HOW A GOOGLE REFUGEE BECAME AI’S MOST PROMISING INFRASTRUCTURE PLAYER

(NBIS), (MSFT), (META), (NVDA), (CRWV), (UBER)

If you blinked, you probably missed Nebius (NBIS) go from tech minnow to headline-snagging AI infrastructure juggernaut. In Q3 ’25, they clocked $146 million in revenue, up a dizzying 355% year-over-year. 

Wall Street yawned because it fell short of a $155 million consensus. But if you’re playing this game for single-quarter beats, you’re not thinking big enough.

But what really matters here isn’t the headline miss, but the velocity. 

The real story is a small European firm muscling its way into the AI arms race and landing deals with both Microsoft (MSFT) and Meta (META). In the current gold rush for AI compute power, that’s like being handed pickaxes by the two biggest miners in the business.

Let’s not mince numbers. Wall Street wants $578 million in full-year 2025 sales. Q3 gives us $146 million. That implies Nebius needs to conjure up $276 million in Q4, nearly doubling quarter-over-quarter. Normally, that’d be a moonshot. But these are not normal times. 

The new London facility just turned on 4000 of NVIDIA’s (NVDA) B300 chips. A fresh Israeli site is about to go live. And the Kansas expansion just brought NVIDIA’s B200s into the fray. This is a company hurtling forward, with demand outpacing supply.

But here’s the crux: demand isn’t the bottleneck – supply is. Specifically, power. AI data centers don’t run on dreams and VC decks; they run on megawatts. And as anyone in the game knows, securing cheap, scalable energy is the unsexy lynchpin of this entire boom. 

If Nebius locks in power at inflated rates, it’s game over before it really begins. But if they can source cheap juice, especially near Tier 2 U.S. grids, they might just become a thorn in CoreWeave’s (CRWV) side.

Speaking of which, CoreWeave still holds the U.S. pole position, trading at a bloated $54 billion to Nebius’s comparatively svelte $27.5 billion. Yet Nebius just landed a Microsoft contract worth $17.5 billion – yes, nearly their entire market cap at the time. 

That’s a backroom handshake most start-ups would sell their cap tables for. It validates the model, the tech stack, and the leadership.

The CEO, an old Yandex hand, isn’t a stranger to big-boy projects. Scaling infrastructure isn’t his first rodeo. In Q3 alone, CapEx exploded to $955 million. That’s no rounding error. And yet, they raised $4.3 billion in equity and convertibles without breaking a sweat. 

As of September, they’re sitting on $4.8 billion in cash. The market is happy to bankroll the shovel sellers of this AI boom…for now.

Of course, profitability is another beast entirely. They flaunt 70.5% gross margins, but depreciation isn’t broken out clearly. 

When you back in $99 million in depreciation, margins might nosedive to low single digits. If you want a sobering metric, there’s your wet towel. 

Still, the comparison with CoreWeave, which books depreciation under GAAP, suggests margins north of 70% aren’t just smoke and mirrors in this sector. There’s real operating leverage here.

The wildcard is Avride, Nebius’s autonomous vehicle moonshot. They just teamed up with Uber (UBER), jointly injecting $375 million into the venture. It’s flashy, it diversifies the portfolio, and it adds a layer of optionality. 

But make no mistake: it’s also a capital hog. Expect Nebius to quietly reduce its stake over time and refocus on the data center meat and potatoes.

Zooming out, there’s a bigger risk that rarely gets airtime: whether the LLM boom, which underpins this entire infrastructure frenzy, can actually sustain itself. 

If OpenAI, Anthropic, or the next shiny unicorn runs out of funding, training slows, and Nebius’s pipelines could dry up. But we’re not there yet. 

For now, hyperscalers are gorging on compute like kids at a candy store. And Nebius is one of the few vendors still building shelves.

Is it priced to perfection? Hardly. Is it risky? Undeniably. But show me a sub-$30 billion company with Microsoft and Meta contracts and a $7-9 billion run rate by 2026, and I’ll show you something rare in this market: an asymmetrical bet where the upside hasn’t been fully priced in – yet.