AI Infrastructure News · June 24, 2026
The AI infrastructure race has shifted from chips to capacity and power. Meta and Microsoft pre-commit tens of billions to multi-year data center leases as global hyperscaler spending reaches $700B in 2026, while $17.5B in nuclear investment targets SMRs for grid-scale power backing the buildout.
CLSK, BTDR) pivot to AI landlords, SpaceX signals a second Colossus facility, and xAI's Colossus hosts frontier model training, demonstrating that infrastructure capacity, not silicon supply, gates next-generation AI.
China claims TOP500 leadership with all-domestic chips while Nvidia processors surge 2-3x on China's black market under export controls; the US supply chain diversifies through Marvell's AI interconnect acquisitions (Celestial AI, XConn) and SK Hynix's HBM validation, as Congress shifts grid costs to tech companies, making power economics the next pricing frontier.
Hyperscaler Spending $700B · Nuclear Investment $17.5B · Hyperscale Data $1.2B · Michigan Capacity 20MW · Nvidia China Black Market 2-3x