Softbank gives UK AI chip maker Graphcore a soft landing
Japanese tech investor has a taste for British chips and bought up the storied silicon designer ARM in 2016
Softbank has scooped up another chunk of the UK’s indigenous silicon industry in the shape of struggling AI chip specialist Graphcore.
Cofounder Nigel Toon confirmed in a statement that the firm had been acquired by Softbank and will become a wholly owned subsidiary operating under the Graphcore name.
It’s reported that Softbank paid less than the $700m investors have allegedly poured into the Bristol-based startup to date. However, it's most recent accounts showed revenue of $2.7m in the year to December 31, 2022, down from $5m the previous year, with operating costs of $206.8m.
Softbank acquired the storied British silicon designer ARM back in 2016. It IPO’d the firm last year but remains its biggest shareholder by far.
Graphcore set up shop in 2016, before the world fully woke up to the coming AI revolution. Nvidia’s domination of the AI chip market – and its subsequent elevation to one of the world’s most valuable companies - was sparked by researchers’ realization its highly parallelized GPU tech could be adapted for model training. It’s fair to say Nvidia has capitalized on that insight, turning itself into one of the world's most valuable companies.
Graphcore, by contrast, set out to build an AI targeted architecture from the outset. Its Mk2 Colossus Intelligence Processing Units, released in 2020, packed 1472 independent cores, each running six threads, and sharing 900 MB of onboard memory.
Its processors are available as processor cards, blade units, and integrated pods, which can be scaled up to a 1024 IPU device. It also offers access to its technology as a cloud service.
It has snagged high profile installations, including the University of Edinburgh’s supercomputing centre, the UK’s Hartree Centre, and the US’ Argonne National Lab.
And there has certainly been evidence that Graphcore’s technology could outperform Nvidia GPUs on at least some AI tasks.
However, it has yet to turn that vast potential power into revenue, while simultaneously struggling with the capital requirements of continuing to develop, productize and scale out its technology.
“This is a tremendous endorsement of our team and their ability to build truly transformative AI technologies at scale, as well as a great outcome for our company,” said Graphcore co-founder and CEO Nigel Toon. “
He said that the unlocking the full potential of AI still required much work to improve efficiency, resilience, and computational power to unlock the full potential of AI. “In SoftBank, we have a partner that can enable the Graphcore team to redefine the landscape for AI technology.”