Jensen Huang: demand solves the AI financing scare — and Nvidia chips are now the loan collateral
Key points
Key takeaways from Yahoo Finance's interview with Nvidia founder and CEO Jensen Huang, live after his GTC Berlin keynote, with Atreides Management CIO Gavin Baker joining in — on the AI financing scare, China, and where the build-out goes next.
Demand solves everything. Asked whether the neoclouds face a shakeout as debt markets wobble, Huang said: "This is going to get solved, and the reason is that the demand is so good" — citing SpaceX, CoreWeave, Nebius and Scale as all needing far more capacity than they have. His one-liner: "With demand, everything will be resolvable."
The DSX Pod: cutting idle capital. A gigawatt data centre costs roughly $50 billion, and four to six months pass between components shipping and the facility earning money. Nvidia's DSX Pod — "an AI factory in a shipping container" — pre-builds more of the system in the supply chain, aiming to take three to four months out of that window, accelerating time to revenue and shrinking the idle cash sitting in unopened hardware.
Underwriting loans against the chips. Huang argues Nvidia systems stay productive for seven to eight years or more — Blackwell and Vera Rubin alike — because CUDA and new software layers keep refreshing the installed base. That durability plus fungibility means "you might be able to underwrite a loan against the asset itself": residual value is high, and "the risk involved in underwriting Nvidia's computing is the lowest of any asset out there" — though "nobody gives credit for that at the moment".
Baker's DeepSeek data point. Gavin Baker said DeepSeek has disclosed it monetises inference at $38 per GPU hour: "Last time I checked, I hadn't heard of anybody renting GPUs anywhere near that high. So maybe this explains why Blackwell prices have been up 17 days in a row."
The bottlenecks keep moving. The constraint chain has run from wafers to CoWoS packaging to HBM memory to silicon photonics to connectors — and now power. But often the power exists and the land, shell, construction or grid connection lags, and permitting varies by country and state. Huang's answer is diversification: Nvidia is "the only computing company that supports every single AI model, every CSP, every neocloud in every country", so it can ship compute wherever a bottleneck clears first.
Open models and sovereign AI. Roughly $600 billion of venture financing went into AI-native startups in the first half of the year, and Huang says 80% use open models — post-trained with their own data into domain-specific "super AIs". The future is hybrid, like cloud: "rent a closed frontier-lab model whenever you can, build your own AI only when you must" — some companies, such as Cadence and Siemens, need sovereign AI.
China. On reports that DeepSeek and Huawei are co-designing chips, Huang said the US "should always take China seriously" — no one manufactures chips, open models and AI researchers at greater scale. "We have to recognise we have very serious competition", and the answer is to compete on innovation; Washington's message to Nvidia, he said, is partly to "go compete around the world" — hence Africa and Central Asia, and ambitions in Brazil and Egypt.
Building the AI economy. After a trip from London to France, Sweden and Berlin, Huang said Nvidia has invested about $10 billion in European companies over the last year and announced 14 more this week. The industry has shifted in the last six months from "invention mode" to "AI useful mode" — and his keynote message to Europe was: stop overthinking, start building.
Read more: Yahoo Finance on YouTube
