Intel CEO Pat Gelsinger believes the entire tech sector is fed up with Nvidia CUDA’s dominance. According to him, it is finally.

Today, when you say AI, you say Nvidia. The rest of the tech industry doesn’t like that. AMD is trying to break Nvidia’s dominance with its MI300 accelerators, and Intel is hoping to play a major role in the AI ​​ecosystem with the integrated accelerator capabilities of its latest Xeon Scalable processors.

That’s all well and good, but Nvidia is currently dominating through a kind of vendor lock-in with CUDA. Nvidia accelerators were the only option for a long time and they work with CUDA, so now everyone works with CUDA, making it less obvious to choose an accelerator manufacturer other than Nvidia.

Eliminate

That won’t last, says Intel CEO Pat Gelsinger. “The entire industry is motivated to eliminate the CUDA market,” he said. Gelsinger refers, among others, to OpenAI and Google, which are themselves developing towards a “Pythonic Programming Layer”. Gelsinger says AI specialists are increasingly looking for open training models.

Openness is also how AMD wants to make a difference, but Gelsinger probably doesn’t want to advertise too much for this party. Nvidia, for its part, recently announced that its H100 accelerator remains the best chip on the market, at least if you fully commit to the Nvidia ecosystem.

The CEO further points out that you only have to train an AI model once. Once this happens, the dependency on CUDA disappears and the discussion changes. “Can he operate the model properly? “Basically, it’s the inference market where the game is played,” he predicts. And as it happens, Intel has fifth-generation Xeon Scalable chiplets on board to make that conclusion easier.