Desk
NVIDIA hardware, software, and developer news.
A physical AI investor says robotics is still in the building-blocks phase, not its ChatGPT moment. We think that is the honest read, and a useful corrective to the hype.
DeepSeek just showed the world the engine room behind its cheap tokens — real production serving code, under a permissive license, with the hardest pieces still held back.
Agents spend much of their time running code and tools, not generating text, and NVIDIA is pitching its Vera CPU as the engine for that work, with a long list of partners behind it.
Spain's national supercomputer is opening its doors to startups, with NVIDIA's startup program as a side entrance. Sovereign compute is starting to look like a real product, not just a slogan.
Amazon's TorchNeuron goes after Nvidia's real moat, the software, and its plan to strip chip-specific code out of PyTorch could matter far more than one more chip option.
Berlin just signed on with NVIDIA to push AI into factories and the Mittelstand. It's a smart bet on Germany's real strength, and a deeper dependence on one American supplier.
Tenstorrent turned down a possible sale and raised $1.5 billion instead, betting that serving open-weight models cheaply is a big enough market to build an independent chip company on.
A trading firm's engineers wrote Mamba-2's forward pass in NVIDIA's Python tile language and beat a Triton baseline. The bigger story is who gets to write fast GPU code now.
Europe's public broadcasters want AI agents in the newsroom without handing over control of their data or editorial judgment. NVIDIA is happy to help write the rules, and sell the hardware.
NVIDIA's new Structured Data Models library makes table predictions without a training step, and a hands-on walkthrough shows the real risk lives in quiet defaults, not the models.
RTX Spark laptops are finally on shelves, and NVIDIA threw a world tour of parties to prove it. The real test starts once the confetti is swept up and owners start using them.
NVIDIA’s RTX Spark pitch: GB10 SoC, up to 128GB unified memory, and local coding agents that don’t need a cloud VM for every loop.