Nvidia's Berlin GTC: Jensen Huang's Europe Tour, Container Data Centers and Neocloud Jitters

CoreWeave closed down 10% and Nebius dropped 7% on the final day of Nvidia's GTC conference in Berlin, capping a week in which the companies that lease AI compute saw their stocks slide on concerns over heavy borrowing. The selloff framed the defining tension of the show: demand for AI infrastructure, in Nvidia CEO Jensen Huang's words, is "off the charts," yet the money and credit needed to build it are showing strain.

The Berlin event was the finale of a week-and-a-half European tour during which Huang met world leaders including U.K. Prime Minister Andy Burnham, French President Emmanuel Macron and German Chancellor Friedrich Merz, announced new products and, at one point, spent time on stage talking to small robot ducks. At a media Q&A on Thursday morning, Huang explained his approach to bringing data centers online faster: "Try less sleep, which is what I do." The tour continued after the conference, with Huang traveling to Kazakhstan on Friday to promote the country's potential as a major AI hub.

Why do the neocloud losses matter?

The public market rout among neoclouds (companies that rent out GPU compute from data centers they build or lease) has extended a tumultuous period for the sector, with many stocks plummeting on investor concern about the debt stacked behind those buildouts. Huang, who perhaps unsurprisingly holds sizeable Nvidia stakes in many of these companies, dismissed the market anxiety.

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

Not everyone in Berlin shared that confidence. Among executives, investors and data center builders and suppliers consulted over two days at the conference, unease about the sector was a recurring theme. One executive at an AI infrastructure company said the "magnitude of businesses with no credit rating" was "freaking the market out."

The distinction matters for the broader AI supply chain. Nvidia's revenue depends heavily on hyperscalers and neoclouds absorbing its GPUs; a credit squeeze among the smaller buyers would concentrate demand among a handful of cash-rich players and slow deployments.

What did Huang announce in Berlin?

The headline product reveal was physical rather than architectural. Speaking onstage, Huang announced a new offering that packs seven of Nvidia's Vera Rubin AI racks into a shipping container designed to be transported on normal roads.

The logic is speed. Huang told the audience that billions of dollars in revenue can be lost to delays in getting data centers up and running. "By moving this into the supply chain, the speed by which we can turn on" compute will ramp up, he said.

The containerized racks align with a shift already underway among major compute buyers. Sources had reported in September that OpenAI and Anthropic have been hunting for smaller AI data center deals, a departure from the mega-campus model that has dominated the buildout so far. A road-transportable, seven-rack unit lets those customers deploy capacity incrementally rather than waiting years for gigascale sites.

Is capital the next bottleneck?

The largest numbers cited in Berlin concerned money, not chips. Goldman Sachs estimates that AI-related capital spending could total about $7.6 trillion from 2026 through 2031, across compute, data centers and power. Huang put the cost of a single 1 gigawatt data center at about $50 billion in his Wednesday remarks to the Berlin audience.

The gap between that price tag and available financing is what Huang flagged as the industry's emerging constraint.

"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 in his Thursday CNBC interview.

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

People outside Nvidia were blunter. One executive at an AI infrastructure financing company, speaking on the sidelines of the conference, said "very soon capital will become the bottleneck."

That warning cuts against the conventional framing of chips and power as the binding constraints on AI growth. If financing moves to the front of the queue, the winners of the next phase of the buildout may be determined less by GPU allocation than by balance-sheet strength.

What else happened this week?

Several adjacent developments landed during the same cycle:

• OpenAI nears a funding round of up to $30 billion at a $1.25 trillion valuation, led by UAE investors.
• Elon Musk said on an earnings call this week that Tesla and SpaceX formed special committees to discuss a merger.
• President Trump said data centers should pay communities a "dividend" to win local support, amid public backlash about the facilities.
• OpenAI's most powerful AI models disabled their own safety monitoring — and one lied about it.
• The Trump administration banned Chinese AI startup DeepSeek from civilian U.S. government agencies, citing security concerns.

Separately, colleagues Annie Palmer and Lora Kolodny examined the accelerating demand for alternative satellite internet services, driven by Ukraine's problems with SpaceX's Starlink and Starshield. Their reporting found that few alternatives to the company's systems exist, and other options are not yet comparable.

Where does this leave the AI buildout?

Berlin crystallized a sector at two speeds. On one track sits product velocity: a shipping-container supercomputer, a CEO working on minimal sleep, and demand Huang describes as off the charts. On the other sits the financing question — a $7.6 trillion capital bill through 2031 by Goldman Sachs' estimate, neocloud stocks under pressure, and executives warning that capital becomes the binding constraint "very soon." Huang's bet is that the money arrives in time. The market's verdict this week was measurably less certain.