Modal will move its Sandboxes product, used to run AI agents, onto Nvidia's Arm-based Vera CPU, its first departure from x86. Its own tests showed Vera handling up to twice as many container tasks as an Intel server, though the results have not been independently verified.
Tenstorrent, which positions itself as a challenger to Nvidia, raised $1.5 billion and is now valued at $6.2 billion. The company chose to stay independent after attracting acquisition interest earlier this year.
After the cost per run of its AI code reviewer doubled, Cloudflare analysed its token usage. Prompt caching and routing requests to the right model cut its inference spending by 68% and made reviews faster.
German Chancellor Friedrich Merz and NVIDIA CEO Jensen Huang discussed Germany's AI roadmap at an event in a former Berlin power station, with a focus on industrial uses. Huang pointed to NVIDIA's long ties with German industry, from gaming and workstations to the car sector.
Amazon has launched TorchNeuron, an open-source backend that lets PyTorch code run on its Trainium AI chips with a one-line change. It enters public beta on October 20 under the Apache 2.0 license, supports Trainium2 and Trainium3, and does not yet cover every PyTorch feature.
Intel has confirmed that its future Razer Lake-AX chips will deliver almost four times the AI performance of its current products. The goal is to run large language models and, above all, to power the next generation of humanoid robots.