Intel CEO Pat Gelsinger unveiled Intel's Core Ultra, fifth-generation Xeon data center chips and its latest AI chip, Gaudi 3, at an event in New York on Thursday. And claimed that inference technology will be more important than artificial intelligence training. When asked, Gelsinger said that NVIDIA's CUDA advantage in the training field will not last forever.

Gelsinger: "You know, the entire industry has incentives to kill the CUDA market." He cited companies such as MLIR, Google, and OpenAI as examples of companies that are moving to a "Pythonic programming layer" to make artificial intelligence training more open.

"We believe CUDA's moat is shallow and small," Gelsinger continued. "Because the industry has an incentive to bring a broader range of technologies to a broad range of training, innovation, data science, etc." But Intel doesn't just rely on training. Instead, it believes that reasoning is the way to go.

Gelsinger believes: "Because the inference happens, once you train the model... you are not relying on CUDA." The key is, can you run the model well? He said Intel will rise to the challenge with Gaudi3, which was demonstrated on stage for the first time today, and will do so with Xeon and edge PCs. "It's not that Intel won't compete in the training space, but fundamentally the inference market is where the game is," Gelsinger said.

The just-announced Gaudi3 will launch next year and will compete with chips from rivals Nvidia and AMD to power large and power-hungry AI models.

The release of Gaudi3 is of great significance to Intel. This chip can give Intel the capital to compete with Nvidia's H100 and AMD's just-launched MI300X.

We know that the most famous AI models, such as OpenAI’s ChatGPT, run on Nvidia GPUs in the cloud. That's one reason why Nvidia's stock price is up nearly 230% so far this year. That's why companies like AMD and now Intel have released their own latest AI chips, which they hope will lure AI companies away from Nvidia's dominance of the market.

While the company is sparse on details, Gaudi3 will compete with Nvidia's H100, a prime choice for companies building massive chip fleets to power artificial intelligence applications, and AMD's upcoming MI300X, which will start shipping to customers in 2024.

Intel has been making Gaudi chips since 2019, when it acquired a chip developer called Habana Labs.

Gelsinger said: "We are already seeing a boom in generative artificial intelligence, which is the star product of 2023.