Jensen Huang, CEO and founder of NVIDIA, discusses the company's position in the AI revolution and its broader accelerated computing mission. The conversation covers NVIDIA's business philosophy, supply chain strategy, competitive dynamics with TPUs and ASICs, and geopolitical considerations around China.

The discussion explores NVIDIA's transformation from a graphics company to the dominant force in AI computing, examining how the company maintains its competitive advantages through ecosystem development, supply chain management, and continuous architectural innovation. Huang addresses concerns about AI commoditizing software and explains why he believes the opposite will occur.

Key topics include NVIDIA's investment strategy in AI companies, the technical differences between GPUs and specialized AI chips, supply chain bottlenecks and scaling challenges, and the complex dynamics of technology export controls with China. The conversation also touches on NVIDIA's philosophy of doing 'as much as needed, as little as possible' while building the world's largest AI ecosystem.

The Electron-to-Token Business Model and Software Commoditization

NVIDIA's core mission is transforming electrons to tokens, requiring 'insanely hard' engineering that won't be commoditized despite software valuation crashes

Huang predicts tool usage will 'skyrocket' as AI agents proliferate: 'The number of agents that are going to be using floor planners and layout tools... is going to skyrocket'

Software companies face temporary valuation pressure because 'agents aren't good enough at using their tools yet,' but this will reverse as capabilities improve

NVIDIA follows the philosophy of doing 'as much as needed, as little as possible' - handling critical components while partnering extensively across the ecosystem

Supply Chain Lock-Up and Scaling Bottlenecks

NVIDIA has nearly $100 billion in explicit purchase commitments, with additional implicit commitments through CEO alignment across the supply chain

Huang spends significant time educating supply chain partners about AI's scale: 'I need to make sure that the entire supply chain understands what is coming at us'

Most supply chain bottlenecks resolve within 2-3 years with proper demand signals: 'Once you can build one, you could build 10. And once you can build 10, you can build a million'

Energy policy represents the primary long-term constraint: 'You can't create an industry without energy... those things take a long time'

CoWoS packaging was successfully scaled after industry focus: 'We swarmed the living daylights out of it... now I think we're in fairly good shape'

GPU Flexibility vs. Specialized AI Chips

NVIDIA built 'accelerated computing, not a tensor processing unit' - supporting diverse applications from molecular dynamics to quantum chromodynamics

Programmability enables algorithmic breakthroughs: 'The only way to really get 10x leaps, 100x leaps, is to fundamentally change the algorithm'

Blackwell achieves 50x efficiency gains over Hopper through new models, MOEs, and system-level co-design rather than just Moore's Law improvements

NVIDIA's install base spans 'several hundred million GPUs' across every cloud, providing unmatched ecosystem reach for developers

TPU adoption is primarily driven by Anthropic: 'Without Anthropic, why would there be any TPU growth at all? It's 100% Anthropic'

Investment Strategy and Missing Early Opportunities

NVIDIA invested up to $30 billion in OpenAI and $10 billion in Anthropic, but missed earlier opportunities due to scale limitations

Huang's regret: 'I didn't deeply internalize that they really had no other options, that a VC would never put in $5, $10 billion'

The company follows a non-winner-picking philosophy: 'When I invest in one of them, I invest in all of them' based on NVIDIA's own unlikely survival story

NVIDIA supports neo-clouds like CoreWeave, Nscale, and Nebius: 'If we didn't support CoreWeave to exist, these AI clouds wouldn't exist'

Investment strategy aligns with 'as little as possible' philosophy - supporting ecosystem partners rather than becoming a cloud provider directly

China Export Controls and Technology Leadership

China has '60% of the world's mainstream chips' manufacturing and '50% of the world's AI researchers' making complete isolation unrealistic

Huang argues current policies risk 'giving up the world' and creating separate ecosystems: 'Why would you want United States to give up the world?'

Chinese companies have abundant energy and can 'gang up more chips' to compensate for process node disadvantages at 7nm vs. advanced nodes

DeepSeek and Huawei's record year demonstrate continued capability: 'The amount of compute they have in China is enormous'

Huang advocates for research dialogue over isolation: 'It is essential that our AI researchers and their AI researchers are actually talking'

Architecture and software stack matter more than raw process nodes: 'Between Hopper and Blackwell... Moore's Law is dead... Blackwell is 50 times Hopper'