AWS adds TorchNeuron to move PyTorch apps onto Trainium

AWS adds TorchNeuron to move PyTorch apps onto Trainium

2 min read•Oct 20, 2026•
Alex Thornton
Alex Thornton

Amazon Web Services released TorchNeuron, a tool that lets AI startups move PyTorch apps onto its Trainium chip, first reported by TechCrunch. That gives startups a simpler route to Amazon’s chip, while putting the tool against Nvidia’s software advantage, which has helped make Nvidia hardware a default choice for AI apps.

TorchNeuron works with PyTorch, the open-source toolkit developers use to build, train and run AI models. Amazon says an app can run on Trainium after one line of configuration, with no code rewrite; because the tool is adapted from PyTorch, developers who know the framework already know how to use it.

Nvidia has spent years ensuring its CUDA software supports AI apps written with PyTorch on Nvidia GPUs and Nvidia’s other hardware. Amazon says about 85% of AI apps are written with PyTorch, and Nvidia hardware is now deeply embedded in the toolkit, essentially by default.

Amazon is also contributing changes upstream to the main PyTorch project, working with the PyTorch community to remove code that ties the framework to specific chips. The stated goal is to make PyTorch more hardware-agnostic for the developer community, not just Trainium users.

AWS argues its cloud gives startups infrastructure to move from prototype to production, rather than relying on a single-purpose compute provider and building everything else around it. The company says its cloud will save startups money even if AWS does not offer the cheapest per-token price, because Trainium handles training and inference on the same chip and is tightly integrated with the infrastructure startups need.

Colin Brace, VP of Technology Software at AWS Annapurna Labs, cited Deepgram as processing 70% more audio per dollar at production-grade accuracy after moving to Trainium with native PyTorch support.

The two efforts work at different levels: TorchNeuron changes how PyTorch apps reach Trainium, while the upstream contribution targets code that ties PyTorch to specific chips. The open question is whether loosening those Nvidia ties gives developers more choice and more affordable options, as Amazon promises, or brings more configuration work.

Plug-in vehicles reach 32% of global registrations in September

Plug-in vehicles reached 32% of global registrations in September, first reported by Cleantechnica. Battery-electric vehicles accounted for 23% of registrations and set a record at 1.54 million, up 8% year over year, while plug-in hybrids fell 9%; year to date in 2026, BEVs were up 12% and PHEVs down 10%.

Global plug-in registrations rose 3% year over year to around 2.16 million in September, close to the all-time record of 2.17 million set last December.

Year to date, plug-in registrations were up 4% year over year: BEVs grew 12%, while PHEVs fell 10%. Their combined share for 2026 stood at 25%, with BEVs at 18% and PHEVs at 7%. The overall growth rate therefore hides a clear split between the two categories.

The Tesla Model Y led the model rankings with a record 165,000 registrations, up 17% year over year. Tesla Model 3 came second with 65,000, enough to move ahead of Geely Xingyuan and BYD Song despite a 4% annual decline.

BYD Song finished fourth at 54,519 registrations, up 19% and the fastest growth rate among the top five. An export push and the new-generation Ultra body helped lift the model.

At the make level, Tesla had a record 232,000 registrations, up 9% year over year. Toyota reached 61,332 for its fourth record in a row, finishing ahead of Volkswagen’s year-best 58,000 to become September’s best-selling legacy make.