Nebius Stock: An NVIDIA Training Pact, a 600 MW Rotterdam Letter, and the Contract That Hasn't Been Signed

Nebius shares fell 12% on Friday, caught in a broad selloff of technology and AI infrastructure names driven by worries over the cost of building data centers and the debt used to pay for them. The move says little about anything Nebius disclosed that day. What it does say is that investors are now grading the entire sector on a single question: can planned capacity be built on financing that holds up?

The distinction matters. A sector-wide wave of selling is not proof that every company caught in it carries the same problems. It does show that shareholders are repricing similar business risks in unison, and Nebius has not been able to step out of that shadow. AP reported widespread losses across tech stocks and concerns about rising financing costs; Reuters had flagged the previous Wednesday that debt-laden AI infrastructure providers were already under pressure as rate expectations climbed. No fresh funding announcement from Nebius triggered the drop.

Gartner's Trillion-Dollar Warning

Gartner added fuel to the skepticism on Wednesday. The research firm cautioned that more than $3.5 trillion in AI infrastructure spending could be exposed to impairment risk by the end of 2030, and that over 80% of neocloud providers might cease operating as standalone businesses. These are projections, not realized losses, and certainly not a verdict on Nebius specifically. As a warning about potential misallocation of capital, though, they carry weight. Dismissing the share price decline as undifferentiated nerves would be wrong. So would reading a sector forecast as one company's fate.

Rotterdam: Intent, Not Capacity

The Rotterdam project shows how tightly growth ambitions and execution risk sit together. According to Het Financieele Dagblad on October 12, Nebius signed a letter of intent to take a "significant portion" of the planned capacity at the Rotterdam Maasvlakte AI data center, a project designed for 600 megawatts. What the company has not done is sign a binding offtake agreement. Construction of the project itself remains uncertain, the report noted.

That gap is the whole story. The letter is evidence of appetite for expansion, not proof of capacity in hand. For investors, the difference grows more important as the financing of large infrastructure projects comes under sharper scrutiny. A final contract could sharpen the picture, but it would not by itself resolve every construction and funding question attached to the project. Those steps deserve separate assessment. Should the project fail to materialize, this particular expansion option simply disappears.

An NVIDIA Classroom Is Not a Data Center

On the operational side, Nebius Academy and NVIDIA announced an expanded collaboration on Friday covering hands-on AI training for developers. The arrangement includes jointly developed courses, offerings from the NVIDIA Deep Learning Institute hosted on the Nebius Academy, and workshops built around Nebius infrastructure, with sessions planned at NVIDIA GTC Berlin.

It is a concrete broadening of the education offering, and it can put the platform in front of the developers who will use it. As a signal for the stock, though, it belongs in the operational column: the subject is training and platform adoption, not the financing of a data center. Extrapolating added revenue or assured economic returns from it would go too far. A more convincing positive case would need several things to line up — the training partnership delivered as announced, Rotterdam reaching firmer contractual stages, and financing fears in the market easing. Only that combination would bring operating prospects and the risk question closer together.

Insider Buying on Autopilot

Director Arne Grimme's share purchases deserve the same sober treatment. An SEC filing documents buys on Thursday and Friday, executed under a Rule 10b5-1 trading plan adopted on June 15. That structure makes it hard to read the purchases as a spontaneous response to the recent selloff. Their value lies precisely in that limitation: personal buying is worth noticing, but a pre-set schedule is a different thing from a snap decision made as market conditions shifted. Media reports on October 12 also pointed to disclosed insider sales, including a sale by Marc Boroditsky executed under a pre-arranged trading plan — which limits how far it can be read as an immediate judgment on the business. Neither direction of insider activity substitutes for examining project contracts or assessing how they are financed.

What Would Actually Change the Story

The bear case needs no new bad company news. If investors stay skeptical of debt-financed data centers, even confirmed partnerships may do little to shift how the risk is priced, and the market will fixate on capital needs and deliverability rather than the number of announced alliances. The bull case starts with a calmer funding backdrop, at which point company-specific progress can carry more weight again.

The next concrete catalyst, then, is a decision on the final offtake contract with Volt. A signed deal would remove a significant caveat; assured execution of the construction would still need to be judged on its own. The bar for a convincing re-rating sits higher than the bar for a headline. For now, deliverability of growth outweighs the signal value of new announcements — and the 12% slide is no proof that Nebius has failed. It is a reminder that the market has moved from announced possibilities to binding commitments, and is waiting to see which ones arrive.