AI Pulse · Frontier Labs Watch

Nvidia and SK Group just locked in a deal worth more than $500 billion to build AI infrastructure, including a commitment to secure next-generation high-bandwidth memory supply and construct data centers that will pull two gigawatts of power. The first facility goes live in 2027, according to CNBC.

This matters because memory supply is now the bottleneck. Every frontier lab, every cloud provider, every company trying to run serious AI workloads competes for the same pool of HBM chips. When Nvidia ties up SK Hynix’s production years in advance, it shapes what everyone else can build and what they’ll charge for it. The recent price cuts we’ve seen from providers like Anthropic’s Opus 5 become possible only when someone upstream secures enough memory to drive volume and efficiency. Without deals like this, you get scarcity, higher costs, and slower rollouts.

For business owners, this is upstream infrastructure news that trickles down fast. If you’re planning to scale AI work in the next two years, you’re betting on whether the labs you use can get their hands on enough compute. Nvidia locking down memory supply now means some providers will have room to cut prices or expand capacity, while others won’t. It also means the gap between frontier models and everything else could widen if smaller players can’t secure similar deals.

This is exactly the kind of shift we track and build into tools like the Omni Command Centre, so you’re not guessing which provider will have capacity when you need it. The infrastructure layer moves slowly until it doesn’t, and then it rewrites the economics of everything on top of it.