TrendMacro conversation with Stephen Witt on Jensen Huang and the secret history of Nvidia
Huang and Nvidia made their luck with years of investment in seemingly dead technologies.
Update to Strategic View
Nvidia's domination of AI compute is the result of years of massive investment in enabling researchers to use the power of their chips' parallel-computing architecture -- when the industry had given up on it. After a decade and billions of dollars spent with nothing to show for it, neural net researchers started to get surprisingly good AI results with Nvidia's platform combining its chips with its CUDA programming tool -- when the AI industry, such as it was, had given up on neural nets as a technique. The big breakthrough -- transformers -- happened at Google, but their woke culture kept them from using it for years. Nvidia has substantially achieved "vendor lock" with its combination of hardware and software services. Its most dangerous competition now comes from ASICs, chips custom-made to accelerate particular AI tasks -- Nvidia missed it, and for now Broadcom is the leader. Nvidia is starting to enjoy dominance in robotics thanks to its Omniverse virtual training tool. Nvidia is critically vulnerable to any disruptions at Taiwan Semiconductor which manufactures virtually all its product, and has no back-up plan. CEO Huang is in China now meeting with Xi as part of Trump's summit party, having been accidentally forgotten on the guest list for the flight over on Air Force One. With China's control of Taiwan in the summit headlines, Huang -- a volatile and narcissistic personality -- has a delicate diplomatic task.