SoftBank has put more than $450 million into Graphcore, the British AI chip company that has spent the better part of a decade trying to build a credible alternative to Nvidia’s dominance in machine intelligence hardware.
What Graphcore actually does
Founded in 2016, Graphcore designs what it calls Intelligence Processing Units, or IPUs. These are chips built from the ground up for machine learning workloads, as opposed to Nvidia’s GPUs, which were originally designed for rendering video game graphics before the AI world discovered they were really good at matrix math.
Graphcore had raised approximately $682 million in total funding prior to this deal, attracting heavyweight backers including Sequoia Capital and Microsoft. The company’s technical chops were never really in question. Its business model, however, was a different story.
Revenue never matched the ambition. The company reportedly struggled to convert technical demonstrations into meaningful commercial traction, a problem that plagued several would-be Nvidia challengers over the past few years.
Why SoftBank wants in
SoftBank’s interest in Graphcore isn’t charity. It’s strategy.
Masayoshi Son has been vocal about positioning SoftBank at the center of the AI infrastructure buildout. The company already controls Arm Holdings, the chip architecture firm whose designs power virtually every smartphone on the planet. Adding Graphcore gives SoftBank a dedicated AI silicon play that complements Arm’s broader licensing business.
Look at the competitive landscape. Nvidia sits at the top with its H100 and B200 chips commanding waitlists and premium pricing. AMD is chipping away at the margins with its Instinct line. Google has its TPUs. Amazon has Trainium. Every major tech company has either built or bought its way into custom AI silicon.
What this means for the AI chip market
Companies like Cerebras, Sambanova, and Habana Labs (acquired by Intel in 2019) have all faced the same fundamental challenge: Nvidia’s CUDA software ecosystem creates switching costs that make technical superiority almost irrelevant.
Graphcore’s IPUs aren’t used for crypto mining, but SoftBank building a competing AI silicon stack could, over time, ease the GPU shortage that has constrained both industries.