One of the clearest signs of a company's power is when its customers cannot easily leave. In artificial intelligence, that company is Nvidia, whose chips train nearly all of the world's advanced AI. So it is notable, if modest, when a customer decides to send some of its work elsewhere. Turing, a Japanese startup building self-driving technology, says it will move about 10% of its AI training onto chips made by AMD, Nvidia's smaller rival, Yahoo Finance reported.

Who Turing is

Turing is not the famous chip designer; it is a five-year-old Japanese company developing the software brains for fully self-driving cars, with the aim of commercializing the technology later this decade. Like almost everyone training large AI models, it has until now leaned entirely on Nvidia's hardware. The decision to diversify came alongside an investment: AMD's venture arm, together with Japan's Mitsubishi Corporation, put money into Turing, according to reports of the funding round. In return, Turing agreed to run part of its computing on AMD's chips.

The jargon, briefly

Two terms help make sense of this. A "GPU" (graphics processing unit) is the specialized chip used to train and run AI; Nvidia makes the dominant ones. "Training" is the compute-hungry process of teaching an AI model from data, as opposed to "inference," the cheaper task of running the finished model. And the reason Nvidia is so hard to leave is not only its chips but its software: a platform called CUDA that developers have built their tools around for nearly two decades. Rewriting all of that to run on a competitor's chips is costly, which is a big part of why Nvidia has kept its lead even when rivals offer competitive hardware.

Why 10% matters more than it sounds

On its own, shifting a tenth of one startup's training is a rounding error in a market Nvidia still dominates, with an estimated 80%-plus share of AI chips. But it matters as a proof point. It shows AMD's accelerators are now capable enough to handle real production training, not just lab demonstrations, and that a customer is willing to bear the switching costs to avoid depending on a single supplier. For any buyer, having a credible second source is also leverage: it strengthens the hand in price negotiations with a vendor that has enjoyed enormous pricing power.

The move fits a broader push across the industry to chip away at Nvidia's moat, both by rivals like AMD building faster hardware and by engineers developing software that lets AI models run across different chips without being locked to CUDA. None of it dethrones Nvidia today.

Why it matters

For investors watching the AI build-out, the significance is directional. Nvidia's extraordinary margins rest on its near-monopoly in training chips; anything that gives customers a real alternative, even at the margin, chips away at that. AMD, long the perennial runner-up, is trying to convert improving hardware into actual design wins, and a named customer moving workloads is exactly the kind of evidence it needs. Whether this becomes a trickle or a flood depends on how quickly the software barriers fall and how aggressively big AI spenders push for a second supplier. For now, Turing's decision is a small crack, worth watching precisely because cracks in a moat this wide have been so rare. This article is informational and not investment advice.