Volantis Raises $88 Million to Build Photonic Inference Systems It Says Offer 30 Times the Memory Bandwidth
The Series A startup uses optical interconnects to attach far more memory to an AI accelerator. All performance figures are company claims.
Apple Tightens macOS Full Disk Access Controls Because of AI Agent Risks
TechCrunch reports that Apple will require very explicit user action before an app gets broad access to files, mail, messages and browsing history.
Meta Releases Muse Wonder, Its First Frontier Model, and Prices It Like GPT and Claude
TechCrunch reports on Meta Superintelligence Lab’s two new models, which power the Muse agent. Benchmark results are Meta’s own.
Microsoft Unveils Maia 300, an Inference Chip It Says Will Become Its Primary Source of Capacity
The four-chiplet accelerator on a 2-nanometer node pairs 288 GB of HBM4 with a dataflow design. The efficiency numbers are Microsoft’s own estimates.
xMEMS Begins Mass Production of Solid-State Chips That Cool AI Glasses, Phones and Data Center Parts
The startup, known for earbud speakers, says its no-moving-parts cooling chips are now in volume production. Performance claims are the company’s own.
Nvidia Contributes Vera Rubin Rack Designs to the Open Compute Project and Details Spectrum-6
At the OCP Global Summit in San Jose, Nvidia said open blueprints and faster networking are meant to remove the bottleneck of how quickly AI factories can be built.
Cisco Ships a 102.4-Terabit Silicon One G300 and Reveals Repluggable Optics With Nvidia
At the Open Compute Project Global Summit, Cisco described new open silicon and optical networking aimed at the network bottleneck in large AI clusters.
Analyst: Nvidia Is Turning the AI Factory Into a Product and Agents Into the Workload
ZK Research’s Zeus Kerravala reads the GTC Berlin announcements as a shift from selling the key AI component to selling an operating model for AI.
At GTC Berlin, Jensen Huang Pitches an AI-Everywhere Europe With Robot Ducks, Humanoids and Phone Calls
The Register’s account of Nvidia’s Berlin keynote covers the DSX POD, physical AI demos, a protest, and a Forrester analyst’s view of Nvidia’s European investment.
Nvidia’s DSX Pods Package Vera Rubin Racks in Factory-Built Containers
At GTC Berlin, Nvidia unveiled prefabricated data center containers it says can cut deployment from six months to as little as two weeks. The Register notes the claims are Nvidia’s own.