm), and OpenAI.
SoftBank was always the most likely suitor and has now been confirmed as the new owner of Graphcore, having paid between $400 million and $500 million for the company. The exact price has not been confirmed.
But what's next?
“Society is seizing the opportunities offered by basic models, generative AI applications, and new approaches to scientific discovery,” said Vikas J. Parekh, Managing Partner at SoftBank Investment Advisers. “Next-generation semiconductors and computing systems are essential on the path to general AI, and we are pleased to collaborate with Graphcore on this mission.”
Graphcore will become a wholly owned subsidiary of SoftBank (which will continue to operate under the Graphcore name) and will likely maintain closer ties to Arm. The company’s headquarters will remain in Bristol, with offices in Cambridge, London, Gdansk and Hsinchu (it previously closed offices in Norway, Japan and South Korea). Co-founder and CEO Nigel Toon will remain in charge of Graphcore.
“This is a great endorsement of our team and their ability to build truly transformative AI technologies at scale, and a great result for our company,” Toon said. “The demand for AI computing is huge and growing. There is much work to be done to improve efficiency, resilience, and computational power to unleash the full potential of AI. In SoftBank, we have a partner that can enable the Graphcore team to redefine the AI technology landscape.”
What the future holds for Graphcore and how SoftBank will profit from its latest acquisition remains to be seen. Licensing its intellectual property to Arm seems like the obvious first step.