Your last hardware decision shouldn’t decide what you build next.

AWS launches TorchNeuron, enabling developers to run existing PyTorch code natively on AWS Trainium AI chips without major code rewrites.

• PyTorch developers can try Trainium with configuration changes instead of code rewrites
• Hugging Face Transformers and popular libraries now support Trainium directly
• Deepgram achieved 60% better price-performance moving speech models to Trainium
• Teams can choose hardware per application based on price, performance, and availability
• Standard PyTorch checkpoints work across different hardware, preserving flexibility
• 100 most-used text-generation model architectures work out of the box

TorchNeuron reduces switching costs and empowers teams to select optimal hardware for each workload without being locked into initial infrastructure decisions.