Jensen Huang – TPU competition, why we…

1. Key Themes

The "Electrons to Tokens" Framework: NVIDIA as the World's Essential Transformation Engine

Jensen reframes NVIDIA not as a chip company or even a software company, but as the entity responsible for transforming electrons into valuable tokens — and argues this transformation is nowhere near commoditized.

"The input is electron, the output is tokens. That is in the middle NVIDIA. And our job is to do as much as necessary, as little as possible to enable that transformation to be done at incredible capabilities." 00:01:54

The moat isn't just hardware — it's the entire supply chain orchestration, the ecosystem of partners across all five layers of AI, and the deliberate philosophy of doing only what others cannot.

"If we didn't create them, nobody would. And I am completely certain of that." 00:45:21

The Supply Chain as Strategic Weapon: Informing, Aligning, and Locking In the Upstream

NVIDIA's $100B+ in purchase commitments is just the visible tip. The deeper moat is Jensen's personal role in educating and aligning upstream CEOs on the size of the coming opportunity — creating willingness to invest that no competitor can replicate.

"Why are they willing to make the investments for me and not someone else? And the reason for that is because they know that I have the capacity to buy it, buy their supply and sell it through my downstream." 00:05:50

"Most of my keynotes... it almost comes across like education. And in fact, that's exactly on my mind. I need to make sure that the entire supply chain upstream and downstream, the ecosystem, understands what is coming at us." 00:07:11

Jensen proactively solves bottlenecks before they become bottlenecks — from CoWoS packaging to silicon photonics — and claims no supply bottleneck lasts more than 2-3 years.

"My point is that none of the bottlenecks last longer than a couple, two, three years. None of them." 00:15:09

The China Chip Debate: NVIDIA's Existential Argument Against Export Controls

Jensen makes the case that export controls are strategically self-defeating for America. His argument: China already has sufficient compute (7nm ≈ Hopper), abundant energy, and 50% of world's AI researchers. Denying chips doesn't stop China — it only concedes the developer ecosystem and global market share to Huawei.

"50% of the world's AI researchers. If you're worried about them, what is the best way to create a safe world? Well, victimizing them, turning them into an enemy likely isn't the best answer." 00:59:44

"The policy that you're advocating resulted in the American telecommunication industry being policyed out of basically the world to the point where we don't control our own telecommunications anymore." 01:25:40

"We shouldn't concede it. If we lose it, we lose it. But why do we concede it?" 01:31:16

2. Contrarian Perspectives

Software Tool Companies Will See Skyrocketing Revenue from AI — Not Get Killed By It

While Wall Street punishes software companies on fears of AI commoditization, Jensen argues the opposite: AI agents will massively multiply the number of tool users, driving usage of companies like Synopsys and Cadence to new highs.

"I think the number of agents are going to grow exponentially. The number of tool users are going to grow exponentially... The number of instances of Synopsys design compiler is going to skyrocket." 00:03:29

"The reason why it hasn't happened yet is because the agents aren't good enough at using their tools yet." 00:04:28

TPUs Are Not a Real Threat to NVIDIA — The Anthropic TPU Deal Is an Anomaly, Not a Trend

Widely cited as evidence that ASICs are displacing GPUs, Jensen flatly dismisses this narrative.

"Anthropic is a unique instance and not a trend. Without Anthropic, why would there be any TPU growth at all? It's 100% Anthropic." 00:37:03

He also reveals the real reason Anthropic uses TPUs has nothing to do with performance: Google and AWS made multi-billion dollar investments into Anthropic early on, and compute usage was part of the deal. NVIDIA simply wasn't in a position to make those investments at the time.

"At the time, I didn't deeply internalize how difficult it would be to build a foundation AI lab like OpenAI and Anthropic... they needed huge investments from the supplier themselves. We just weren't in a position to make the multibillion dollar investment into Anthropic so that they could use our compute." 00:39:33

Architecture Matters Far More Than Process Node — Moore's Law Is Dead as a Competitive Moat

Jensen argues the common assumption that China is stuck at 7nm and therefore crippled is deeply wrong. Architecture improvements, not process node jumps, drive real performance gains.

"Is Blackwell 50 times more advanced lithography than Hopper? Not even close. Between Hopper and Blackwell, from the transistors themselves, call it 75%... Blackwell is 50 times Hopper. My point is architecture matters." 00:32:44

This is both a defense of China's capabilities AND a self-justification for NVIDIA's own lead. It cuts both ways.

The Real Bottleneck in AI Is Not Chips — It's Plumbers and Energy

While everyone debates TSMC capacity and EUV machines, Jensen points to a far more overlooked constraint that cannot be solved with capital: physical infrastructure trades and energy policy.

"It's the stuff that's downstream from us. Energy policies that prevent energy from... you can't grow, you can't create an industry without energy." 00:15:39

"More chip capacity, that's a two, three-year problem. More co-op capacity, two, three-year problem." 00:16:08

NVIDIA Deliberately Does Not Price-Gouge During Chip Shortages — And This Is a Competitive Advantage

Counterintuitively, NVIDIA refuses highest-bidder allocation and holds firm prices even during massive supply shortages. Most would assume this leaves money on the table. Jensen argues it builds the dependability that allows NVIDIA to be the foundation others bet their businesses on.

"We never do that. We set our price. We set our price. And then people decide to buy it or not... I prefer to be dependable, to be the foundation of the industry." 00:54:16

3. Companies Identified

CoreWeave Neo-cloud / AI infrastructure company. Jensen credits NVIDIA with enabling CoreWeave's existence, and it's been reported NVIDIA is backstopping up to $6.3B with $2B invested. Mentioned as a key node in NVIDIA's downstream ecosystem strategy.

"If we didn't support CoreWeave to exist, these NeoClouds, these AI Clouds wouldn't exist. If we didn't help CoreWeave exist, they would not exist." 00:46:15

Micron Memory manufacturer. Highlighted as an early believer in Jensen's AI vision, making outsized bets on HBM and LPDDR when others hadn't yet committed.

"Sanjay and the Micron team, I still remember the meeting really well, where I was clear about exactly what's going to happen and why it's going to happen... They really invested in it. And it obviously has been tremendous for the company." 00:11:49

Lumentum / Coherent Silicon photonics ecosystem suppliers. Jensen highlights these as examples of NVIDIA proactively building future supply chain infrastructure years in advance, specifically for optical interconnects at scale.

"The investments that we've done with Lumentum and Coherent and all of the silicon photonics ecosystem. The last several years, we really reshaped the ecosystem and the supply chain of silicon photonics." 00:12:17

Synopsys / Cadence EDA software tool companies. Jensen argues they are massively underappreciated beneficiaries of the AI agent wave — not victims of it.

"The number of agents that are today were limited by the number of engineers. Tomorrow, those engineers are going to be supported by a bunch of agents and we're going to be exploring the design space like you've never seen explored before." 00:03:59

Crusoe AI-native cloud provider. Among the first to deploy Blackwell and Blackwell Ultra; also deploying Vera Rubin. Notable for cross-user, cross-GPU KV caching for inference — delivering 10x faster time-to-first-token and 5x better throughput than VLLM. Mentioned as an example of a Neo-cloud that NVIDIA has helped build.

"Most inference engines already do KV caching for a single user's forward passes. But Crusoe does it across users and GPUs." 00:23:52

Huawei Chinese tech/chip company. Jensen uses Huawei as his primary evidence that export controls have backfired — Huawei had its largest year ever, went into AI chips, and is now developing silicon photonics to gang chips at scale.

"What's the evidence? Huawei just had the largest single year in the history of the company." 00:08:26

NScale / Nebius Neo-cloud AI infrastructure companies. Mentioned alongside CoreWeave as examples of NVIDIA-enabled AI cloud ecosystem companies that wouldn't exist without NVIDIA's support.

"If we didn't support NScale, they wouldn't be where they are today. If we didn't support Nebius, they wouldn't be where they are today." 00:46:36

4. People Identified

Sanjay Mehrotra (CEO, Micron) CEO of Micron Technology. Highlighted as an early, visionary partner who believed in Jensen's AI thesis and made decisive bets on HBM when others hesitated.

"Sanjay and the Micron team, I still remember the meeting really well, where I was clear about exactly what's going to happen... They really doubled down on it." 00:11:49

Dylan Patel (SemiAnalysis) Semiconductor analyst. Referenced multiple times by Jensen as a reliable, credible external source for NVIDIA performance benchmarks and supply chain data. Jensen specifically references his "Inference Max" benchmark as the gold standard no competitor has beaten.

"Dylan's, right, Inference Max is sitting out there for everybody to use. And not one TPU won't come." 00:32:24

5. Operating Insights

"As Much as Needed, As Little as Possible" — The Philosophy Behind NVIDIA's Business Model Decisions

Jensen articulates a precise operating philosophy that explains every major strategic choice NVIDIA makes: only do what no one else will do; partner for everything else. This applies to whether to become a cloud, a financier, a foundation model lab, or a chip foundry.

"We should do as much as needed, as little as possible. And what that means is the work that we do with building our computing platform, if we don't do it, I genuinely believe it doesn't get done." 00:44:34

Operators and investors should stress-test every new initiative against this question: Is this something that won't happen unless we do it? If someone else will build it anyway, focus resources elsewhere.

GTC as a Supply Chain Alignment Tool, Not Just a Product Launch

Jensen uses GTC not primarily as a consumer marketing event but as a mechanism to inform, align, and inspire the entire upstream and downstream supply chain simultaneously — so that every partner can "see firsthand all the things that I tell them."

"I bring them together so that the downstream could see the upstream, the upstream could see the downstream, and all of them could see all the advances in AI." 00:06:20

This is a replicable playbook for any platform company: create a venue where your entire ecosystem gathers so they can validate the opportunity directly, reducing the need for individual persuasion campaigns.

Prefetching Bottlenecks Years in Advance Through Ecosystem Investment

NVIDIA identifies supply chain bottlenecks 3-5 years out, then proactively builds the ecosystem — through investment, patent licensing, and new technology development — before the constraint becomes acute. CoWoS packaging and silicon photonics are cited as examples.

"We're prefetching the bottlenecks years in advance... We built up an entire supply chain around TSMC. We partnered with them on CoWoS, invented a whole bunch of technology. We licensed those patents to the supply chain, keep it nice and open." 00:12:17

6. Overlooked Insights

NVIDIA's Strategic Investment in Foundation Labs Was Driven By Regret — And Is Now a Deliberate Competitive Moat

Jensen openly admits that NVIDIA's failure to invest in Anthropic early enough was a strategic error — not a conscious choice. Anthropic went to Google and AWS because those companies wrote checks; NVIDIA didn't. Jensen has now corrected this, investing in both OpenAI (~$30B reported) and Anthropic, but at far higher valuations.

The non-obvious insight: NVIDIA is now structurally committed to being an investor in foundation model companies as a business necessity, not just financial opportunity. If they don't invest, the labs go to whoever will — and those investors get supply commitments in return. This means NVIDIA's investment portfolio is effectively a customer acquisition and retention tool disguised as venture capital.

"My mistake is I didn't deeply internalize that they really had no other options... In return, used their compute. We just weren't in a position to do so at the time. But I'm not going to make that same mistake again." 00:40:01

"When I was able to, when Anthropic came to us, I'm delighted to be an investor, delighted to help them scale." 00:40:57

Groq Acquisition Signals NVIDIA Is Quietly Entering the Inference Latency Market Segment

Buried in a long answer about chip architecture diversity, Jensen reveals that NVIDIA recently acquired Groq and is folding it into the CUDA ecosystem. The rationale: tokens now have differentiated value based on response time, creating a premium segment for ultra-low-latency inference that justifies different hardware economics.

This is a massive, underreported strategic move. Groq's LPU architecture is purpose-built for inference speed — and NVIDIA is now absorbing it rather than competing with it, while using it to capture a new high-value market segment that didn't exist "a couple of years ago."

"Recently we added Groq. And we're going to fold Groq into our CUDA ecosystem... the value of tokens have gone up so high that you could have different pricing of tokens... if I can give them much more responsive tokens so that they're even more productive than they are today, I would pay for it." 00:37:28