Desk
NVIDIA hardware, software, and developer news.
NVIDIA is turning Cosmos into a full assembly line for physical AI, and the real product here is synthetic data for the dangerous moments nobody wants to film.
NVIDIA wants AI data centers to behave like flexible grid citizens instead of giant fixed loads, and the energy industry's biggest names are signing on.
European retail is quietly proving the business case for small, post-trained open models, and the Shopify numbers are the ones every merchant platform will be quoting.
Open weather AI is getting local, and the people adopting it first are the ones with money on the line: insurers, energy producers and forecasters.
European governments are buying AI on the condition that they keep the keys, and open models are becoming the price of entry for selling into the public sector.
Perplexity and NVIDIA found a way to make a model teach itself without the teacher cheating, and the payoff is training that may need far less compute.
NVIDIA's Python profiler makes GPU measurement almost effortless, which is exactly why its default settings deserve more attention than they get.
NVIDIA is pitching a pre-trained model that reads a company's database as-is and predicts churn or fraud without a training run, and the benchmark claims are strong enough to take seriously.
Nvidia's answer to Europe's slow infrastructure is a data center in a box, and Berlin got a pitch where Nvidia sits under every industry it named.
NVIDIA's open forecasting model leans on a frozen pretrained encoder plus a lookup of similar past patterns, and it ships with explanations. That last part may matter most on a factory floor.
NVIDIA used GTC Berlin to argue that agents belong on the desk as much as in the cloud, and the pitch is strongest where it promises data never has to leave the room.
Google Cloud's A5X instances put Vera Rubin racks in reach of rented customers next month, and the first tenant is telling: a lab betting everything on reinforcement learning from experience.