A national emergency meant to shield America's power grid from Chinese sabotage is now the thing utility planners blame for stalling the batteries that AI data centers need most.
On August 26, President Trump signed Executive Order 14420, declaring a national emergency over foreign-made equipment sitting inside America's bulk-power system, the high-voltage transmission lines, substations and grid control gear that keep electricity flowing. The order, invoked under the International Emergency Economic Powers Act and the National Emergencies Act, gives the Department of Energy sweeping authority to block transactions involving equipment from countries deemed national security threats, and it doesn't stop at future purchases. It reaches backward to gear already installed and humming on the grid today.
The order never names China outright. It doesn't have to. China dominates manufacturing for the exact hardware the order targets: roughly 80% of global solar inverter and battery-cell capacity, according to International Energy Agency data cited in reporting on the order, and about 60% of the lithium-ion batteries currently installed in U.S. grid storage projects. Hoodline and other outlets covering the signing noted that nearly 40% of inverters installed in the U.S. in 2025 came from Chinese suppliers. That's not a niche supply chain. That's most of it.
Energy Secretary Chris Wright has been building the case for this for months, framing it less as a trade fight and more as an existential bottleneck. He's argued publicly that the grid has to modernize or it will become the thing that slows AI down, not chips, not capital, the wires and transformers themselves. Worldwide data center power demand is projected to hit 133 gigawatts in 2026, up from 105 gigawatts in 2025, according to Gartner, and AI-optimized servers alone are expected to account for 31% of that draw. Power, not silicon, has become the constraint everyone in the industry quietly admits they didn't plan for fast enough.
Here's the problem. The equipment this order restricts, batteries and inverters especially, is the same equipment utilities are racing to install to handle the load AI is adding to the grid right now. BloombergNEF has identified more than 23 gigawatts of large-scale energy storage projects that have already reached financial close or are under construction in the U.S., and the firm told Utility Dive that a meaningful share of that pipeline now faces delay or outright cancellation while developers wait for the Department of Energy to say which suppliers are actually banned.
AI data centers have pushed America's largest power grid to its breaking point PJM Interconnection, the US grid serving 67 million people, missed its power supply reliability target in July 2026 as AI data center demand drove capacity costs up more than tenfold. Federal officials are now debating breaking the operator apart, and founders planning GPU clusters face a hard infrastructure ceiling no amount of fundraising can... - AI data centers power grid strain - PJM capacity auction megawatt shortage
That uncertainty has teeth. American utilities were already waiting more than two years for large power transformers before this order, a shortage that predates the Trump administration entirely. Pulling Chinese suppliers out of an already strained supply chain, without yet telling anyone which replacement vendors are acceptable, doesn't shrink the backlog. Canary Media reported the order's language is vague enough that developers are reconsidering signed contracts and shopping for new suppliers before they even know the final rules.
The Department of Energy has 120 days from the signing to publish those rules, a deadline that lands on December 24. Until then, every utility with a Chinese-sourced battery contract in the pipeline is making a bet on what Washington decides a backdoor risk actually looks like.
Frankly, the timing is awkward. The same administration pushing this emergency order announced in late September that it would spend nearly $2 billion on 31 grid upgrade projects across 26 states, aimed at squeezing more than 23 gigawatts of additional transmission capacity out of the existing network using real-time sensors and smarter routing. That's the Energy Department trying to add grid capacity with one hand while an emergency order it also wrote makes the parts needed to add that capacity harder to source with the other.
Wright has said the long-term fix is dispatchable power, mainly natural gas in the near term, because he doesn't think either the U.S. or China can run an AI buildout on renewables alone. He's probably right that gas plants solve the generation side faster than permitting fights over new transmission lines ever will. But generation isn't the part breaking right now. Storage and inverters are, and this order just made the ones utilities already know how to buy harder to get.
None of this means the security concern is manufactured. Grid operators and cybersecurity researchers have warned for years that remote-access backdoors in foreign-made grid hardware are a real and underappreciated risk, and an adversary with access to substation control equipment is a legitimate worry in a way that, say, a banned social media app is not. The administration isn't wrong that the risk exists. It's that the fix, as written, lands on the exact equipment category the AI buildout can least afford to lose right now, and nobody in Washington has said yet what happens after December 24.
Also read: The 10-Year Treasury Yield Just Broke a Record That Stood Since 2007 • Bond investors are starting to say no to the AI debt binge • Cisco turns Webex into an AI agent workspace with Claude built in
Mitsubishi Bets 7.5 Billion Dollars That Gas Power Is AI's Real Bottleneck Mitsubishi Corporation closed its largest-ever acquisition on July 15, a $7.5 billion deal for Aethon Energy's Haynesville shale gas assets. The bet reaches beyond LNG exports to Japan and into the gas-fired power now feeding America's AI data center buildout. - AI infrastructure power consumption costs - natural gas demand for data centers
This article is posted in AI News, check it out for more related stories.
Join the discussion #
Open in the community → Almost there. Sign in and your reply posts straight away.