NVIDIA has established an immense and resilient supply chain, with purchase commitments in the hundreds of billions and the capacity to scale to a trillion dollars. This is underpinned by decades-long, trust-based relationships with key partners like TSMC, often without formal contracts.
Jensen Huang's core thesis is that the scaling of general-purpose computing, as described by Moore's Law, has effectively ended. The future of performance gains lies in domain-specific acceleration, which offers 10x to 100x improvements over the incremental gains from traditional CPUs.
Huang argues forcefully that US chip export restrictions on China are a strategic error. He contends that these policies will not stop China but will instead force it to build a parallel, competitive AI ecosystem, ultimately eroding the global dominance of the American tech stack.
Despite the rise of custom silicon like Google's TPUs and AWS's Tranium, Huang is confident in NVIDIA's moat. He argues that performance comes from full-stack optimization and algorithmic advances (like the 50x gain from Hopper to Blackwell), an ecosystem that in-house chips cannot easily replicate.
Contrary to fears of AI replacing software jobs, Huang predicts an exponential increase in both AI agents and the use of existing software tools. He believes agents will act as force multipliers, driving up the number of instances and overall usage of software, much like AI in radiology increased the demand for radiologists.
Keep pulling the thread on Jensen Huang.