“We think of the CUDA moat as shallow and small,”
Pat Gelsinger says inference will matter more than training, and inference has “no CUDA dependency.” That happens to be the market where Intel has a product. He wants “100% of the datacenter AI TAM” through chips, accelerators and foundry deals with Nvidia itself. A moat that is shallow and small does not need a CEO on stage at the NASDAQ to drain it.