DeepSeek Takes 44 Percent of OpenRouter Traffic at One Ninety-First of Claude’s Price
Four DeepSeek variants carry 44 percent of OpenRouter tokens. Claude Opus holds 4.4 percent at 89 times the input price.
The Model Topping the HuggingFace Chart Is Not the One People Run
A community quant of DeepSeek V4.1 has 12 times the downloads of the original. Seven of the top 30 trending models are re-uploads.
Microsoft Cut $5 Billion From Capex. The Reason Was Buildings, Not Demand.
Hyperscaler capex lands between $803B and $828B this year. The first cut came from late data center capacity, and that reprices 2027 GPU capacity.
Your GPU Cluster Has 268 Settings. NVIDIA Put Them Under Version Control.
NVIDIA ships a validated Kubernetes runtime for Dynamo as a pinned recipe. The cluster setting nobody wrote down is the one that breaks.
Failing Costs the Attacker Nothing, Which Is Why a 53 Minute Kernel Exploit Works
CVE-2026-98432 turns a Linux TCP race into remote root in 53 minutes, and Go 1.24 widened who is exposed. The retry loop is the story.
Nine AI Platforms Were Breached in October. The Attacker Wanted Your Inference Key.
Nine AI platform breaches in October, and two stolen keys that burned 2.5 million model calls. Why your quota is counted one region at a time.
One robots.txt Line Now Says No to AI Training. Nobody Has to Read It.
A new robots.txt training signal replaces a block per provider with one policy token. Four AI companies back it, and any crawler may ignore it.
Google Capped the Gemini Bill. Provisioned Throughput Keeps Running.
Google Cloud spend caps are generally available for four services. The fuse stops on demand AI traffic and leaves reserved capacity billing.
The Seller Took the Residual Risk. Your Stack Did Not Get the Same Deal
Four deals put the seller behind the resale value of AI hardware. AI capacity allocation for 2028 is now a contract, not a purchase order.
Your 2027 Cluster Has a Host CPU Problem, and Intel Just Said So
Intel sold every server CPU it could make. The server CPU shortage runs to 2028, and host density per rack is now the constraint.