📰 Key Takeaways

On its Q3 earnings call Thursday, Nebius announced planned 2027 spending of $40 billion, well above Wall Street expectations. The money comes mainly from customer prepayments and loans backed by customer contracts, essentially using the future value of its data centers to pay for building them. That approach gets riskier if projects slip or get cancelled, and CEO Arkady Volozh admitted: “Building at this scale is hard, delays are the industry reality, and we’re no exception.”

The biggest worry is the Microsoft contract. It’s worth up to $19.4 billion and covers AI compute capacity at the Vineland, New Jersey data center, but it’s already months behind. Only a bit more than half is expected to be delivered by year-end, with the rest pushed to Q2 2027, and those batches were originally scheduled to be deployed in 2025 through 2026. According to Nebius’s contract summary, if delivery is late and no substitute capacity is offered, Microsoft can terminate the affected services after a grace period, and the contract includes financial penalties for late delivery. Nebius says Microsoft “took part in the planning process,” but it hasn’t said whether Microsoft agreed to the new dates, and Microsoft hasn’t publicly cancelled anything.

The direct cause of the delay is that DataOne, the third-party operator of the site, shut down its temporary power generation earlier than planned and switched to Bloom Energy fuel cells. But it wasn’t entirely voluntary: in August, investigative outlet Floodlight found at least 45 natural gas generators running at the site without permits, and New Jersey regulators fined the operator $1.1 million for violating federal law. DataOne also faces a lawsuit from nearby residents over the constant humming noise.

There are also reports that Microsoft has become more cautious about neocloud partners that can’t deliver on time, and Nebius’s financing leans more and more on customer repayments. If capacity doesn’t come online on schedule, costs keep piling up while revenue gets pushed back. See the original article for the full details.

💬 JudyAI Lab Take

Nebius is raising 2027 capex to $40 billion, funded by customer prepayments and contract-backed loans, while its biggest customer, Microsoft, has a $19.4 billion contract that’s running into delivery delays. This story shows the financial structure behind the AI compute buildout.

For AI builders, this case shows that compute supply isn’t just a question of whether you can get GPUs. The Vineland delay traces back to a problem with the site operator’s temporary power setup, which led to a $1.1 million fine and a residents’ lawsuit, and ultimately hit contract delivery. When financing leans more on customer repayments, a one-quarter slip in capacity means costs pile up first while revenue gets pushed back. The late-delivery penalties and termination clauses also mean Microsoft holds the upper hand in this kind of deal. In other words, a compute provider’s reliability depends on very physical things like power, sites, permits, and funding, not just technical ability.

If your product depends heavily on a single cloud or compute provider, start by checking the delivery deadlines and fallback options in your contract, and think about whether you could switch to a second source if you had to.

📅 Source Info

🔗 Further Reading