OpenAI, 구매자들이 Nvidia 대안을 시험하는 가운데 AMD Cloud 용량을 임차할 예정이다.
OpenAI announced on Monday, Nov. 16, that it would rent cloud computing capacity from TensorWave, a startup that runs only Advanced Micro Devices (AMD) chips. OpenAI, Anthropic and Meta have signaled plans to use AMD chips since late 2025. TensorWave projects that its annual recurring revenue will rise from $125 million at the end of 2025 to $210 million by the end of 2026.
Superlinked decided to deploy AMD chips after several months of testing. CEO Daniel Svonava said the chips suit small open-source models, such as those from DeepSeek, and that their lower cost can improve margins.
Google agreed in July to guarantee up to $44 billion in data center leases against tenant defaults. Cirrascale has become one of the first providers outside Google Cloud to offer computing capacity on Google's Tensor Processing Units (TPUs). Autolane, a delivery platform, has shifted some evaluation work from graphics processing units (GPUs) to lower-cost TPUs since September, according to co-founder Chad Agate.
Use of TPUs at scale remains limited among customers of DoiT. CEO Vadim Solovey said 16 of the cloud software company's roughly 5,000 customers use TPUs at that level. He characterized the chips as an additional supplier rather than a replacement.
Nvidia's third-quarter revenue was more than eight times AMD's in the same period. Megaport CEO Michael Reid said some customers who asked about AMD after failing to secure Nvidia chips dropped the plan when they faced a six-month wait for AMD, compared with 12 to 14 weeks for Nvidia. Reid also described Nvidia's business and marketing support as a competitive strength, saying it helps connect customers with financing contacts and data center resources.