NVIDIA: The Central Bank of the Global AI Supply Chain - How Impenetrable Is Jensen Huang's Built Unbeatable Moat?

Gavin Baker believes that NVIDIA has become the "central bank" of the AI supply chain with an extremely hard-to-replicate moat, by vertically integrating the supply chain, locking in TSMC's wafer production capacity and the supply of key global components, and building a financible data center ecosystem. NVIDIA's residual value guarantee mechanism allows its data centers to be financed with a low equity ratio, attracting institutions including Blackstone and KKR to participate, while the financing cost for competing products such as TPU is significantly higher.

Recently, in a recent podcast episode released by a16z, renowned investor Gavin Baker and a16z partner David George had an in-depth dialogue, and both sides agreed that Nvidia is "in a very, very favorable position."

Baker believes that Jensen Huang has built an unassailable moat. Through vertically integrating the supply chain, locking TSMC's wafer capacity and the global supply of key components, and building a financizable data center ecosystem, Nvidia has become the "central bank" of the AI supply chain, with a moat that is extremely difficult to replicate. The host David George also said: "The past 26 years have taught me one thing — never bet against Jensen Huang."

01 Nine Types of Chips, One Unified Ecosystem

Gavin described Nvidia's current product matrix: Nine types of chips — a variety of acceleration chips, CPUs, Ethernet switches, two types of GPUs, plus InfiniBand.

This is not a simple product line expansion. Gavin stated that Nvidia's strategy is "vertical integration with horizontal openness" — even if a truly excellent competitive chip emerges, it will almost certainly perform better if it can access Nvidia's ecosystem.

This means competitors are faced with a dilemma: confront head-on, or integrate. Gavin's only piece of advice for all semiconductor CEOs is: "The only thing you need to say is 'Thank you Jensen Huang, thank you for creating this opportunity, how can we cooperate with you'."

He added that every 1% of market share is currently worth roughly $100 billion, "Find a niche market, and capturing 1% of it is enough."

02 Supply Chain Lock-in: The Hardest-to-Replicate Barrier

Nvidia's moat lies not only in its chip design capabilities, but also in its control over the supply chain.

According to Gavin's introduction on the podcast, Nvidia has locked 70% to 80% of the global key supply, including TSMC's wafer capacity, DRAM capacity, NAND capacity, laser production capacity, capacitor production capacity, and everything needed to build server racks.

"Over the past 15 years, he has upgraded bets of billions of dollars made every two to three years into bets of hundreds of billions of dollars, while bringing the entire supply chain and financing system along with him." Gavin said.

This scale of supply chain integration means that even if a competitor launches a chip with equivalent performance, it will face the realistic constraint of no available production capacity. Gavin pointed out directly: "Hardware is hard, the real world is hard. And Jensen Huang, with his current scale and speed, has brought the entire supply chain and financing system along with him at the same time, which is really difficult to replicate."

03 Residual Value Guarantee: Making Financing a Moat

The most easily overlooked part of Nvidia's moat is its financing structure.

Gavin broke down this mechanism in detail during the dialogue: Assuming an Nvidia data center costs $50 billion, the buyer only needs $15 billion in equity, and the remaining $35 billion can be covered through financing. Institutions including Blackstone, KKR, Apollo, Goldman Sachs, and JPMorgan Chase are willing to participate in the financing, and the core reason is that Nvidia provides a residual value guarantee.

The key point is: as long as the residual value guarantee is lower than Nvidia's gross profit from selling chips to data centers, Nvidia bears almost zero risk, and can also earn revenue sharing from it.

In contrast, Gavin pointed out that TPU may be the second most financizable option, "but it probably requires at least double the equity investment, and the financing interest rate is higher."

"Capital cost is a huge advantage, which is why you just want to be part of his ecosystem." Gavin said.

David further added that Nvidia uses residual value guarantees to help small and medium-sized players compete with Anthropic and OpenAI, "just like he supported NeoClouds in the past — this is essentially universalizing computing power, which is beneficial to the world."

04 Open Source Is a Boon, Not a Threat

There is a concern in the market that the rise of open source models will compress Nvidia's profit margins. Gavin's judgment on this is completely opposite.

"Some people actually think this is a huge risk to his business — the logic is completely reversed." Gavin said, "Open source means that the profit margin of Tokens generated on Nvidia GPUs may drop from 90% to 40%, but this means more Tokens will be consumed, which in turn requires more computing power. In a supply-constrained world, this is a huge boon for him."

Gavin also pointed out that Nvidia's incentive mechanism naturally favors AI fragmentation, model diversification and distributed computing power, "which is completely consistent with the national interests of the United States." He is the biggest advocate of open source, which in turn strengthens his business rather than weakens it.

05 The Challenger's Dilemma: Don't Provoke Michael Jordan

For competitors trying to challenge Nvidia, Gavin used a recurring metaphor.

"Sometimes you see someone making arrogant remarks in front of him, just like seeing someone talking nonsense to Michael Jordan when he is at the top of his game — it's the 50th game of the regular season, he's a little bored, then some young player who thinks highly of himself decides to provoke him, and then... that's my favorite moment to watch."

He took the TPU team as an example, believing that it once "tugged at Superman's cape" with unsatisfactory results . For Apple's self-developed AI chip Jalapeno, Gavin gave certain affirmation — "the first truly competitive in-house ASIC I have ever seen, built in a very short period of time, it deserves the credit it is due" — but he also said, "Jalapeno is tugging at Superman's cape, let's wait and see."

Gavin also mentioned a structural reason explaining why general-purpose GPUs are difficult to be replaced by dedicated chips: The three mainstream Chinese open source models DeepSeek, Kimmy and Qwen evolve in very different ways, "they can all run on general-purpose GPUs, but if you want to go dedicated, you need general-purpose chips to cope with the uncertainty of this evolution."

06 Chip Transaction Structures Reveal Real Preferences

Gavin also provided a unique perspective to judge the real preferences of the market: look at the transaction structures signed between chip companies and their customers.

He divided the transaction structures into four tiers, from best to worst: chip companies directly invest in customers (such as Google and Amazon's TPU and Trainium transactions with Anthropic); residual value guarantee structures (Blackstone and KKR participating in financing); warrants linked to fixed Token prices; and simply providing warrants ("which may have negative NPV").

"Through the hierarchy of these four structures, you can infer the real customer preferences. Nvidia's transactions are usually very good, and smart people are participating in his transactions — that speaks for itself." Gavin said.

This article is from the WeChat official account "Hard AI", author: a researcher focusing on technology production and research, published with authorization from 36Kr.