Nvidia is the central bank of the AI supply chain—just how deep is the moat built by Jensen Huang?

In a recent podcast episode released by a16z, renowned investor Gavin Baker and a16z partner David George had an in-depth conversation, with both agreeing that Nvidia is "in a very, very good position."

Baker believes that Jensen Huang has built a deep moat. Nvidia, through vertically integrated supply chains, securing TSMC's wafer capacity and global key component supply, and building a financeable data center ecosystem, has become the "central bank" of the AI supply chain, a moat that is extremely difficult to replicate. Host David George also stated: "The past 26 years have taught me one thing—never lose a bet against Jensen Huang."

Nine types of chips, one ecosystem

Gavin described NVIDIA's current product portfolio: nine chips—multiple accelerator chips, CPUs, Ethernet switches, two GPUs, and InfiniBand.

This is not simply a product line expansion. Gavin stated that Nvidia's strategy is "vertically integrated but horizontally open"—even if a truly excellent competing chip emerges, it will almost certainly perform better as long as it can be integrated into Nvidia's ecosystem.

This means competitors face a dilemma: confront them head-on, or integrate. Gavin's advice to all semiconductor CEOs is just one sentence: "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 each 1% market share is worth about $100 billion today, "Find a niche market and capture 1% is enough."

Supply chain lock-in: The most difficult barrier to replicate

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

According to Gavin in the podcast, Nvidia has secured 70% to 80% of the world’s key supply, including TSMC’s wafer capacity, DRAM capacity, NAND capacity, laser capacity, capacitor capacity, and everything needed to build racks.

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

This scale of supply chain integration means that even if competitors produce chips with comparable performance, they will face the practical constraint of not being able to acquire the necessary production capacity. Gavin pointed out directly: "Hardware is difficult, the real world is difficult. Huang Renxun, with his current scale and speed, has brought the entire supply chain and financing system with him at the same time, which is truly difficult to replicate."

Residual value guarantee: making financing a moat

The most easily overlooked aspect of Nvidia's competitive advantage is its financing structure.

In the conversation, Gavin broke down this mechanism in detail: Assuming an Nvidia data center costs $50 billion, the buyer only needs $15 billion in equity, and the remaining $35 billion can be raised through financing. The core reason why institutions like Blackstone, KKR, Apollo, Goldman Sachs, and JPMorgan Chase are willing to participate in the financing is that Nvidia provides residual value guarantees.

The key point is that as long as the residual value guarantee is lower than the gross profit Nvidia makes from selling chips to data centers, Nvidia faces almost zero risk and can still receive a share of the revenue.

In contrast, Gavin points out that TPUs may be the second financing option, "but would probably require at least double the equity investment and have higher financing rates."

"The cost of capital is a major advantage, which is why you want to be part of its ecosystem," Gavin said.

David further added that Nvidia helps smaller players compete with Anthropic and OpenAI through residual value guarantees, "just like he supported NeoClouds in the past—this is essentially about making computing power more accessible and beneficial to the world."

Open source is a benefit, not a threat.

There is a concern in the market that the rise of the open-source model will squeeze Nvidia's profit margins. Gavin's view is completely the opposite.

"Some people actually think this is a huge risk to his business—the logic is completely backwards," Gavin said. "Open source means that the profit margin of tokens produced on Nvidia GPUs might drop from 90% to 40%, but it also means that more tokens will be consumed, which requires more computing power. In a world with limited supply, this is a huge benefit for him."

Gavin also pointed out that Nvidia's incentive mechanisms naturally favor AI fragmentation, model diversification, and decentralized computing power, "which is entirely consistent with the national interests of the United States." He is a major advocate of open source, which has strengthened, rather than weakened, his business.

The Challenger's Dilemma: Don't Mess with Michael Jordan

Gavin used a recurring metaphor to describe competitors trying to challenge Nvidia.

"Sometimes you see people talking nonsense in front of him, like when Michael Jordan is on fire and you're talking trash to him—like in Game 50 of the regular season, he was getting a little bored, and some self-important young player decided to mess with him, and then…that was my favorite moment to watch. "

He cited the TPU team as an example, arguing that they had "pulled the cloak from Superman," with less than ideal results . Regarding Apple's self-developed AI chip, Jalapeno, Gavin offered some praise— "The first truly competitive internal ASIC I've seen, made in a fairly short time, deserves due recognition"—but also added, "Jalapeno is pulling the cloak from Superman; we'll wait and see."

Gavin also mentioned a structural reason why general-purpose GPUs are difficult to replace with dedicated chips: DeepSeek, Kimmy, and Qwen, three mainstream Chinese open-source models, have evolved in very different ways. "They can all run on general-purpose GPUs, but if you want to specialize, you need general-purpose chips to cope with the uncertainty of this evolution."

Chip trading structure reveals true preferences

Gavin also offers a unique perspective on judging true market preferences: looking at the transaction structures that chip companies sign with their customers.

He categorized transaction structures into four tiers, from best to worst: direct investment from chip companies to their clients (such as Google and Amazon's TPU and Tranium transactions with Anthropic); residual value guarantee structures (Blackstone and KKR participating in financing); warrants pegged to a fixed token price; and simply issuing warrants ("this could be a negative NPV").

"By understanding these four levels of structure, you can deduce the true customer preferences. Nvidia typically makes very good deals, and smart people are involved in their trades—that speaks volumes," Gavin said.

Risk warning and disclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.