Category: AI Infrastructure
-

The Plumbing Behind Your AI Tools Just Got Rebuilt
Most people will never notice this happened, and that is exactly the point. Model Context Protocol, the open standard that lets AI models like Claude plug into outside tools and data without everyone writing custom connectors, just got its biggest update since it launched. The new spec, dated July 28th, drops the old requirement that…
-

Nvidia Just Bet $1 Billion on Korea’s AI Backbone
Nvidia, the Korean internet giant Naver, and the investment firm Brookfield just agreed to triple the size of a data center campus in Sejong, South Korea, taking it from 55 megawatts to 200 megawatts. The total price tag on this expansion is 10 billion dollars. Break down who is paying for what and the deal…
-

Microsoft Is Feeding Its Own AI Before It Feeds Azure
Business Insider talked to people inside Microsoft who described something the company has never said out loud in public quite this plainly. When GPU capacity is scarce, and it is scarce right now, Microsoft’s own AI products get served first. Azure customers get whatever is left over. This is not really new information if you…
-

Nvidia Might Be About to Backstop OpenAI’s Biggest Bet
The Wall Street Journal reported that Nvidia is in talks to guarantee roughly 250 billion dollars in financing so OpenAI can lease a massive data center that SoftBank’s energy arm is building in southern Ohio, on the site of a former uranium enrichment plant. The campus is meant to run at 10 gigawatts, which is…
-

Anthropic Is Asking SK Hynix to Help It Build Its Own Chips
Anthropic just asked SK Hynix for parts to build its own chips. Not buy more of somebody else’s supply. Build its own, from scratch. SK Group’s chairman, Chey Tae Won, said it himself at an AI summit in San Francisco, standing right next to Anthropic CEO Dario Amodei. He called it remarkable that a company…
-

China’s Nvidia Challenger Just Filed for an IPO
A Shanghai chip designer most people outside China have never heard of just filed to go public in Hong Kong, and the timing tells you a lot about where this race is headed. MetaX makes GPUs built to replace Nvidia hardware inside Chinese data centers. Its current chip runs at roughly three quarters of the…
-

AMD Finally Built Something Nvidia Should Worry About
AMD stopped playing catch up this week. At its Advancing AI event, CEO Lisa Su unveiled Helios, a full rack-scale AI system built to go head to head with Nvidia’s Vera Rubin. Seventy two Instinct MI455X chips, 31 terabytes of memory, all wired together with AMD’s own networking and software stack. Su says Helios beats…
-

OpenAI Just Bet 30 Billion on Compute
OpenAI announced a $30 billion data center investment this week, and it is a statement about where frontier AI lives. Not in software. Not in clever ideas. In raw, permanent, physical infrastructure. Thirty billion dollars in servers and power. That is what it costs now to train and run frontier models at scale. That is…
-

Microsoft and Mistral Just Bet Billions on European AI
Microsoft and Mistral announced a major expansion of their partnership on July 21, and the bet underneath it is crystal clear: Europe is going to be the battleground for sovereign AI. The deal is multibillion dollars. The structure is thousands of the latest Nvidia Vera Rubin GPUs deployed in Europe. The mission is to give…
-

Google’s Chip Just Got Very Specific
Google is building a server chip called Frozen v2 that hardwires Gemini’s architecture directly into silicon. Not weights. Architecture. The blueprint itself becomes hardware. The payoff: 6 to 10 times the token throughput per watt compared to general-purpose TPUs. This is a bet on a settling architecture. Once your model design is baked into silicon,…
