Nvidia dominates the market for graphics processing units used to train artificial intelligence models, although rivals like AMD and Intel are trying to catch up. Now some scrappy startups are entering the fray with alternative designs for AI chips they claim work better and more efficiently than Nvidia’s GPUs. Other startups are targeting Nvidia’s app-writing software, which keeps companies using its chips.
Young companies such as d-Matrix and Rain Neuromorphics are pitching their chips and software as a way for companies to reduce the costs of training and running machine-learning models compared with Nvidia’s products. Tiny Corp and Modular are developing alternatives to Cuda, an Nvidia programming language that lets developers speed up their applications and that only works with Nvidia’s GPUs. Three of the startups we profile here—Qyber, Modular and MatX—were founded by engineers who previously worked at Google, which sells AI chips that aim to rival Nvidia’s. That’s a sign ambitious tech entrepreneurs feel they can make a dent in the sector, despite the deep-pocketed competition.