Nvidia unveils custom high-bandwidth memory promising higher bandwidth and lower power use - but who will actually get to use it?
THE SO WHAT
A custom NVHBM spec claiming 30% more bandwidth and 15% less power than HBM4E tightens Nvidia’s grip on the full AI stack—silicon plus memory. For buyers, the real constraint won’t be performance, it’ll be access and allocation, so lock in supply commitments early if you depend on top-bin GPUs.
READ THE SOURCE
MORE FROM THE WIRE
Deep & Emerging TechGoogle patches multiple Chrome bugs including a V8 flaw being used in attacks
Twelve Chrome vulns with yet another V8 zero-day—plus the EU’s 24-hour disclosure clock—turn browser patching into a compliance race. CISOs should tighten auto-update policies and logging around browser versions before regulators start asking for receipts.
Deep & Emerging TechMan Arrested After Showing Up at Police Station With a Truck Full of Flock Cameras
Physically ripping out Flock cameras and livestreaming it is a reminder that surveillance infrastructure has a social license, not just a contract. Cities and vendors relying on ALPR networks should expect more direct pushback and factor community sentiment into deployment plans.
Deep & Emerging TechCarney’s Push to Build Canadian Data Centers is Running Into Backlash
Community and political backlash to Mark Carney’s data center push in Canada mirrors US fights—power, water, and land use are now gating factors for AI infra as much as chips. If you rely on hyperscale capacity, start mapping geographic and permitting risk alongside your GPU forecasts.
Deep & Emerging TechThink of Nvidia’s $13 billion deal for Hugging Face as a form of ‘health insurance’
Spending $13B on Hugging Face is Nvidia underwriting the open-source ecosystem to keep model diversity—and thus GPU demand—high. If you’re building on open models, expect tighter coupling to Nvidia’s stack over time, from optimized runtimes to bundled credits, and plan accordingly for lock-in and pricing power.