Eight vLLM Advisories Closed at Once, in a Bite
Eight vLLM advisories close in 0.32.0. The worst vLLM vulnerability skips the insecure serialization flag entirely.
Vera Rubin Lost 13 Percent of Its Bandwidth, in a Bite
NVIDIA cut Vera Rubin bandwidth 12.7 percent per GPU without a release. The eight GPU board lost 26 percent.
A Token Fell 77 Percent, in a Bite
AI token prices fell 77 percent in five months. Darkbloom told 800 Mac owners to expect no recovery, and leases priced on that unit are next.
The Capex Number Moved Because Buildings Slid, in a Bite
Hyperscaler capex lands at $803B to $828B for 2026 and Microsoft cut $5B, blaming late data center capacity.
NVIDIA Pinned 16 Components to a Recipe, in a Bite
NVIDIA ships a validated Kubernetes runtime for Dynamo as a pinned recipe, and switching intent alone moves 41 values.
A Kernel Race Becomes Remote Root in 53 Minutes, in a Bite
CVE-2026-98432 turns a Linux TCP race into remote root in 53 minutes, and a failed attempt costs nothing.
Nine AI Platforms Lost Customer Keys in One Month, in a Bite
Nine AI platform breaches in October and the two stolen keys behind 2.5 million model calls.
The Host CPU Is the Shortage Now, in a Bite
Intel sold every server CPU it could make. Host CPUs belong in the 2027 lead-time tracker beside memory, not below it.
The Trillion Parameter Model Fits on Four GPUs, in a Bite
Mistral Large 4 opened its weights with 1 trillion parameters. On four GPUs the limit is KV cache capacity, not the benchmark.
The TPU Sale Now Comes With a Buyback, in a Bite
Alphabet's filing says some TPU sales carry a buyback the buyer can force, booked as a financial guarantee once the chips move on.