Nvidia GPU Supply Crisis Forces Cloud Providers to Ration Capacity
Claire Weston
Nvidia's GPU shortage has spread across the cloud computing ecosystem, with emerging cloud providers seeing four to five companies competing for every chip; the supply-demand imbalance is reshaping pricing power and startup survival across the compute market.
How severe is the GPU shortage?
An Nvidia sales rep advised a small AI startup to lease idle GPUs from a Qatari government-linked entity — a sign the scramble has reached sovereign channels.
Emerging cloud provider Nebius — a GPU leasing platform — says four to five companies are competing for every Nvidia AI chip in its facilities.
This means → the gap is no longer about waiting weeks in a queue — it is about having money and still not getting chips.
Who decides which customers get GPUs?
Nebius holds three capacity-allocation meetings per week, screening which clients receive resources.
Americas GM Dan Lawrence: "We meet three times a week to discuss which customers come in and which ones we want to pick."
In plain terms = GPUs are not first-come-first-served. The supplier picks the customer, not the other way around.
Do higher prices cool demand?
Nebius Chief Revenue Officer Marc Boroditsky told analysts in May that the company had already raised prices.
Even at higher prices, every chip type remains fully sold out.
This means → the GPU market is in a seller-pricing phase. Price hikes do not suppress demand — the shortage far exceeds what price elasticity can absorb.
How are startups coping?
AI startup The Biological Computing Company originally contracted with Nebius but switched to Amazon Web Services (AWS) over the past month or two — Nebius could not supply the capacity it needed, and AWS had cut GPU rental prices.
This reflects a bind: smaller platforms offer better pricing but lack inventory; hyperscalers have inventory but set their own terms.
In plain terms = startups are forced to hop between cloud platforms — whoever can deliver chips wins the contract.
What is Nebius's survival strategy?
Roughly 75% of its customers are companies that have maxed out traditional cloud servers; the other 25% migrate all workloads to Nebius.
The client base clusters around AI model developers, robotics firms, and code-generation tool makers.
Nebius does not compete on low prices. It actively screens for startups "with sufficient funding to keep scaling." Lawrence: "We don't see ourselves as a low-cost provider."
This means → Nebius's logic is to trade scarce resources for high-quality customers, not to chase volume.
Where does this shortage ultimately point?
Whether the GPU supply bottleneck eases in the near term will directly determine who holds pricing power in the compute-cloud market.
For startups, the cost of securing compute is a more immediate survival question than model capability.
This reflects a broader shift: AI's core bottleneck has moved from "are models good enough" to "is there enough compute, and can you afford it."
Content is for reference only, not financial advice.