Two smaller cloud providers, Runpod and Nebius, are picking up business from AI startups frustrated by limited access to graphics processing units on Amazon Web Services. The shift highlights how resource constraints at the biggest cloud platforms are opening doors for competitors.
Why the GPU crunch matters
Training and running large language models requires massive amounts of GPU compute. AWS, the dominant cloud provider, has struggled to keep up with demand as AI startups and enterprises alike rush to deploy generative AI. The shortages have led to long wait times and allocation limits for some customers, especially those without long-term contracts or reserved capacity.
Runpod and Nebius, both smaller players, have been able to offer more immediate access to high-end GPUs like Nvidia's H100s. For startups racing to get products to market, that speed can be a decisive factor.
What Runpod and Nebius are offering
Runpod, which started as a platform for renting GPU power for AI training, has been expanding its customer base among early-stage AI companies. Nebius, a cloud infrastructure provider spun out of the Russian internet company Yandex, has also been targeting the same segment. Both companies emphasize flexibility and lower costs compared to the big three cloud providers.
Neither company has disclosed exact pricing or customer numbers, but the trend is clear: AI startups are increasingly willing to try non-traditional cloud services when the incumbents can't deliver.
What this means for the cloud market
The GPU shortage is not just a temporary hiccup. Analysts expect demand for AI compute to keep growing for years. If Runpod and Nebius can maintain reliable service and scale up, they could become permanent fixtures in the AI infrastructure landscape. That would break the near-monopoly that AWS, Microsoft Azure, and Google Cloud have held over cloud computing for the past decade.
For now, the big players are not standing still. AWS has announced plans to invest more in its own custom AI chips and expand data center capacity. But those investments take time. In the meantime, startups need GPUs today.
The question is whether Runpod and Nebius can hold onto the customers they're winning now once the supply constraints ease. That will depend on how well they perform under pressure and whether they can build the kind of trust that keeps startups from jumping back to AWS the moment a new instance becomes available.
