Dear Aventine Readers, 

A few months ago, a power outage in Spain brought the country to a halt for almost twenty-four hours. And while renewable energy didn’t cause the problem, it likely contributed to it. Around the world, electrical grids are straining as a result of increasing electricity demand and the need to accommodate diverse energy sources such as natural gas, nuclear power, solar and wind. In this issue we look at the enormous challenge of making our grids more resilient at a time of skyrocketing energy demands. 

Also in this issue: 

Thanks so much for reading! 

Danielle Mattoon
Executive Director, Aventine

Subscribe

Subscribe to our newsletter and be kept up to date on upcoming Aventine projects

Shortly after midday on April 28, the Iberian peninsula was plunged into chaos.

In an instant, Spain’s electricity grid lost around 15 gigawatts of power, or about 60 percent of its generation capacity. Across Spain and Portugal, cellphone networks failed, trains stopped, factories shut down and hospitals switched to emergency generators. Power was not restored across most of the region until the following morning. Spain’s energy minister, Sara Aagesen, confirmed that a sudden loss of power generation in Granada, followed by two more failures seconds later in Badajoz and Seville, triggered a series of grid disconnections that ultimately led to the blackouts.

The underlying cause of those failures and how they rippled through the grid to cause a nationwide blackout are yet to be determined, said Aagesen. But the incident has prompted speculation about the lack of resilience demonstrated by Spain’s energy network and what needs to change on grids around the world to ensure electricity security in the face of higher demand.

Aventine spoke with engineers, policy experts and energy consultants to understand the current state of the global power supply and what’s needed to improve it. They described grids that are straining as a result of increasing electricity demand and emerging complications due to the need to combine traditional energy sources like natural gas and nuclear power with renewables like solar and wind. They also pointed to a slew of new technologies that could increase grid reliability, and warned that the path ahead will be both expensive and complex. 

“This is a real hard problem,” said Frank Felder, an independent electricity consultant and former director of the Rutgers Energy Institute. “You’re trying to control stuff at the millisecond level … all the way out to transmission interconnections, which could take, at least in the US, 10 to 15 years to plan and build.”

Increasing grid strain

Energy grids have never been, and will never be, perfect. They are complex systems, and they can fail for all sorts of reasons. In 2003, parts of the Midwestern and Northeastern United States including New York City, as well as some parts of Canada, were plunged into darkness, a failure that affected as many as 55 million people and cost as much as $10 billion. That outage was complex — partly caused by power imbalances, human error and a computer glitch. In 2021, Texas suffered a major power crisis that left an estimated 10 million people without electricity and cost as much as $195 billion. It resulted from a lack of winterization of power generation facilities. The Department of Energy estimates that power outages cost the US $150 billion each year. 

Power grids are so complex, with so many inputs and demands, that there can be countless causes for blackout-producing errors. Now, with grids around the world experiencing more strain than ever while also accommodating more renewable energy supplies, we are “asking things from the grid for which the grid was not originally designed," said Alexis Kwasinski, an associate professor of electrical and computer engineering at the University of Pittsburgh who specializes in critical infrastructure resilience. 

At the same time, said Michael Mastrandrea, research director of the Climate and Energy Policy Program at Stanford University’s Woods Institute for the Environment, climate change is exacerbating the problem. Heat waves place greater demand on air conditioning, while extreme weather such as hurricanes and snowstorms puts physical strain on infrastructure with increasing regularity. And as nations become increasingly electrified, and our economies become increasingly dependent on electricity, the cost of outages will only grow, said Felder. 

“We are in a period of transition where the grid that we had isn't the grid that we need in the future,” said Mastrandrea.

The challenge of adding renewables

In the hours and days following Spain’s recent blackout, some commentators pointed to the nation’s high reliance on renewables and their inherent intermittence and unpredictability as a potential culprit; before the country lost power, wind and solar accounted for just under 70 percent of the nation's electricity generation. Yet many experts, including those who spoke with Aventine, explained that a failure of renewables was unlikely to have been the root cause of the problem, not least because the outage occurred on a sunny day just after noon, a peak time for solar generation. What is possible, though, said the experts, is that Spain’s grid may not be set up to handle such a large proportion of renewables during moments of instability; that might have exacerbated problems elsewhere on the grid after the Granada outage and contributed to the blackout. 

Variability of supply is one challenge renewable energy poses to the grid. Another is that renewable energy doesn't deliver electricity at a steady frequency; instead, its frequency matches that of whatever grid it’s hooked up to. Energy grids must operate at a consistent frequency to work properly, and if the frequency on the grid varies significantly, power sources are designed to disconnect from the grid as a safety measure. If and when that happens, renewables can exacerbate the effect and contribute to a shutdown. In Spain on April 28, the loss of two generation sources led to a fall in frequency from 50Hz to 48Hz, which caused an automatic disconnection of the French grid from the Iberian Peninsula and a cascade of other generation failures. 

Technologies can mitigate such risk. Battery storage, for example, can be used to inject electricity of a steady frequency into the grid when an outage occurs. It’s also possible to build so-called “grid-forming” inverters for renewables that help maintain stability. Advanced inverter technology is in development and not yet widely deployed, but batteries are an established technology, and Spain’s lack of investment in them has been widely discussed, even just days before the blackout occurred. For context, California has about 250 times as much battery storage as Spain to support about 30 percent less installed generation capacity. 

Spain’s power grid was also made more vulnerable by being relatively isolated. Grids can be stabilized by a neighboring grid during times of stress or inconsistency. A good example is the interconnection between the UK and Norway, known as the North Sea Link, which enables the two nations to share renewable power — hydro produced in Norway and wind from the UK — to balance their energy supplies. Spain is only weakly connected to the rest of continental Europe through its short border with France, limiting its access to support. Building out new transmission lines and interconnections is a slow process. According to the International Energy Agency analysis, new high- or extra-high-voltage lines can take between four and thirteen years to build in Europe or the US, often running into huge delays because of planning requirements or opposition.

Tech tools for a smarter grid

The concept of a smart grid has existed for years, but the technology to make it happen — from sensors to AI algorithms — is increasingly available and affordable.These improvements are not a substitute for building out grid infrastructure, but they could have significant impact on grid reliability. 

For instance, the grid could be blanketed by sensors that allow energy providers to keep more accurate tabs on supply, demand and hardware faults, helping them react faster when things go wrong. Artificial intelligence could be used to better forecast supply and demand, helping grid operators to plan energy generation or predict maintenance schedules to avoid hardware failures. Felder pointed out how these systems could help with all sorts of esoteric optimization problems, such as calculating the most efficient voltages to carry on power lines for various weather conditions or dynamically adjusting how power is routed during periods of high demand.

Demand response — an approach that attempts to tune demand to supply, rather than the other way around — could also be deployed far more widely. While the concept has existed for decades, it is becoming a more realistic proposition, said Kwasinski, as devices in both commercial and residential buildings become more internet-connected, allowing energy customers to opt in to schemes that, say, delay the charging of electric vehicles during periods of intense energy use.

Local micro-grids could also have an impact on residential demand, said Mark Jacobson, professor of civil and environmental engineering at Stanford University. For instance, domestic solar and batteries, when used at scale, could help reduce the demand that a heat wave puts on a grid, with homes simply powering their own AC units rather than drawing power from a larger network. Jacobson said that the impact of micro-grids is already observable in California, with demand around noon — the point at which domestic solar generation is at its peak — now falling.

Who gets the grid ready for the future?

With so many options available to help boost the resilience of the grid, the choice of which ones to implement, and in what order, is far from straightforward. One option is to do everything at once. “From a pure theoretical energy policy perspective, you have to do all [of it] at the same time,” said Benjamin Sovacool, a professor of energy policy with positions at Boston University, the University of Sussex in the UK and Aarhus University in Denmark. “Does it make sense to get the perfect high voltage transmission network designed if your supply is really shoddy?”

Yet that is a dauntingly expensive task. In a fall 2023 report, the International Energy Agency estimated that annual global investment in energy grids needed to double to more than $600 billion a year by 2030 in order for nations to meet their climate and energy security goals. In December 2024, BloombergNEF estimated that the figure needs to be more than $800 billion by the same date, close to three times as much as the current rate of investment.

A big part of the financing challenge is the structure of the energy market. While the production of energy remains highly lucrative, with companies vying to produce electricity at competitive rates, the transmission and distribution of electricity is handled by natural monopolies at the national or regional level, said Rahmat Poudineh, head of electricity research at the Oxford Institute for Energy Studies. The organizations which own and operate grids, “do not face any competition, and their revenue is regulated,” he said, which means that there is little incentive or ability to invest in new infrastructure. Particularly at the local distribution level, “it is absolutely a fix-it-if-it's-broken approach at the moment,” said Sovacool. “They're not doing preemptive investment.”

That leaves policymakers to consider a trade off, said Poudineh. “Should we define our objective as just a clean power system, or should we redefine the objective as a clean and resilient power system?” If resilience is a goal, he said, just like carbon reduction is a goal in many nations, then it should be treated as one, with targets and appropriate incentives in place. That would require policymakers to calculate the cumulative potential cost of imposing new mandates on our grid — such as, say, demanding that all new renewables installations use grid-forming inverters — and establish whether the extra resilience of those policies might outweigh the potential cost of future blackouts. Alongside policy, we may also see insurers and financiers increasingly demanding commitments to resilience when they support large-scale infrastructure projects. 

For residents of Spain, it is clearly too late for those sorts of cost-benefit calculations and mandates. But one thing is increasingly clear: The trade-off continues to slide in the direction of favoring grid enhancements rather than sticking with the status quo. “In this modern economy, [with] the level of electrification that we are aiming to achieve,” said Poudineh, “resiliency of the system becomes much more valuable.”

Researchers are working to decode the language of dolphins. Credit: Stocksy

Can AI help us talk to animals? You may have seen headlines earlier this month about prizes being handed out for advances in interspecies communication. Perhaps you raised your eyebrow at the Dr. Dolittle headlines. But make no mistake: This is a serious endeavor, and one in which much progress is already being made. The prize — perhaps unfortunately branded with the name Coller Dolittle Challenge — will offer $500,000 cash or up to $10 million equity investment to the first research team that can demonstrate that an animal both understood and responded meaningfully to a message sent by humans. That’s not yet happened. In the meantime, the challenge is awarding a $100,000 prize each year to a group that has made significant progress toward that goal. This year, it went to a lab at the Woods Hole Oceanographic Institution in Massachusetts for research in decoding the language of bottlenose dolphins. What once seemed far-fetched is now increasingly plausible thanks to AI, and such work has become serious science focused on all kinds of creatures. Other shortlisted teams for this year’s prize were looking at nightingales, cuttlefish and marmoset monkeys. Crows, sperm whales, pigs and dogs are among the animals being studied by other research groups. A Nature report last year described how AI could identify individual sperm whales by their vocalizations, and found that elephants and marmoset monkeys give names to their companions. Soon, researchers hope, AI’s pattern-recognition abilities will let it generate strings of sounds that make sense to animals in the same way that, say, ChatGPT can put together words for humans. The missing — and perhaps most important — part of the puzzle is extracting the meaning of those noises, a complex task that will demand zoological expertise and an understanding of communication beyond just sound, such as physical movement. Oddly, that means that we’re likely to communicate with animals before we actually understand what we’re saying to them.

AI in everything, everywhere, ASAP. Google clearly believes that its AI is ready for prime time, and that it should be everywhere. In a recent blog post based on remarks from the company’s annual developer conference last month, Google CEO Sundar Pichai wrote that AI is moving from “research to reality,” and explained how Gemini, the company’s flagship large language model, is being embedded across almost all of Google’s products. The list of more than 100 new features is too long to detail, but it included: AI Mode in search, which replaces traditional search results with a fully AI-generated page; a Gemini app that can analyze what your phone’s camera sees in real time; near-instant audio translation of foreign languages in Google Meet video calls; and an AI-enabled version of Android Auto that lets drivers ask about car problems or for information about their destination. “More intelligence is available, for everyone, everywhere,” Pichai told the crowd. Google’s reach gives it an enormous advantage here: Billions of users will simply start using these features in everyday products by default. Open AI, without Google's massive platform, is now countering, with a $6.5 billion acquisition of Jony Ive’s AI startup, io. The goal there, said OpenAI CEO Sam Altman, is to build “a new family of products” for the AI era. Ive wouldn’t tell the New York Times what those products might look like, but said that they would “elevate humanity.” The Wall Street Journal has reported that the device may take the form of … a necklace. Whether that can elevate humanity remains to be seen. What is certain, though, is that tech companies aren’t just doubling down on building AI. They’re now doubling down on getting it in front of as many people as possible, as fast as they can.

Fiber optic cables are helping us understand the world beneath our feet. Fiber-optic cables like the ones telecom companies use can be used to detect underground activity like earthquakes and geothermal activity because subterranean vibrations subtly affect the way that light travels down them. Now, so-called dark fiber cables — ones that aren’t in active use by telecom companies — are being employed specifically for the purpose of sussing out underground activity. In May of this year, New Scientist reports, a team from ETH Zurich in Switzerland started using four 30-mile dark fibers in Athens, Greece, to examine 350 square miles beneath the city for seismic activity. A project in the UK will soon use fibers linking the cities of Southampton, London and Cambridge to learn if the detection of vibrations can help utilities and construction companies monitor underground pipes and building work. The early results are promising. In California, similar experiments have helped identify geothermal reservoirs that could be tapped for power. In Iceland, they detected early signs of volcanic activity. And cables deployed under the ocean floor have sensed earthquakes fast enough that the technique could be used to provide early tsunami warnings. New research has also suggested that active telecom fibers could be used for data collection, potentially opening up the entire world’s fiber network as a listening system to probe the sub-surface of the planet. It will never provide a complete map, as fiber tends to be laid only between high-population areas, but it could transform our ability to monitor Earth's hidden activity.

Power hungry, from MIT Technology Review
21,000 words, or about 90 minutes, across six stories

We know that AI systems consume enormous amounts of power and produce significant carbon emissions. What we don’t know is exactly how bad the problem is, largely because companies remain tight-lipped on the details around training and running their models. This package of stories from MIT Technology Review aims to clear up that murky picture. The standout piece is an attempt to estimate AI’s energy use with more precision than before. It's an almost impossible task, but the magazine’s reporters worked closely with AI researchers to piece together a detailed view of how much power different models consume and what that looks like when extrapolated to global levels. Other stories in the package tackle the water demands of data centers, the reality of how quickly nuclear power can scale to meet AI’s energy needs, and a sobering look at what will likely power the AI boom in the short term (spoiler alert: it’s probably gas). 

Are groundbreaking science discoveries becoming harder to find? From Nature
3,400 words, or about 14 minutes

Sustaining Moore’s Law — which describes the rate of improvement of chips over time — now takes 18 times as many semiconductor researchers as it did in the 1970s. In drug development, the number of new treatments approved per billion dollars of R&D spending has halved roughly every nine years since 1950. And yet, from 1956 to 2020, US research spending grew 11-fold. What’s going on? This Nature story dives into that question, exploring how we assess scientific novelty and examining why progress increasingly feels like an uphill battle. Among the culprits for the slowdown in progress: mounting bureaucracy, ballooning costs of experimental hardware and the ever-rising learning curve to reach the knowledge frontier. Another explanation: Maybe all the low hanging fruit has been plucked? The bad news is that it does seem as though discovery is slowing, and in the US, budget cuts to research risk making it worse. The good news: Programs to incentivize research are flourishing in many other parts of the world. Whether those programs succeed could reshape the future of innovation.

ScamGPT: GenAI and the automation of fraud, from Data & Society
8,700 words, or about 35 minutes

Artificial intelligence is making many jobs more efficient. That includes the work of criminals. This report from the nonprofit research group Data & Society explores how AI is being used by fraudsters and finds that it’s making scams easier to carry out, more scalable, and harder to detect. Criminals can now clone voices, write persuasive messages, create fake avatars and even mimic customer support chatbots from legitimate companies, all with minimal technical skill. But perhaps the most important takeaway from the report is that stopping this wave of AI-enabled fraud will require more than just better technology. Instead, the authors argue, we need a multipronged response: public education to raise awareness, corporate responsibility to prevent AI from being used for fraud, and policy and legal frameworks to regulate misuse of generative AI. In short, protecting people from AI-driven scams will demand a coordinated, collective effort, something that as yet does not exist.