nvidia

Cerebras unveiled its CS-4 rack-scale inference system, claiming 30x faster performance than GPUs, though independent analysis finds the chip inside is an overclocked WSE-3, not a new design.

JetPack 7.2.1 adds T3000 emulation on the Jetson Thor AGX Developer Kit plus new automated video-pipeline tools built on PyNvVideoCodec 2.2.

Nvidia's open-weight Nemotron 3.5 Lightning targets the repetitive execution layer of AI agents, launching alongside a NeMo Switchyard library that routes tasks between it and frontier models.

Red Hat and partners including NVIDIA, Microsoft, and IBM Research launched asago, an open source project that automates translating AI governance policy into deployable safety controls.

Thinking Machines Lab shipped its first model, Inkling, an open-weight 975-billion-parameter system that tops US open-weight benchmarks but still trails China's leading models.

NVIDIA's open-weight TwoTower model reuses a frozen Nemotron backbone and trains only a denoiser tower, reaching 2.42x faster generation at 98.7% of baseline quality.

Nvidia joined a $190M round backing a startup building an open scale-up switch to rival its own NVLink fabric.

IDC says Nvidia took the No. 1 revenue spot in data center Ethernet switching in Q1 2026 with $2.1 billion and a 21.5% share, edging Arista.

At Automate 2026, NVIDIA unveiled Halos for Robotics, a three-layer functional-safety system spanning IGX Thor compute, Halos OS, and an ANAB-accredited inspection lab, with Agility Robotics as first adopter.

The startup says its 3nm Napier inference chip turns matrix multipliers into adders, with commercial sales of a 72-chip system slated for the second half of 2027.

The 550-billion-parameter model activates 55 billion parameters per token, ships under the Linux Foundation's OpenMDW-1.1 license, and trades blows with China's Kimi-K2.6 on coding benchmarks.

NVIDIA and Unitree announced a jointly developed open humanoid robot reference design built on Unitree's H2 Plus chassis, with 75 total degrees of freedom and Jetson Thor compute, due from Unitree in late 2026.