People Most Likely to Break Nvidia’s Chip Grip

Lisa Su rebuilt AMD into a serious computing leader and now offers the only broad accelerator roadmap with enough scale to challenge Nvidia directly. Customers need that second source.

Hock Tan positioned Broadcom inside hyperscalers’ custom AI chips and the networking that connects them. He attacks Nvidia through customers large enough to design their own alternatives.

Andrew Feldman has kept Cerebras committed to wafer-scale computing through years of skepticism. The machines offer a radically different way to train and serve large models.

Sid Sheth leads d-Matrix, which places computation close to memory to reduce inference cost and delay. That bottleneck matters increasingly as trained models serve more users.

Ebrahim Hussain founded Architect Labs to use AI in the chip-design process itself. Its claim of dramatically faster development could change who is able to create specialized processors.

Jeff Dean has guided Google’s AI systems and TPU program across several generations. Google’s ability to train major models without depending entirely on Nvidia is a powerful proof point.

Rene Haas leads Arm as its designs expand through cloud servers and specialized accelerators. Arm’s licensing model lets many companies build alternatives without creating an instruction set from scratch.

Gavin Uberti co-founded Etched around a chip designed specifically for transformer inference. The narrow bet could deliver exceptional efficiency, though architecture changes would create obvious risk.

Jonathan Ross co-founded Groq around an architecture built for extremely fast, predictable inference. If deployed AI becomes the larger cost, that specialization becomes more valuable.

Zhao Lidong co-founded Enflame to build domestic AI accelerators for China’s data centers. The September IPO will test whether the company can convert strategic demand into sustained execution.