One Encoded Character Bypassed Cisco SD-WAN Manager Authentication
One encoded character is enough for admin on Cisco Catalyst SD-WAN Manager. CVE-2026-76504 is exploited in the wild and the federal deadline is October 3.
The AI Buildout’s Real Number Is 180 Gigawatts, Not 420
Legrand models 180 GW of installed data center power by 2030 against 420 GW announced. Powered landlords are repricing leases above 60 percent.
Vultr’s $1.2 Billion Bet Is Ethernet Inside the GPU Rack, Not NVLink
HPE booked its first AMD Helios AI rack order, $1.2 billion from Vultr. Six Juniper switches inside each rack put Ethernet into the GPU scale-up domain.
AMD Bought the Lab That Tells It What to Build Next
AMD is buying World Labs for $8.2 billion to steer its chip roadmap, but the software gap with CUDA is still the real problem.
Vera Rubin NVL72 Reached Production, and the Benchmark Moved to the Rack
CoreWeave put NVIDIA Vera Rubin NVL72 into production with Cognition, and the number that now matters is tokens per megawatt.
A 1 MW Rack Broke the 54-Volt Bus. The Fix Is 800 Volts DC.
800 VDC power delivery is moving from roadmap to product as Infineon and Eaton pair silicon carbide with solid-state transformers.
GPU Rental Prices Are Going Up. The Token Price Is Not.
GPU rental prices are rising as Nebius lifts H100 and B300 rates up to 21% on 1 October, even as cost per token keeps falling.
Cerebras and Gimlet Bet 100 Megawatts That Two Kinds of Silicon Beat One
Cerebras and Gimlet plan 100 megawatts of disaggregated inference, splitting prefill and decode across wafer-scale silicon and GPUs.
Anthropic’s Biggest IPO Number Is Not the Loss. It Is the Compute Bill.
The Anthropic IPO prospectus puts compute at 58 percent of operating expense and $518 billion of forward infrastructure obligations.
Four Countries Announced GPU Clusters This Week. Almost None Said What They Cost.
Sovereign AI infrastructure announcements name the GPUs and skip the megawatts, the cost and the offtaker. Here is what four Asian clusters disclosed.