There’s a new category of cloud company that didn’t exist three years ago. They don’t sell general-purpose compute. They don’t compete with AWS on storage or databases. They do one thing: rent GPUs to AI companies at massive scale.

They’re called neoclouds, and they’re growing at rates that make traditional cloud providers look sluggish.

CoreWeave went from a crypto mining operation to a $12B revenue business. Nebius is targeting $7-9B in annualized revenue by end of 2026. Lambda just raised $1.5B to build “AI factories.” And they’re all backed by the same customer: companies that need GPUs faster than Nvidia can ship them.

What Is a Neocloud?

Traditional cloud providers - AWS, Azure, GCP - are general-purpose platforms. They offer compute, storage, databases, networking, and hundreds of managed services. AI workloads are one category among many.

Neoclouds are different. Their entire infrastructure is optimized for GPU-heavy AI workloads:

  • GPU-first architecture - clusters designed for training and inference, not web servers
  • High-bandwidth networking - InfiniBand or NVLink fabrics that traditional clouds don’t offer
  • Faster provisioning - spin up thousands of GPUs in minutes, not weeks
  • Simpler pricing - pay per GPU-hour, no complex SKU maze

The trade-off: they can’t host your React app or run your PostgreSQL database. They’re specialists, not generalists. And in AI, specialization wins.

The Big Three

CoreWeave (CRWV)

Origin story: Founded as a crypto mining company. When the crypto market crashed, they pivoted all their GPU capacity to AI cloud computing. It might be the most successful pivot in recent tech history.

The numbers:

  • Revenue: ~$5B in 2025 (300% YoY growth), projected $12B in 2026, ~$20B by 2027
  • IPO: Nasdaq March 2025, raised $1.5B at $40/share. Stock jumped 80% - biggest tech IPO since 2021
  • GPUs: 45,000+ GPUs - largest private GPU provider in North America
  • Backlog: $55.6B in contracted revenue, providing multi-year visibility
  • Key deals: $6.3B Nvidia deal (buying unused cloud capacity through 2032), $14B Meta deal

The bull case: CoreWeave is the “AWS of AI.” They have the largest non-hyperscaler GPU fleet, long-term contracts with Nvidia itself (Nvidia is both their supplier and customer), and revenue growing at triple-digit rates.

The bear case: $10B+ in debt with $1.2B in annual interest expenses. Non-GAAP gross margins are 65%, but the company is still net-loss making. If AI demand slows or GPU prices drop, the debt becomes crushing. This is a leveraged bet on AI demand never declining.

Nebius (NBIS)

Origin story: Spun out of Yandex (Russia’s Google) in 2024. The team brought deep infrastructure expertise from building one of the world’s largest search engines and pivoted to AI cloud.

The numbers:

  • Revenue: $1.25B ARR at end of 2025, targeting $7-9B annualized run rate by end of 2026
  • Growth: 521% to 900% implied growth - the fastest in the entire AI infrastructure space
  • Stock: Up 160%+ in 2025
  • Key deals: $17.4-19.4B Microsoft deal (5 years), $3B Meta contract
  • Data centers: Scaling from 2 (2024) to 16 by end of 2026, across US and Europe
  • Power: Targeting 3+ GW of contracted power capacity

The bull case: Nebius has the most aggressive growth trajectory. The Microsoft and Meta contracts provide revenue visibility that most startups would kill for. Average selling prices increased 50%+ in Q4 2025, showing pricing power. Analysts see 84% upside from current levels.

The bear case: Execution risk is enormous. Going from 2 data centers to 16 in two years while growing revenue 7x requires flawless operational execution. The Yandex heritage raises geopolitical questions for some investors, despite the full corporate separation.

Lambda (Private, potential 2026 IPO)

Origin story: Started in 2012 as a deep learning research tool company. Pivoted to GPU cloud when they realized researchers needed hardware more than software.

The numbers:

  • Funding: $2.44B raised, $4B valuation (Series E, November 2025)
  • Latest round: $1.5B led by TWG Global
  • Infrastructure: Building “AI factories” - Kansas City facility launching early 2026 with 10,000+ Blackwell Ultra GPUs
  • Key deals: Multi-billion dollar, multi-year Microsoft partnership

The bull case: Lambda is the most developer-friendly neocloud. Their platform is designed for researchers and AI teams who want raw GPU access without cloud complexity. The Microsoft partnership validates the model and could pave the way for an IPO in 2026, following CoreWeave’s playbook.

The bear case: Still private, so financials are opaque. Smaller scale than CoreWeave and Nebius. The GPU cloud market is getting crowded.

How They Compare

Metric CoreWeave Nebius Lambda
Status Public (CRWV) Public (NBIS) Private ($4B valuation)
2025 Revenue ~$5B ~$1.25B ARR Undisclosed
2026 Target ~$12B $7-9B ARR N/A
Growth rate ~134% 521-900% N/A
GPU fleet 45,000+ Expanding rapidly 10,000+ (new facility)
Biggest contract $14B Meta $19.4B Microsoft Multi-B Microsoft
Revenue backlog $55.6B Multi-year contracts N/A
Key differentiator Scale + Nvidia relationship Growth speed + Microsoft/Meta deals Developer experience

Why Neoclouds Exist (And Why AWS Isn’t Enough)

You might wonder: why don’t AI companies just use AWS or Azure? Three reasons:

1. Availability. Getting 10,000 GPUs on AWS can take months. CoreWeave can provision them in minutes. When you’re in an AI arms race and every week of training time matters, availability is everything.

2. Networking. AI training requires GPUs to communicate constantly. Traditional cloud networking wasn’t designed for this. Neoclouds build InfiniBand and NVLink fabrics specifically optimized for GPU-to-GPU communication. The bandwidth difference is 5-10x.

3. Cost. Neoclouds don’t carry the overhead of running 200+ cloud services. They run GPUs. That specialization translates to 20-40% lower costs for equivalent GPU time compared to hyperscaler cloud.

The hyperscalers know this. That’s why Microsoft has deals with both Nebius and Lambda - they’re outsourcing some GPU capacity to specialists rather than building everything themselves.

The Risks

Debt. CoreWeave has $10B+ in debt. Building GPU data centers requires massive upfront capital. If AI demand plateaus or GPU prices drop significantly, the debt burden could become unsustainable.

Customer concentration. Nebius gets a huge chunk of revenue from Microsoft and Meta. CoreWeave’s biggest deals are with Nvidia and Meta. Losing a single customer would be devastating.

GPU supply. All three companies depend on Nvidia shipping GPUs on time. TSMC’s CoWoS packaging capacity is already sold out for 2026. Any supply chain disruption hits neoclouds harder than hyperscalers who have more diversified infrastructure.

The hyperscaler response. AWS, Azure, and GCP are investing heavily in AI-optimized infrastructure. Azure already runs massive GPU clusters. If hyperscalers close the availability and networking gap, the neocloud value proposition weakens.

Commoditization. As more GPU cloud providers enter the market, GPU rental becomes a commodity. Margins compress. The company with the lowest cost of capital (hyperscalers) wins a commodity war.

What It Means for Investors

Neoclouds are the highest-growth segment in all of tech right now. Triple-digit revenue growth, massive backlogs, and direct exposure to the AI infrastructure buildout.

But they’re also highly leveraged bets on a single thesis: that AI compute demand will keep growing at the current pace for years. If it does, these companies are dramatically undervalued. If it doesn’t - if AI spending slows, if GPU prices drop, if hyperscalers catch up - the downside is severe.

The smart framing: neoclouds are the picks-and-shovels play on the AI gold rush, but with significant debt and concentration risk. They’re not safe investments. They’re high-conviction bets on the most important trend in technology.

And if you believe AI infrastructure spending will hit $1 trillion by 2030 (as multiple forecasts suggest), then CoreWeave at $12B revenue and Nebius at $7-9B are still in the early innings.