Every gold rush has a consistent pattern. The people who find gold rarely get rich. The ones who sell pickaxes, jeans, and shovels do. The AI gold rush is proving to be exactly the same — and the numbers are starting to prove it.
CoreWeave: The Quiet $66 Billion Backlog
CoreWeave is a cloud provider that was built from scratch for GPU-accelerated AI workloads. It doesn't try to be AWS — it specializes in the kind of compute that training and inference require.
The numbers are staggering: $66.8 billion in revenue backlog as of early 2026. The company generated $3.6 billion in revenue in the first nine months of 2025 alone — a 204% year-over-year increase. CoreWeave projects $12.1 billion in revenue for 2026.
In April 2026, CoreWeave signed a multi-year agreement with Anthropic to provide Nvidia GPU capacity for production-scale Claude inference workloads. That deal alone tells you everything about who holds the leverage in this market.
The company's physical infrastructure is expanding at an extraordinary pace: 850 megawatts across 43 data centers by end of 2025, with projections to exceed 1.7 gigawatts by end of 2026. That's enough power to run a small city.
Scale AI: $2 Billion in Annual Revenue
Every AI model is only as good as the data it was trained on. Scale AI provides the data labeling, curation, and annotation infrastructure that frontier model companies depend on.
Revenue hit approximately $2 billion in 2025, doubling year-over-year from $870 million in 2024. In June 2025 alone: Meta purchased a 49% non-voting stake for $14.8 billion, and Scale AI closed a $14.3 billion funding round. Total funding: $15.9 billion. Current valuation: $29 billion.
The Meta deal came with a twist — Google and OpenAI cut ties with Scale AI mid-year as a result, showing the delicate political balance of being the industry's data provider.
The Infrastructure Stack
Think of AI infrastructure in layers:
- Base: NVIDIA chips, cloud providers (AWS, Google Cloud, Azure), and specialists like CoreWeave
- Data: Labeling, curation, synthetic data generation — Scale AI's territory
- Tooling: Evaluation frameworks, observability, fine-tuning platforms
The counterintuitive insight: profit margins at the base and middle layers are often higher than at the application layer. Building a new AI app is competitive. Providing the infrastructure that every AI app needs is a durable business.
What This Means for Enterprises
When you're making infrastructure decisions — which cloud to use, which data providers to trust, which tooling to build on — you're not just making a technical decision. You're making a strategic bet on who the ecosystem winners will be.
The companies that figured out how to partner with CoreWeave and Scale AI early have a significant advantage. The ones still evaluating are finding that the best providers are increasingly booked out, priced up, or aligned with competitors.
