Aethir signs B300 multi-year contracts

GPU cloud used to be sold like burst capacity — grab some H100s, run a training job, move on. This story is different. Aethir is now tied to a multi-year, nine-figure enterprise contract built around NVIDIA’s new B300 chips, and that changes the signal. It says at least some AI customers are no longer treating compute as a short-term rental problem. They’re treating it like core infrastructure. ### What actually got signed? The concrete deal is not just “Aethir signed contracts.” Axe Compute said on April 22 that it closed a $260 million multi-year enterprise GPU contract powered by Aethir, centered on a 2,304-GPU NVIDIA B300 cluster. Aethir framed it as one of the biggest enterprise compute deals yet in decentralized GPU infrastructure, which is the part that makes this more than routine cloud sales copy. ### Why does the B300 part matter? Because B300 is not old inventory being repackaged. NVIDIA’s DGX B300 uses eight B300 Tensor Core GPUs built on Blackwell Ultra, and NVIDIA is pitching it for modern AI factories — especially heavy training, inference, and reasoning workloads. NVIDIA says DGX B300 delivers 1.5x dense FP4 performance and 2x attention performance over DGX B200, so a buyer locking in thousands of these is buying into the next compute generation, not just filling a temporary gap. (ecosystem.aethir.com) ### What did Aethir say before this? Aethir had already been teeing this up. In March, it said it was the first decentralized AI infrastructure network deploying B300 GPUs across multiple regions, with the hardware aimed at advanced training, multimodal models, and agent workloads. So the April contract matters because it looks like the first big proof point after that rollout claim — not just a roadmap slide, but an actual customer commitment measured in years and hundreds of millions. (nvidia.com) ### Why is “multi-year” the real story? Because multi-year compute deals shift the risk around. If you are a startup or enterprise customer, you usually worry that GPU prices spike, supply disappears, or your vendor reprices you once your model starts working. A longer contract can smooth that out. But it also means you are making a view on future demand, model economics, and financing. Basically, compute starts to look less like on-demand cloud and more like leased industrial capacity. (blog.aethir.com) ### Why does this matter for decentralized cloud? Decentralized GPU cloud has often been sold as a cheaper or more flexible alternative to hyperscalers. The knock on it was always whether serious enterprises would trust it for production-scale workloads. A 2,304-GPU B300 cluster tied to a multi-year contract is a direct answer to that doubt — at least for one customer. It suggests decentralized supply can be packaged, financed, and sold in a way that looks a lot more like mainstream infrastructure procurement. (ecosystem.aethir.com) ### What’s this about hedging compute prices? This is the part that makes the story feel early but important. On May 12, CME Group and Silicon Data said they plan to launch compute futures later in 2026, pending regulatory review. If that market actually develops, buyers and sellers of GPU capacity could hedge price swings the way airlines hedge fuel or chip buyers hedge commodities. That would make multi-year GPU contracts easier to price — and easier to finance. (ecosystem.aethir.com) ### So what changes now? The immediate change is not that every startup suddenly signs a three-year B300 reservation. The change is that the menu of options is widening. You can rent compute by the hour, reserve it in bulk, or maybe soon hedge it financially too. Once that happens, GPU infrastructure stops being just a technical bottleneck and becomes a treasury problem. ### Bottom line? Aethir’s B300 contract signal matters because it points to a new phase — GPU compute getting locked up, financed, and eventually hedged like serious industrial capacity rather than rented like spare cloud time. (investor.cmegroup.com)