
NVIDIA announces new RTX PRO 6000 Blackwell GPU series for workstations
Up to 96GB of GDDR6 ECC memory
At GTC 2025, NVIDIA has unveiled its new RTX PRO 6000 Blackwell GPU series for workstations, including a full lineup of desktop and laptop GPUs, as well as a data center-oriented RTX PRO 6000 Server Edition one.

Top fabless chipmakers rake in $249.8 billion
Half of this went to Nvidia
The semiconductor industry in 2024 was an all-you-can-eat buffet for AI chipmakers, with the top ten fabless firms raking in a cool $249.8 billion—nearly half of which was trousered by Nvidia.

Micron ships HBM3E and SOCAMM products for AI servers
Memory wizardry is the secret sauce behind AI
Memory outfit Micron is shipping both HBM3E and SOCAMM products for AI servers, and claims its chips will be the secret sauce behind the AI boom.

Intel releases XeSS 2 SDK
Enters the AI upscaling race
Troubled Chipzilla has finally pulled its finger out and released the XeSS 2 SDK, hoping to claw back some relevance in the AI upscaling arms race.

MSI snubs AMD’s RDNA 4 lineup
Agreed to see other people
MSI has quietly ghosted AMD’s latest Radeon GPUs, skipping the entire RDNA 4 lineup and leaving the Radeon RX 9070 series without one of its usual partners.

Bolt Graphics brings RISC-V to the GPU market
Hail Zeus
Bolt Graphics, a startup out of Sunnyvale, California, has announced its Zeus GPU platform, promising to thrash Nvidia’s GeForce RTX 5090 in path tracing and FP64 compute workloads.

Beijing boffins create chips from Bismuth
More digestible than silicon
Troubled Chipzilla and its mates at TSMC might want to start sweating because a bunch of Beijing boffins have emerged from their smoke-filled labs with a non-silicon based transistor that’s faster and more efficient than anything out there.

Nvidia’s Blackwell spin spotted
Sales not as good as claimed
Nvidia is trying to convince everyone that its shiny new RTX Blackwell cards have outsold the previous generation 2:1 in the first five weeks.

Google claims its new AI needs less hardware
Same performance as DeepSeek AI’s R1 with a single Nvidia H100 GPU i
Search outfit Google is claiming that its latest open-source model, Gemma 3, can match nearly the same performance as DeepSeek AI’s R1—while using just a single Nvidia H100 GPU instead of 32.

Meta training its own AI chips
RISC-V business
Meta is quietly testing RISC–V–based AI training chips to kick Nvidia out of its wallet.