The big challenge to OpenAI's $100 billion deal with Nvidia: Access to power (www.businessinsider.com)

Nvidia’s announced $100 billion investment in OpenAI locks in massive hardware access — including capacity tied to roughly 10 gigawatts of high‑performance GPUs — but exposes a critical missing ingredient: electricity. That 10 GW of new demand is roughly equivalent to New York City’s summer peak and comes on top of industry estimates that the U.S. will need about 60 GW of new power to serve data centers by decade’s end. Utilities are already overwhelmed: permitting, transmission upgrades and interconnections take years, and the current data‑center construction pipeline has slowed (CBRE reports a 17.5% drop in H1 2025), with power availability cited as the key constraint. The technical and strategic implications are immediate. Securing chips and capital no longer guarantees deployment without grid capacity, long lead times for transmission upgrades, and regulatory hurdles. Big tech is responding by self‑sourcing generation (on‑site gas turbines), buying large off‑take deals (including nuclear), and investing in emerging options like SMRs, hydrogen fuel cells and fusion R&D — but those paths carry environmental, regulatory and timing risks. Experts view the Nvidia‑OpenAI tie as strategically brilliant, yet underscore that getting cumulative power at scale is the “silent bottleneck” of the AI race; resolving it will require massive grid investment, creative commercial power sourcing, and years of infrastructure work.