The Amsterdam-based AI cloud firm signed a roughly $3 billion deal with Meta to supply high-performance GPU infrastructure over the next five years — its second multibillion-dollar hyperscaler contract in just two months.
The move cements Nebius as one of the most aggressive players in the so-called neocloud scene — a new class of infrastructure companies racing to keep up with insatiable demand for AI compute. But there’s a catch: this growth is burning cash fast.
Revenue rockets, but losses deepen
Nebius’ Q3 numbers read like a paradox.
Revenue surged 355% year-over-year to $146.1 million, but losses widened to over $100 million, up from about $39.7 million a year ago. The culprit? A massive ramp-up in capital spending.
The company poured nearly $955.5 million into data centers, GPUs, and land — over five times what it spent in the same quarter last year. It’s a “spend big or get left behind” playbook reminiscent of early-era AWS or Tesla Gigafactories.
“Nebius is basically betting that owning infrastructure at scale will decide who wins the next decade of AI,” one industry analyst told.
Meta joins the rush for GPU power
For Meta, the deal is as much about survival as scale. Its Llama and generative AI ambitions are bottlenecked by GPU supply — something giants like Microsoft and Amazon have also struggled with.
Nebius said it would deploy the required capacity for Meta within the next three months. Demand, however, is so intense that the contract was “limited by available capacity,” according to founder and CEO Arkady Volozh.
This follows Nebius’ $17.4 billion deal with Microsoft in September, marking the second hyperscaler partnership in under a quarter.
The neocloud moment
Nebius and rival CoreWeave are spearheading a movement reshaping how AI compute is distributed. Instead of building everything in-house, Big Tech is increasingly outsourcing GPU clusters to specialised players with faster deployment cycles.
That shift signals a structural change in cloud economics — one where agility and GPU access might trump scale.
“Nebius is effectively a GPU utility provider,” said one VC tracking the space. “They don’t need to out-Amazon Amazon. They just need to feed the AI boom.”
Can Nebius sustain the speed?
Nebius’ shareholder letter targets an annualised run-rate revenue of $7 billion to $9 billion by end-2026, up from about $551 million today. That’s a breathtaking pace — and it depends entirely on whether the firm can unlock enough capacity fast enough.
“The only real limitation on our revenue growth,” Volozh wrote, “has been the amount of capacity we’ve been able to bring online.”
Translation: the GPUs are the gold, and supply chains are the minefield.
Market cools on the heat
Despite the record growth and blockbuster Meta deal, Nebius’ stock slipped more than 3% in early trading. Investors appear wary of ballooning costs and long-term profitability.
Still, for now, Nebius sits at the center of the AI gold rush — the rare middleman that can sell shovels to both miners and magnates.
Conclusion
Nebius is on fire — in both growth and burn rate. Its $3 billion Meta contract signals rising demand for neocloud infrastructure, but the race to scale might be its toughest challenge yet.
If 2023–24 was the year of AI models, 2025 is clearly the year of AI infrastructure.