HOUSTON, March 23, 2026, 11:59 CDT
Google sounded the alarm Monday over what it sees as lagging U.S. electricity development, flagging power as an obstacle on par with semiconductors for the next generation of AI data centers. “We are concerned that we are not full throttle on energy,” said Alphabet and Google President Ruth Porat, speaking at CERAWeek in Houston. Reuters
The squeeze is real now: big cloud players are scaling up computing muscle much faster than utilities can put up new grids and substations. U.S. Energy Information Administration data shows power demand has climbed 1.7% annually since 2020, after a long stretch of barely any change, thanks mostly to sprawling computing hubs. Looking ahead, the agency sees load picking up 1.9% in 2026, then 2.5% in 2027. Last week, Google said it’s ready to shift up to 1 gigawatt of its data center load for demand response — enough juice for about 750,000 homes — with Michael Terrell, who heads advanced energy at the company, calling it “a really important tool for meeting future demand.” Reuters
Nvidia and startup Emerald AI on Monday announced partnerships with AES, Constellation, Invenergy, NextEra, Nscale, and Vistra to develop “flexible” AI factories. Their approach: build power generation and storage right at or next to data center sites, aiming to accelerate grid hookups. Nvidia CEO Jensen Huang pointed out that energy, compute, networking, and cooling need to be engineered as a package. Constellation CEO Joe Dominguez weighed in, saying, “We don’t have a supply problem — we have a peak problem.” NVIDIA Newsroom
Project plans are shifting fast. On March 20, the U.S. Department of Energy said SoftBank and AEP Ohio intend to overhaul federal land in southern Ohio, building a massive data-center campus with 10 gigawatts of capacity and adding 10 GW of fresh generation—most of it gas, some 9.2 GW. Officials expect construction to get underway this year.
Washington’s been pushing tech companies down a similar path. According to a White House fact sheet dated March 4, Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI have agreed to secure or develop new power for their data centers, and they’ll also cover the cost of required delivery upgrades—rather than passing those expenses on to households.
Google isn’t the only tech player hitting this wall. Back in February, Amazon Web Services flagged to Reuters that getting grid connections is now a top constraint for its European growth. Pamela MacDougall, who oversees energy markets and regulation for AWS in EMEA, laid it out: tying into the region’s transmission network can drag on for as long as seven years—while building the actual data center takes closer to two.
Alphabet picked up $0.61 to trade at $301.61 by late morning in New York. Nvidia tacked on $2.87, reaching $175.57. Among AI-related power stocks, Constellation advanced $7.68 to $289.67, Vistra pushed $5.73 higher to $151.75, and Talen Energy jumped $13.09, hitting $316.06.
The bottleneck isn’t going away quickly. PJM—the country’s biggest grid operator—might trigger a wave of direct power deals with data-center operators under its February proposal targeting major new users, analysts said. Still, both James West at Melius Research and Rick Pederson from Bow River Capital warned that drawn-out permitting, state-level sign-offs, and lengthy interconnection queues could stall any “flurry” of agreements. Reuters
No sign of relief yet. The Electric Power Research Institute figures data centers might soak up as much as 9% of U.S. electricity by 2030—that’s more than double the 4% expected in 2025. S&P Global, for its part, pointed out last week that global data-center power demand could jump anywhere from 12% to 16% annually between 2025 and 2030.
