Dear Aventine Readers, 

One byproduct of data centers' unquenchable thirst for energy is the emergence of an oddball mix of technologies designed for long term energy storage. Rusting batteries, giant balloons, underground caverns of compressed air.  All these are being put to use as ways to shore up the grid. The question is: Will any of them become cost effective enough to break into mainstream use? 

Plus:

Thanks for reading! 

Danielle Mattoon 
Executive Director, Aventine

Subscribe

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

A two-hour drive north from Los Angeles, something unusual is happening underground.

A company called Hydrostor is hollowing out vast caverns in the bedrock of Kern County, California, with a plan to turn them into giant storage tanks of reserve energy. When there’s spare power on the local grid, the company will use it to compress air into the caverns, then seal it with water. When demand for power spikes, air can be released to drive a turbine that generates electricity. 

This is one of a string of novel technologies intended to fill a gap in our electricity grids. Lithium-ion batteries, which today dominate grid storage, can typically store and provide only a few hours of energy. Gas peaker plants can run indefinitely, but they burn fossil fuels, are inefficient and obviously don’t help capture spare resources from renewables. In between sits what’s known as long-duration energy storage, or LDES, an umbrella term for technologies that can soak up spare electricity and then release it over periods ranging from several hours to several days. As Doug Vine, director of energy analysis at the Center for Climate and Energy Solutions put it, LDES "takes renewable resources and makes them into clean, firm resources."

The best-known version is pumped hydroelectricity (or hydro), a process by which surplus electricity is used to force water uphill into a reservoir, where it is stored until demand returns and the water is released through a turbine to generate power. But pumped hydro depends on geography and a raft of these newer technologies don’t. 

Giant pillows inflated with carbon dioxide. Batteries that work by deliberately rusting iron pellets. Compressed air pumped into underground voids. These are some of the technologies vying for a piece of the energy storage action. 

Less than five years ago, LDES methods like these would have been dismissed as unproven and too expensive. Now, thanks to AI’s thirst for electricity, hyperscalers like Google and Meta are signing deals to put them into use. The question is whether the new attention and investment will be enough to drive down costs and reduce the technology’s risk to the point at which it becomes a routine part of the grid.

Making the grid work harder

Curtis VanWalleghem, Hydrostor's co-founder and CEO, spends a lot of his time talking to utility companies. From those conversations, one thing has become clear: "Load growth is back with AI data centers." Goldman Sachs recently predicted that US data center demand for electricity will more than double to 66 gigawatts by 2027, up from 31 gigawatts in 2025, driven largely by AI.

That is forcing utilities to find new ways to expand capacity. LDES, VanWalleghem argues, lets them squeeze more out of what they already have. He estimates that only 60 percent of the grid’s transmission capacity is being used; with strategically located storage such as LDES, this can be increased to over 80 percent.

Hydrostor's Kern County plant is by far its largest to date. It will be able to deliver 500 megawatts — enough for 400,000 homes and more than twice the amount of its previous biggest facility — for eight hours. The infrastructure required to build the plant is mostly off-the-shelf, borrowed from mining and conventional power generation, and the company claims it can build a facility pretty much anywhere.

Hydrostor is not the only LDES company working directly with utilities. Form Energy, based in Somerville, Massachusetts, whose batteries store electricity by oxidizing iron pellets, is working with utilities in Minnesota, Maine and Georgia on projects that will have a collective capacity of over 100 megawatts. Energy Dome, an Italian startup, is building a 20-megawatt facility for Alliant Energy in Wisconsin; it uses energy to condense CO₂ into a liquid, then expands it back into a vast pillow-shaped dome to drive a turbine. 

But there is a catch to all these technologies, said Zeenat Hameed, an energy storage analyst at Wood Mackenzie: cost. For now, she said, only compressed air comes close to being cost competitive with lithium-ion, and even then only above eight hours of discharge, which lithium-ion doesn’t yet stretch to. VanWalleghem concedes that the 500-megawatt Kern County project will be the first Hydrostor facility that makes economic sense, a result of its scale.

Powering up data centers

AI companies are less concerned about cost than utilities are. The value of computational power is so large compared to the cost of electricity that “there is no premium on electricity that will force [the companies] to shift their [approach to electricity] consumption,” said Dharik Mallapragada, a professor at NYU Tandon School of Engineering who models clean energy systems. 

Waiting lists for new grid connections for data centers in the United States now run to about five years on average, according to research from Lawrence Berkeley National Laboratory. The most popular stopgap measure has been to use on-site gas turbines, but the turbines have their own waiting lists, with lead times reportedly stretching to seven years. Combine that with hyperscalers' lingering climate commitments — Google recently reaffirmed its desire to be carbon-free by 2030 — and these companies are now exploring how they could build long-duration energy storage facilities to get their data centers online faster, even at prices at which utilities might balk.

“The utility industry historically has been maligned for being extremely risk averse,” said Mateo Jaramillo, CEO of Form. Now, he said, “Google's there saying, ‘Yeah, we'll support this project.’”

Indeed, Google has struck a deal with Form and Xcel Energy, a Minnesota utility. Form will build 300 megawatts of storage at Google’s Pine Island data center campus, which will be combined with 1,400 megawatts of wind and 200 megawatts of solar generation capacity. The first batteries are expected to be delivered in 2028. Google has also entered into a partnership with Energy Dome, but has not yet announced details of any projects. Meta, meanwhile, has cut a deal with Noon Energy, which produces energy storage systems built around technology similar to fuel cells, but which use CO₂ in place of hydrogen. Meta has reserved a total of 1 gigawatt of storage from the startup, but the project will start with a 25-megawatt proof of concept in 2028. And like Google, Crusoe, an AI-focused data center developer, is working with Form, with 12 gigawatt-hours of capacity reserved for delivery starting in 2027.

Mallapragada and VanWalleghem both see this as a bridging solution for the tech companies while they wait for permanent connections to the grid. Being properly tied into the wider grid would, in the long run, offer Big Tech more reliability and more flexibility than building their own independent electricity infrastructure, they said. For now, going it alone is simply faster.

How LDES goes mainstream

The bigger question is whether this wave of interest is enough to push LDES into long-term routine use.

Hameed is skeptical. One of the reasons lithium-ion became a feasible grid technology, she pointed out, was a glut of manufacturing capacity, much of it built on misjudged forecasts of electric-vehicle demand. The resulting price collapse made batteries cheap enough for utilities to use them at scale. Without a similarly dramatic fall in price, she expects LDES technology to remain a niche. Her own forecasts predict lithium-ion holding roughly 85 percent of the global grid storage market through 2034, with compressed air capturing just 3 percent.

Jaramillo contends that his order book is proof that there’s market demand. Form is committed to building 75 gigawatt-hours of storage, and the company has said it is sold out of factory output through 2028. “The fundamental bottleneck is not demand at this point,” he said. “It really is manufacturing.” What’s currently unclear is whether existing deployment volumes will be enough to help companies reduce prices fast enough for broader adoption. Mallapragada argues that it will depend on how intensively the systems are used: If LDES systems are built but used only sparingly, because sun shines and wind blows, companies may not learn enough about failure rates, maintenance needs or durability to improve their products.

Government policy could tip the balance. Vine points out that California, Massachusetts, New York and most recently Virginia have all introduced LDES procurement mandates. Virginia — already home to the largest concentration of data centers in the world, with many more on the way — has set the most ambitious target, requiring 4.5 gigawatts of LDES capacity by 2045. Geopolitics could also help. The cheapest lithium-ion batteries are still made in China, and if Washington decides it wants grid storage that is less dependent on Chinese battery supply chains, LDES technologies could benefit.

None of this guarantees that rusting batteries, giant pillows and underground caverns will become the immediate future of the grid. Yet over the longer term, LDES seems likely to become increasingly necessary. Jaramillo points out that while AI may be accelerating growth in electricity demand, other trends, such as the electrification of vehicle fleets and industrial processes, will continue to put strain on the existing grid. “I don't think that goes away, even if … demand from hyperscalers softens a little bit,” he said.

Workers in China manufacture solar panels for export. Getty

China’s solar industry has become a victim of its own success. After spending years building the world's dominant solar manufacturing base, Chinese producers are now facing a painful reckoning. Three problems are converging at once, reports The Economist. The first is severe overcapacity. China's factories can now produce around 1,000 gigawatts' worth of solar panels annually, yet the entire world installed only about 600 gigawatts in 2025. The second is that China's domestic market is saturated. The country's grid is already struggling to absorb the vast quantities of solar power being produced, and new installations in the country are expected to fall by as much as 43 percent this year. The third is geopolitics. Tariffs, industrial policy and national security restrictions in the West are making it harder for Chinese firms to export their way out of the problem. The result is a glut of hardware and a brutal price war, with panels selling for less than the cost of making them. Since 2024, more than 40 Chinese solar firms have gone bankrupt, been acquired or been delisted from stock exchanges. Meanwhile the five largest manufacturers have collectively shed around a third of their workforce. Attempts by the industry to stabilize itself through production quotas and price floors have largely failed, and government support has evaporated. The conflict in Iran has boosted demand and lifted exports somewhat, but that is unlikely to solve the industry's deeper problems. Stronger demand from overseas markets could help solve the problem, but if that doesn’t materialize, a painful shakeout of the industry is likely to be what realigns supply with demand. 

Wastewater is starting to heat homes. Every hot shower, toilet flush, laundry and dishwasher cycle sends heat down the drain. Increasingly, cities are finding ways to recover it. The idea is to use heat pumps to extract thermal energy from wastewater and transfer it into separate water loops that provide heat to nearby buildings. The new Canadian publication, Be Giant, describes a growing number of projects embracing the approach. Vancouver's False Creek neighborhood was an early adopter, becoming home to North America's first large-scale raw sewage heat recovery system when it opened in 2010. Today, it provides heating to around 10,000 residents. Now there’s a new wave of projects. A major Vancouver development is building a $30 million district energy system that will draw heat from a downtown sewer main to serve 6,000 apartments across 11 towers. Toronto Western Hospital expects a wastewater project that is part of its expansion to provide 90 percent of its heating and cooling needs. And a planned carbon-neutral community on a former airport site outside Edmonton will combine wastewater heat recovery with geothermal boreholes to serve 30,000 people. The idea is spreading south, too. Canary Media reports that in Denver, a planned thermal energy network will connect downtown buildings through an underground water loop that draws heat from multiple sources, including sewage, geothermal wells and conventional heat pumps. The projects are a reminder that a fair amount of energy is hiding in plain sight; we just have to figure out how to use it. 

Brain-computer interfaces are getting regulatory approval. In China, at least. A device called NEO, developed by Shanghai startup Neuracle Technology in partnership with Tsinghua University, has been approved for clinical. According to MIT Technology Review, it is the world's first invasive brain-computer interface to receive such approval, potentially opening the door for thousands of patients to benefit from the technology. NEO is designed for people with paralysis caused by spinal cord injuries. The device works alongside brain-controlled robotic gloves that help patients rehabilitate hand function. One participant, Dong Hui, who could barely lift his arm and had lost the use of his fingers, regained the ability to write his name after 11 months of treatment. So far, Neuracle has conducted 36 clinical studies of the system. Part of the reason it reached approval ahead of competitors may be its relatively conservative design: Unlike more invasive systems such as Neuralink’s, NEO sits on the brain's outer protective membrane rather than penetrating directly into the cortex. The device also benefited from strong government backing. Chinese regulators placed it on an accelerated approval pathway and authorities have already begun integrating the technology into the country's health insurance framework. In fact, China’s ambition to get this sort of hardware deployed may be the biggest milestone of all. Brain-computer interfaces have long been the preserve of labs and clinical trials; China may now be showing the rest of the world what it means to put them into practice.

Easier to start, harder to win, from The Economist
5,000 words,or about 20 minutes

This wide-ranging farewell essay from The Economist's outgoing defense editor argues that warfare is undergoing a "transparency revolution." Across conflicts from Ukraine to Iran, advances in sensors, precision weapons, communications networks and AI are making it easier than ever to find targets and strike them. Drones are the most obvious example. In Ukraine, they’ve helped create vast "kill zones" stretching miles behind the front lines, making large-scale ground warfare extraordinarily costly. They’ve also undermined the advantage of air superiority. The decades-old cornerstone of military strategy is now less useful because mass-produced drones create a constantly shifting layer of risk below 13,000 feet. But it isn’t just drones. Increasingly sophisticated surveillance means that concentrations of troops, vehicles and equipment are easier to detect than ever before. And AI systems can reportedly identify thousands of potential targets each day — far more than human analysts can, and in some cases more than militaries can realistically attack. The essay argues that these technologies create the illusion that victory should be easier. But ongoing wars suggest the opposite: These technologies seem to be making it harder to make decisive strikes that lead to victories. 

Our Warming Planet Is a Petri Dish for New and Deadly Microbes, from The New Yorker
5,100 words or about 25 minutes

Any story that begins with flesh-eating bacteria is unlikely to be all sunshine and rainbows. So it is with this piece, which explores how climate change is reshaping the microbial world in ways that could pose growing risks to human health and entire ecosystems. Take your pick of worrying places to start. Warming oceans are helping dangerous bacteria expand into new regions. Thawing glaciers and permafrost are releasing dormant microbes and antibiotic-resistant genes that have been locked away for decades or even millennia. Rising temperatures on land are helping fungi adapt to warmer conditions, undermining one of humanity's natural defenses against infection: our relatively high body temperature, which currently protects us from many fungal diseases. Microbes are evolving alongside the changing climate, adapting to new environments and exploiting new opportunities. And while scientists are racing to monitor how bacteria, fungi, viruses and other microorganisms are responding as the planet warms, the scale of the task is daunting. “The microbes will be fine,” one researcher told The New Yorker “They are running the planet, and they will continue to run the planet.” It’s the rest of us who need to worry.

The Painful Truth About Long Covid, from Wired
7,700 words, or about 30 minutes

If there's one thing this piece makes clear, it's that long Covid is one of the most contested illnesses in modern medicine. Even defining it is difficult. The National Academies of Sciences, Engineering, and Medicine, for example, use an incredibly broad definition requiring at least one persistent symptom following a Covid infection, whether that infection was recognized at the time or not. Uncertainty is perhaps even worse when you get to potential treatments. A systematic review published in The BMJ found that cognitive behavioral therapy and exercise have the strongest evidence behind them. Yet those approaches are deeply controversial among many patients, and the author notes that they were at odds with research discussions at last year's Long Covid International Conference. Part of the challenge is that long Covid inherited decades of cultural baggage from conditions such as myalgic encephalomyelitis and chronic fatigue syndrome. Some critics dismiss these illnesses as primarily psychological, but many patients and advocates push back against that idea. The result is a polarized research environment. The article describes patients who report significant improvements through various forms of "brain retraining" therapy. It also recounts death threats directed at researchers who support those approaches. The piece doesn’t pretend to resolve these disputes. But it does point out a major sticking point: Long Covid patients want better evidence about what works and what doesn't, but the atmosphere surrounding the condition may be making it impossible to produce that evidence.**