The AI cloud market is quietly dividing over one unglamorous but revealing question: how long should a compute contract run? According to The Information, Nebius and CoreWeave are promoting short-term deals to customers, while Amazon Web Services is pushing in the opposite direction and locking clients into long commitments. That split says a lot about who’s confident, who’s hedging, and where the GPU business is heading.
What’s actually happening
The newer GPU specialists, often called neoclouds, are selling flexibility. Nebius and CoreWeave are dangling shorter contracts so customers can grab capacity without betting years of budget on a single provider. AWS, the incumbent, is doing what incumbents do. It’s using scale and a deep product stack to sign customers to longer terms.
The contrast is the story. In a market where GPU supply has been the constraint, whoever controls capacity usually sets the terms. The fact that some providers are now offering shorter, looser deals hints that the supply picture is loosening, at least in spots.
Why it matters now
Think about what each side is signaling.
- Short-term deals win customers who want optionality. Prices on AI compute keep moving, new chips keep landing, and nobody wants to be stuck paying yesterday’s rate on hardware that’s a generation behind.
- Long-term deals give AWS predictable revenue and give customers guaranteed access. For a company building a product that depends on steady compute, that certainty is worth something.
Here’s the tension. Short contracts are great for buyers until capacity gets tight again, at which point flexibility turns into scrambling for supply. Long contracts protect access but risk locking you into pricing and chips that age fast.
The competitive read
CoreWeave and Nebius are fighting the classic challenger’s battle. They can’t out-scale Amazon, so they compete on terms, speed, and specialization. Offering short deals is a wedge to pull customers away from hyperscalers who want to own the whole relationship.
AWS playing the long game makes sense too. Amazon doesn’t need to win on flexibility. It needs to win on lock-in, integration, and the promise that it’ll still be standing and stocked years from now. That’s a bet on staying power, and it’s a reasonable one.
This is the same pattern that shaped the early public cloud era, just compressed and running on far more expensive hardware.
Where this goes in the next 1-2 years
Expect the contract question to become a real negotiating lever, not a footnote. A few things to watch:
- Pricing pressure. More short-term options mean more price competition. That’s good for buyers and rough on providers carrying heavy debt to fund GPU buildouts.
- Supply swings. If capacity tightens again around a hot new chip, the flexibility advantage flips fast. Short-term buyers could get squeezed.
- Financial strain on neoclouds. Short contracts plus billions in hardware financing is a tricky combination. The providers touting flexibility are the same ones under the most scrutiny about how they’ll pay for it.
Practical takeaways
If you’re buying AI compute, the move isn’t to pick a camp. It’s to split your bets.
- Lock in a baseline of guaranteed capacity for workloads you know you’ll run.
- Keep a flexible layer on short-term deals for experiments and spiky demand.
- Read the supply signals. When providers start offering shorter terms, that’s usually your window to negotiate harder.
For everyone else watching the industry, contract length is now a tell. It shows you which providers think supply is loosening and which are betting on scarcity holding.
The GPU gold rush isn’t just about who has the most chips anymore. It’s about who structures the deal that survives the next price swing. The Information has more detail on how each provider is positioning, and it’s worth the read.