As artificial intelligence systems grow more powerful and pervasive, they’re creating an unprecedented energy crisis that’s forcing industries to confront fundamental safety questions. While much attention focuses on AI’s computational demands, a quieter revolution is happening in the energy infrastructure that powers these systems – and it’s revealing critical gaps in how we manage risk at scale.
The Battery Safety Revolution
The recent update to UL 9540A, published in March 2026, represents a pivotal shift in how we approach energy storage safety. This standard now requires large-scale fire testing to demonstrate that thermal runaway events in one battery energy storage system (BESS) won’t propagate to adjacent units. It’s a response to high-profile incidents like the Moss Landing fires in California that forced evacuations and highlighted how battery failures can escalate beyond individual containers.
But here’s the crucial context: these safety standards are evolving just as AI companies are building massive energy infrastructure to power their data centers. According to TechCrunch reporting, Microsoft is constructing a natural gas power plant in West Texas capable of producing 5 gigawatts of electricity – enough to power millions of homes. Google is building a 933 MW natural gas plant in North Texas, while Meta is adding seven natural gas power plants to its Hyperion data center in Louisiana, bringing capacity to 7.46 GW.
The Infrastructure Bottleneck
These massive energy projects face significant challenges that intersect with the battery safety conversation. Ars Technica reports that nearly 50% of US data center projects planned for this year are delayed or canceled, primarily due to former President Donald Trump’s tariffs on Chinese imports. Wait times for essential power infrastructure components like transformers and batteries have increased from 24-30 months to up to five years.
“The hardest stock to source in our marketplace is Anthropic,” says Glen Anderson, president of Rainmaker Securities, highlighting how infrastructure constraints affect even AI companies’ market positions. “There’s just no sellers.”
Why This Matters for Business Leaders
The convergence of stricter battery safety standards and AI’s energy demands creates a perfect storm for decision-makers:
- Compliance complexity: The 6th Edition of UL 9540A now aligns with NFPA 855-2026, which requires performance-based fire safety outcomes rather than prescriptive design alone. This means companies must demonstrate non-propagation between BESS units through actual testing, not just theoretical analysis.
- Supply chain realities: With turbine prices expected to rise 195% by year-end relative to 2019 prices, and companies unable to place new turbine orders until 2028, the cost of energy infrastructure is skyrocketing. This affects not just AI companies but any business relying on stable, affordable power.
- Community resistance: A Harvard/MIT poll found Americans are more worried about how data centers might alter their communities than about rising utility bills. Research indicates AI data centers can create “heat islands,” increasing temperatures in communities and impacting rainfall patterns and pollution.
The Cognitive Surrender Factor
Amid these complex challenges, research from the University of Pennsylvania reveals another layer of concern: “cognitive surrender.” In a study of 1,372 participants and over 9,500 trials, users accepted faulty AI reasoning 73.2% of the time. When AI was accurate, users accepted its reasoning 93% of the time, but even when it was faulty, they still accepted reasoning 80% of the time.
This matters because as companies increasingly rely on AI for energy management and safety compliance decisions, human oversight becomes critical. Adding incentives increased likelihood to overrule faulty AI by 19 percentage points, while time pressure decreased tendency to correct faulty AI by 12 percentage points – suggesting that rushed decisions in high-pressure energy situations could lead to dangerous oversights.
The Path Forward
The relationship between standards like UL 9540A and CSA/ANSI C800:25 illustrates the balanced approach needed. UL 9540A establishes baseline certification requirements, while CSA/ANSI C800:25 provides deeper performance characterization to support engineering judgment and site-specific compliance. Used together, they form a more complete safety narrative.
For businesses navigating this landscape, the key is recognizing that AI’s energy demands and safety requirements are interconnected challenges. Companies that proactively engage with both certification and performance-based testing will be best positioned to demonstrate compliance, earn stakeholder confidence, and accelerate deployment – but only if they maintain critical human oversight in their decision-making processes.
The U.S. Geological Survey estimates enough natural gas in one region to supply the U.S. for 10 months, but with growth in shale gas production slowing considerably in key regions, the race is on to develop sustainable, safe energy solutions that can power the AI revolution without compromising safety or community well-being.