Extreme weather events, such as hurricanes, winter storms, and tornadoes, have become a major cause of large-scale electric power outages in recent years, causing billions of dollars in losses.

In a new study, researchers have analyzed power outage data and corresponding weather records from several major service territories on the East Coast of the United States. They found that excessive weather stress and planning vulnerabilities at specific grid nodes are key drivers of prolonged local outages, which spread to the whole system. The authors use their findings to suggest ways to reduce customer outages.

The study, published in the INFORMS Journal on Data Science, was conducted by researchers at Carnegie Mellon University, Moonshot for Electric Grid, the Georgia Institute of Technology, Argonne National Laboratory, the University of Maryland, and the University of Illinois Urbana-Champaign.

“Resilience—the capability of withstanding, adapting to, and recovering from a large-scale disruption—has become a top priority for the power sector,” explains Shixiang (Woody) Zhu, assistant professor of data analytics at Carnegie Mellon’s Heinz College, who led the study. “But a system-level understanding of power grid resilience remains limited, despite the importance of accurately assessing this capability.”

After extensive losses as a result of extreme weather in the early 2000s, U.S. regulatory entities at different levels asked the industry to investigate the resilience of the power grid and adopt measures against extreme weather. But for a variety of reasons, identifying the key factors that contribute to the massive blackouts has long been a very complicated problem.

In this study, researchers used a spatiotemporal model and adopted a data-driven approach to analyze quarter-hourly, customer-level power outage data and corresponding weather records in Georgia, Massachusetts, North Carolina, and South Carolina.

They defined power grid resilience as infrastructural resistance to extreme weather and operational recoverability from such damages. Their model captures three important factors of infrastructural resistance that are closely tied to large-scale power outages: planning vulnerability, maintenance sufficiency, and criticality.

The researchers’ model suggests that local power outages directly induced by extreme weather were a nonlinear response to the accumulation of weather effects and caused subsequent large-scale and long-term blackouts by spreading failures through some critical nodes in power networks. Simulations showed that targeted interventions, such as isolating critical nodes and protecting vulnerable nodes from transient faults, could reduce customer outages by 45.5% and 49.5%, respectively. Among the study’s additional findings:

• Outage rates in metropolitan or economically strong areas were generally lower due to less vegetation, more underground or steel-structure-supported power lines, and adequate repair resources. Thus, the electricity infrastructures in those areas are less vulnerable to extreme weather events and more recoverable if damage to an infrastructure occurs.
• In contrast, rural areas, especially those with terrains like mountains, forests, rivers, and deserts, were hard to access and locate a fault, which inevitably delayed recovery from outages. Also, those economically weak areas usually lacked the resources to maintain or upgrade their electricity infrastructures, which became increasingly vulnerable to extreme weather events, resulting in relatively high outage rates.
• The direction (from source to target) of the spread of an outage typically followed the direction in which power flowed: An area with large-generation capacity or dense transmission network facilities (e.g., substations) was probably a hub of outage propagation. Such an area was more likely a mid-sized urban area, which could be developed to host several transmission or generation facilities, but was not a big load center that dominantly attracted power flows.

“Our study suggests there are planning and operational measures that can prevent and mitigate weather-induced power outages,” says Feng Qiu from Argonne National Lab, who coauthored the study. “Among these is reducing the interdependency of power grids by improving their operational flexibility and embracing diversified sources with distributed locations and versatile operation schemes.”

Insights such as these, the authors say, can inform strategies for decision makers to enhance grid resilience and reduce the likelihood of future disruptions.

A new study from MIT suggests the biggest and most computationally intensive AI models may soon offer diminishing returns compared to smaller models. By mapping scaling laws against continued improvements in model efficiency, the researchers found that it could become harder to wring leaps in performance from giant models whereas efficiency gains could make models running on more modest hardware increasingly capable over the next decade.

“In the next five to 10 years, things are very likely to start narrowing,” says Neil Thompson, a computer scientist and professor at MIT involved in the study.

Leaps in efficiency, like those seen with DeepSeek’s remarkably low-cost model in January, have already served as a reality check for the AI industry, which is accustomed to burning massive amounts of compute.

As things stand, a frontier model from a company like OpenAI is currently much better than a model trained with a fraction of the compute from an academic lab. While the MIT team’s prediction might not hold if, for example, new training methods like reinforcement learning produce surprising new results, they suggest that big AI firms will have less of an edge in the future.

Hans Gundlach, a research scientist at MIT who led the analysis, became interested in the issue due to the unwieldy nature of running cutting edge models. Together with Thompson and Jayson Lynch, another research scientist at MIT, he mapped out the future performance of frontier models compared to those built with more modest computational means. Gundlach says the predicted trend is especially pronounced for the reasoning models that are now in vogue, which rely more on extra computation during inference.

Thompson says the results show the value of honing an algorithm as well as scaling up compute. “If you are spending a lot of money training these models, then you should absolutely be spending some of it trying to develop more efficient algorithms, because that can matter hugely,” he adds.

The study is particularly interesting given today’s AI infrastructure boom (or should we say “bubble”?)—which shows little sign of slowing down.

OpenAI and other US tech firms have signed hundred-billion-dollar deals to build AI infrastructure in the United States. “The world needs much more compute,” OpenAI’s president, Greg Brockman, proclaimed this week as he announced a partnership between OpenAI and Broadcom for custom AI chips.

A growing number of experts are questioning the soundness of these deals. Roughly 60 percent of the cost of building a data center goes toward GPUs, which tend to depreciate quickly. Partnerships between the major players also appear circular and opaque.