What happened
Nvidia and SK Group locked in a $500 billion, multi-year partnership to build AI infrastructure at scale, according to a CryptoBriefing report Sunday citing the announcement. SK Group is the Korean conglomerate that owns SK Hynix, the world's dominant supplier of high-bandwidth memory used on Nvidia's H100, H200 and Blackwell-class accelerators, and SK Telecom, which has been expanding its data-center and AI cloud business.
The deal, as reported, wraps chip supply, memory, energy sourcing and data-center capacity into a single vendor-customer commitment stretching over several years. Neither side has published a full breakdown of how the $500 billion figure is phased, but the framing puts it in the same tier as the largest multi-year infrastructure commitments the semiconductor industry has ever disclosed.
The announcement lands with Nvidia already the single most valuable public company tied to the AI capex cycle, and with SK Hynix running near capacity on HBM3E orders through 2026.
Why it matters
For AI, the bottleneck stopped being GPUs alone a year ago. It's HBM, power and rack space. This deal names all three in one document.
SK Hynix supplies the memory that gates Nvidia's shipment volumes. SK Telecom runs the sites where those GPUs get racked in Asia. Bundling them under one commitment tells the market Nvidia is willing to pre-buy the choke points instead of hoping they clear.
For crypto, the read-through runs through a specific corner of the market. AI-linked tokens - decentralized compute networks, GPU rental protocols, inference marketplaces - price off the same physical shortage. When the biggest buyer of GPUs locks in half a trillion dollars of forward capacity with the biggest HBM supplier, it hardens the case that compute stays scarce and expensive well into 2027.