AI chips

South Korean chip designer Mobilint is leading a humanoid robot AI chip project in a ₩800bn state programme, with testing planned on LG Electronics robots.

Tesla and Nvidia will put 3.7tn won ($2.8bn) into Samsung Electro-Mechanics chip-substrate lines, Maeil reports. Neither Tesla nor SEMCO has confirmed it.

NVIDIA says six suppliers will expand substrate, packaging, testing and optics capacity, including a $1.1bn advance for KYEC's US test plants.

Qualcomm will cut jobs worldwide in a "performance-focused reduction", CEO Cristiano Amon told staff in a memo seen by Reuters. No headcount was given.

Google has open-sourced TorchTPU, its native PyTorch stack for TPUs built with Meta, and released llm-d 1.0, an open Kubernetes stack for AI inference.

ByteDance, Alibaba and Tencent ordered over 3mn Nvidia RTX PRO 5500 cards worth 5bn+, Reuters sources say. Nvidia warns it may not fill them.

doubleAI opened early access to Maxwell, a Code Execution Model that rewrites binaries as programs run. Speed-ups are company claims.

The PyTorch blog announces TorchNeuron, an open-source native backend for AWS Trainium. Code goes public on 20 Oct at PyTorch Conference; .to('neuron').

Jim Keller's Tenstorrent closed a $1.5B Series E at a $6.2B post-money valuation, led by Trace Capital and Innovation X, to scale Galaxy servers and IP.

SED: Nvidia asked Samsung, SK hynix and Micron to make NVHBM base dies on TSMC 3nm and standardise core dies on 1c DRAM. Unconfirmed; Samsung DS accepted.