AWS releases TorchNeuron to move PyTorch AI apps onto Trainium
Amazon says the tool needs only a configuration change and plans to contribute changes that make PyTorch less dependent on specific chips.
Modal plans to add Nvidia processors alongside Intel and AMD chips
The cloud startup said early tests of Nvidia's Vera chip doubled AI agents' work output, though adopting the design will require software changes.
Cloud-computing startup Modal said on October 19 that it plans to use Nvidia's Vera central processor to host AI agents, adding a supplier for workloads it has so far handled exclusively with Intel and AMD chips.
Modal rents computing capacity to AI developers. It said preliminary testing found that Vera doubled the work completed by agents, which perform extended tasks for users. Supporting Nvidia's processor design will require substantial changes to Modal's software.
Chief Executive Erik Bernhardsson said Modal would retain its existing suppliers. The company plans to operate Nvidia, Intel and AMD hardware together, assigning workloads according to performance. Bernhardsson said the comparisons used Intel and AMD processors currently available; both rivals are developing newer chips that could reduce Nvidia's advantage.
Modal raised $750 million at a valuation of $15.75 billion in a round led by Accel. The startup said it was on track for $1 billion in sales over the next year. It also reported that service usage had tripled and revenue had increased more than 15 times year-over-year.
Nvidia has told investors it anticipates at least $20 billion in Vera sales this fiscal year and estimates the central processor market at roughly $200 billion. The company has also said that OpenAI, Anthropic, Oracle and Perplexity plan to use the processors.