On the ground at Nvidia GTC Berlin: Jensen Huang, robot ducks and neocloud distress

Jensen Huang has been busy.

Nvidia's GTC in Berlin was the grand finale of a week-and-a-half whistle stop European tour. Huang met with world leaders — including U.K. Prime Minister Andy Burnham, French President Emmanuel Macron and German Chancellor Friedrich Merz — announced new products and even spent time talking to small robot ducks on stage.

The Nvidia chief must be tired. At a media Q&A on Thursday morning, he gave reporters an insight into his mantra. Talking about bringing data centers online faster, Huang said: "Try less sleep, which is what I do."

And the trip's not over for Huang. Today, he's in Kazakhstan, touting the country's potential as a major AI hub.

Amid all the activity in Germany, I spent two days in Berlin running around meeting with neocloud execs, investors and data center builders and suppliers.

Here are the main themes that came up.

Neocloud distress

The final day of the conference came amid a selloff among public neocloud stocks. CoreWeave ended the day down 10% and Nebius dropped 7% by market close.

It continued a tumultuous period for companies leasing AI compute, with many of their stocks plummeting on concerns over heavy borrowing.

Huang — perhaps unsurprisingly considering Nvidia's sizeable stakes in many of those companies — was more bullish.

"If you look at the fundamentals, the demand for the neo clouds is off the charts," Huang told CNBC.

Despite that, there was concern about the neocloud sector among many of the folk I spoke to at GTC. The "magnitude of businesses with no credit rating" was "freaking the market out," said one exec at an AI infrastructure company I met with.

Data centers are getting smaller

Speaking onstage in Berlin, Huang announced a new product that packs seven of Nvidia's Vera Rubin AI racks into a shipping container and is designed to be transported on normal roads.

It's a play to get AI compute online more quickly amid huge demand. Huang told the audience that billions of dollars in revenue can be lost by delays in getting data centers up and running.

"By moving this into the supply chain, the speed by which we can turn on" will ramp up, he said.

Major compute consumers are already exploring opportunities for smaller data center deployments. Sources had told me in September that OpenAI and Anthropic have been hunting for smaller AI data center deals.

Capital could be the next big bottleneck

AI infrastructure is hugely expensive to build.

Goldman Sachs estimates that AI-related capital spending could total about $7.6 trillion from 2026 through 2031, across compute, data centers and power.

A 1 gigawatt data center costs about $50 billion, Huang told the Berlin audience on Wednesday.

"What is missing in the middle is the working capital to fuel that growth… because there's just so much demand across so many sectors of AI, there's a scarcity of capital," he said during his CNBC interview on Thursday.

"The demand of the capital is much greater than the availability of the capital," he told my colleague Arjun Kharpal, though the CEO added: "That's going to get solved."

One person I spoke to on the sidelines, an exec at an AI infrastructure financing company, said "very soon capital will become the bottleneck."