Major cloud providers such as "Microsoft", "Amazon" and "CoreWeave" have reserved graphics processing unit (GPU) capacities primarily for "Anthropic", "OpenAI" and their own internal projects. This situation has caused small and medium-sized artificial intelligence startups to remain in queues for several months. 3dnews.ru reports on this.
Due to the shortage of computing resources, rental rates have increased by more than 25 percent over the last six months. According to "Azure" data, the chip deficit is expected to continue until at least the end of 2026.
Currently, many startups cite the GPU shortage as the primary obstacle to their development.
A three-tier hierarchy for GPU allocation has been implemented in the "Microsoft" system, in which large customers have priority. To rent the new "Blackwell" chips, a customer must order at least 1,000 units and sign a long-term contract worth tens of millions of dollars.
As the market situation worsens, some startups are moving toward building their own infrastructure. For instance, the company "Collide" announced it would invest $500,000 to create its own private computing cluster to avoid dependency on external providers.
Read “Zamin” on Telegram!
