OpenAI's new chip deals raise a tough question: Where will all the power come from? (www.businessinsider.com)

OpenAI has struck a multibillion-dollar, five-year deal with AMD to deploy 6 gigawatts of AI chips — incremental to a separate 10-gigawatt pact with Nvidia announced last week — and will receive warrants for up to 160 million AMD shares (about 1%) as deployments hit gigawatt milestones. AMD says the first gigawatt should begin coming online in the second half of 2026, but full rollout depends on whether OpenAI and partners can secure enough power. The structure ties financial incentives to deployment pace, underscoring how critical raw compute capacity is to OpenAI’s growth plans. The deals spotlight a growing bottleneck: U.S. utilities estimate roughly 60 GW of new power will be needed by 2030 to serve AI and other loads, yet grid upgrades take years of planning and permitting. Industry voices and recent projects show a pivot toward on-site generation and self-supply — OpenAI already runs a natural-gas plant at its Abilene Stargate site, while xAI has used mobile turbines and bought a former gas-plant site. The technical and market implication is profound: hyperscalers may increasingly build bespoke power plants, reshaping data-center economics, permitting regimes, and emissions profiles while accelerating investments in grid-scale generation, storage, and local energy solutions.