Dear Aventine Readers,

Our energy supply has always been constrained by a simple reality: for the most part, electricity needs to be consumed as soon as it’s generated. In the last year, that has changed as large-scale batteries now allow generators to store up electrons at a cost low enough to be  competitive with fossil fuels. That’s a revolution that could expand our energy supply, bring down prices and help stabilize our grid. The big beneficiary so far is renewable power, especially solar, which combined with storage to account for 90 percent of all new energy added to the grid in the first three months of this year. Like most revolutions, this one was years in the making.

Also in this issue:

Thanks for reading,

Danielle Mattoon
Executive Director, Aventine

Subscribe

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

The clichéd knock on solar power is that it produces electricity only when the sun shines. And on the brightest days, when panels can make more power than the grid can use, those extra electrons go to waste. The long hoped-for solution has been a battery big and cheap enough to store solar energy, then release it to the grid as needed. 

In the last six months, thanks to a combination of factors ranging from technological advances to shifts in geopolitics, that solution has arrived. As a result, the cost of power from a combination of solar panels and what is called utility-scale energy storage — essentially giant racks of batteries — is now competitive with power generated by fossil fuels in the US, according to new estimates.

That’s a huge achievement. And inside the energy industry, it’s seen as a tipping point in the trajectory of renewable power. An astounding 90 percent of all new power added to the US grid in the first quarter of 2026 came through the combination of solar panels plus storage. This happened despite inflation, global supply chain shocks, a trade war and the elimination of some tax incentives by the Trump administration. 

“Battery storage is probably one of the more revolutionary things that's happened on the electric grid since the invention of the electric grid,” said Seth Feaster, an energy data analyst at the Institute for Energy Economics and Financial Analysis. “It's hard to convey to people just how radical this change actually is.”

In the short term, falling hardware costs are fueling more ambitious energy projects, including some already in the pipeline, like the partnership between Google and Xcel Energy to build a 30,000 mWh battery in Minnesota, which would be the largest grid battery project in the world to date. In the longer term, grid storage could radically remake our national power infrastructure, accelerating the transition to renewable energy while helping stabilize the shaky US grid — making energy more reliable, less prone to blackouts and perhaps even cheaper for consumers. 

So how did we get here? 

Like most revolutions, this one didn’t happen overnight. Decades of research have helped make batteries cheaper and more energy efficient. In the early 2000s, China made a massive investment in battery manufacturing, much of it for electric vehicles, which drove down costs.  One of the most significant changes happened in 2020 when manufacturers shifted away from nickel-manganese-cobalt formulations and toward cells using lithium-iron-phosphate chemistry, which rely on cheaper and more abundant elements. That shift contributed to an approximately 40 percent drop in the cost of batteries, even as lithium prices have gone up. Overall, the price of utility storage per kilowatt hour has plummeted 93 percent since 2010.

There were also factors beyond the industry’s control that converged in the last few years to boost the prospects for batteries. In 2018, the Federal Energy Regulatory Commission issued an order to allow electricity storage systems to compete in wholesale energy markets, but grid operators sued to block the change. In 2020, a court ruling upheld the FERC order, firing the starting gun for a national race to build batteries. 

Additionally, several factors put upward pressure on the price of energy from fossil fuels, making renewable options more attractive. In 2022, Russia’s full-scale invasion of Ukraine raised prices for natural gas, the main fuel for flexible power generation in the US. This year, the US conflict with Iran blocked shipping in the Strait of Hormuz, further driving up energy prices. Finally, there’s now a five-year (or more) wait for gas turbines, which are in demand to power off-the-grid generators for data centers.

One result of all this has been to make batteries more economically competitive. For the first time, the average cost of producing power over the lifetime of the plant (called the levelized cost of energy or LCOE) of hybrid solar-plus-storage facilities in the US is on par with natural gas generators, according to a report published recently by the International Renewable Energy Agency. The LCOE for renewables plus storage in sunny and windy parts of the country is around $102 per megawatt-hour, about the same price for a new gas turbine plant (a gigawatt can continuously power about 800,000 homes). There are now more than 300 solar plus storage plants across the US. 

Another is that we’ve built a lot more battery capacity. The US now has nearly 50 gigawatts of battery capacity on its power network, up from 1.65 GW in 2020, and about half that capacity is coupled with renewable power. Even with that increase in storage, grid batteries are still only about 5 percent of the overall capacity in the US, so we could build “tons more than we have today,” said Wesley Cole, an energy systems engineer at the National Laboratory of the Rockies (formerly the National Renewable Energy Laboratory). 

How utility storage could benefit consumers

The embrace of batteries defies politics and policy divides. In deep blue California, a renewable energy booster, grid storage has seen a 2,100 percent spike since 2019. In freewheeling, market-driven Texas, grid-scale battery capacity increased 4,100 percent between 2020 and 2024. 

“Battery-storage-and-solar has won in both cases, which I think is really fascinating,” Feaster said. “That tells you that it's not just the policy that's doing this, it's the economics.” 

But battery storage can also have a direct impact on consumers, as residents of Texas have experienced firsthand. In 2023, ERCOT, the main Texas power grid operator, issued more than a dozen alerts to their customers asking them to conserve power during extreme heat that pushed the grid to its limit, causing local brownouts. Then, between 2023 and 2025, ERCOT tripled its grid storage, which can respond more quickly than conventional power plants to surges to meet demand.  In 2025 — even though electricity demand surpassed 2023 levels — ERCOT didn’t issue a single alert. 

Another consumer benefit: While electricity prices are rising across much of the US, utility-scale batteries have helped slow their ascent by making the market between generators and grids more efficient. 

How much can battery storage grow?

The US now has enough manufacturing capacity to meet all of its domestic demand for grid-scale storage. (For context, China still controls more than 80 percent of global battery cell production worldwide.) More US players, including some traditional automakers, are getting into the battery game. Ford announced in May that it is launching a grid-scale energy storage subsidiary. General Motors is developing a sodium ion battery for the grid, building on its R&D infrastructure for EVs. Sodium ion batteries are not as energy-dense as lithium ion cells but sodium is far more abundant than lithium, doesn’t need active cooling and could potentially last decades. To date, sodium ion batteries are a tiny percentage of grid storage in the US but the potential savings and worries over lithium supplies could help them grow. 

At the same time, the current generation of lithium ion batteries and their derivatives have yet to bottom out in price, and most of the grid battery projects underway still use them. “Battery packs are still getting cheaper,” Cole said. “As costs come down, the amount of battery storage that makes sense to build increases.” 

The question now is how much more utility storage can grow. Batteries are hitting the same speed bumps that hamper the rest of the energy sector. The paperwork, modeling, legal and environmental reviews to connect a new generator to the power grid can take more than four years. Despite potential delays, the industry is bullish. At the end of 2025 in the US alone, there were more than 2,000 gigawatts of new solar capacity and storage waiting to plug in. 

 Credit: Nerthuz

An overlooked organ could be key to helping us live longer. Directly behind the sternum is a tiny organ called the thymus. Its job: to produce immune cells that protect us from infections and cancer. The trouble is, the thymus steadily shrinks with age, becoming less effective over time. For decades, scientists weren't entirely sure if that mattered. But Nature explains that a 2023 study found that people whose thymus had been surgically removed were nearly three times more likely to die and twice as likely to develop cancer within five years than patients whose thymus had been left intact. Two studies published earlier this year linked naturally shrunken thymuses to higher risks of death, cancer and cardiovascular disease, and poorer responses to cancer immunotherapy. Some researchers argue that these results show only correlation, not causation, but several startups beg to disagree and are working on ways to rejuvenate the thymus as a means of boosting longevity. The technique furthest along, according to Nature, is developed by a startup called Intervene Immune. A growth hormone mixture, injected four times a week, appears to help stimulate the thymus, and the treatment seems to have turned back the biological age of test subjects by an average of 2.5 years. But this trial involves just 10 people. Other startups, including TECregen, Zag Bio, Thymofox, Tolerance Bio, Thymmune Therapeutics and Lightning Biotherapeutics, are developing competing treatments. There are significant challenges for success. For one, the thymus is difficult to target because many of its cells lack distinctive molecular markers. For another, many of the candidate therapies must be injected intravenously, which is a barrier for long-term treatment. Nevertheless, it’s a potentially productive new avenue in longevity research.

Don't panic about China's new Kimi K3 model. At least not yet. Until a few days ago, most people had never heard of Kimi, an AI model developed by the Chinese company Moonshot. Then the arrival of the K3 version of the model late last week was greeted with headlines that claimed it had erased America's AI lead. Not just that: Moonshot plans to release the model's weights — the numerical parameters that encode what it learned when it was trained — on July 27, allowing anyone with sufficient computing power to run or modify the model themselves. Initially, people feared this meant that capabilities equal to Anthropic's Fable 5 would soon be available everywhere, for free. Stock markets were roiled, America was suddenly losing the AI race and national security was in tatters. A few days out, the picture is clearer. Kimi K3 may be good, but Moonshot acknowledges that performance "trails the most powerful proprietary models." Releasing the weights of a model is, as The Economist recently argued, "a logical strategy for laggards" — a way to "woo foreign customers with cheaper models" while the gap persists. But that could change with increased capability: When China does build something genuinely dangerous, Transformer argues, its government may move to restrict distribution. Indeed, Reuters reports that Beijing is considering limits on overseas access to its best models. (Washington, remember, only started paying real attention to frontier model releases when Mythos became potentially dangerous.) Then there's the commercial side. Moonshot doesn't have sufficient computing capacity to keep up with customer demand for the model, but because it's open-weight, cloud providers like AWS and Microsoft Azure will soon be able to host it, potentially pulling customers away from Anthropic and OpenAI. It’s also important to keep in mind that DeepSeek didn't achieve levels of adoption high enough to damage the rise of America's homegrown AI companies as many initially feared, so it remains to be seen whether K3 will dent them either. The one thing this does make absolutely clear is that China is competitive, which could prompt a response from Washington. Whatever the ultimate fallout, it might be that the repercussions of Kimi K3's release prove more profound than the model itself.

Terrorist groups are using AI to fine-tune their operations. New research reveals how extremists are using the latest large language models to design explosives, upgrade weapons and brainstorm attacks. The study, published by Antonia Juelich, an international security researcher at the University of Cambridge, and reported by The New York Times, is based on interviews with 27 former members of Boko Haram. It describes an almost workplace-like approach to adopting AI. Training sessions were delivered through international jihadist networks, with participants learning how to extract the most useful answers from AI systems and work around their safety restrictions. One interviewee explained: "You type in the question or use your voice and [AI] gives you a detailed answer … It is like a human robot! We used it a lot." According to the report, members of Boko Haram have used models including OpenAI's ChatGPT, Anthropic's Claude, Google's Gemini and xAI's Grok. One problem AI companies face in deterring this kind of usage is that many requests are dual-use: In one case, The Times explains, terrorists used the tools to learn how to jump motorcycles higher and farther, in order to leap over a defensive trench around a military complex. It would be almost impossible to prevent a model from sharing such information. More explicitly violent requests — which AI companies try to filter out and reject — were reportedly handled by switching between different AI models, exploiting differences in each company's safety guardrails. Another recent report, published by the Center for Strategic and International Studies, argued that AI will strengthen the operations of terrorist groups in another way: by enhancing their ability to use fraud and deception to raise money to sustain their work.

The cost of healthcare for aging societies might not be as ruinous as we thought. The common assumption is that as current populations age, healthcare costs will explode, fewer workers will be left to pay for them and economic growth will suffer. Not so fast, according to three new research papers described by The Economist. First, a paper by David Cutler and Lev Klarnet from Harvard University describes how America spent $1 trillion less on healthcare in 2024 than forecasters expected in 2010. Hospital costs have risen much more slowly than anticipated, with annual growth falling from 2.3 percentage points above inflation between 2000 and 2010 to just 0.5 percentage points since 2011. Second, a paper by Liran Einav of Stanford University and Amy Finkelstein of MIT looked at Medicare survey data between 1992 and 2019 and found that expected lifetime Medicare costs rose by just 6 percent, compared with a 14 percent increase in lifetime Social Security costs. People are living longer, but much of that extra time is being spent in relatively good health. That still leaves the broader economic challenge of supporting aging populations with shrinking numbers of working-age adults. On this point, a third paper, authored by Daron Acemoglu and David Autor from MIT among others, points to a more optimistic scenario. Drawing on several decades’ worth of cross-country and US local labor market data, it shows that aging societies aren't poorer: Areas with steeper fertility declines show higher productivity growth among working-age people. The authors argue that necessity has pushed these societies to adopt labor-saving technologies, and doing so offsets the drag from demographic change. None of this research guarantees an easy ride for the economy in the years ahead, but it’s a welcome alternative to what has become the alarmist default narrative.

Who cleans up after the vibe-coding party? from The Financial Times
5,200 words, or about 20 minutes

Drunk on the excitement of vibecoding, wannabe software engineers have created the software equivalent of garbage piling up on the side of a road, presenting a huge challenge for those who have to clean it up. Tools like Anthropic's Claude Code and OpenAI's Codex have enabled a flood of software contributions from excitable new software developers, including updates to the thousands of open-source software projects on which modern life is built. Yet the people who maintain those important projects — almost all of whom are volunteers — have been overwhelmed by the volume and low quality of these new contributions. So much so that, in some cases, maintainers have shut down outside contributions entirely, because the potential for causing damage is too high and the burden of quality control for AI-generated code too great to make the endeavor worthwhile. This story describes the problem, as well as the broader impact of AI coding tools and the way they are changing the work of software development. Perhaps the most important point it makes is this: Both tech companies and the public celebrate the act of building things. Simple maintenance needs more respect. 

Inside America’s Shoplifting Lab, from Slate
7,000 words, or about 28 minutes

Come for the insight into the techno-dystopian realities of crime reduction; stay for the skewering of the US shoplifting crisis. As some would tell it: The nation is undergoing a soaring incidence of petty theft; some reports suggest it may be up as much as 93 percent since the Covid-19 pandemic. As a result, worried retailers are flocking to the Loss Prevention Research Council lab at the University of Florida for solutions. The lab is a test bed for hundreds of anti-theft solutions, including AI-equipped cameras, tiny RFID tags, locked cases, weight-sensing shelves, a holographic salesperson and … a lot of signs. And retailers are flocking: The lab works with at least 190 retailers and 170 security vendors, all of whom pay annual membership fees for access to its research and advice. And therein lies a rub: As the author, Alexander Sammon, explains, the data underpinning the crisis narrative around shoplifting is largely generated by the retail industry, and much of the research around the fight back is generated by a lab funded by those same retailers. Numbers around the incidence of shoplifting are highly contested, and some reports suggest it may actually be flat or falling. Regardless of the numbers, retailers want protection, which is good business for the loss prevention lab. 

Why We Demand Perfect Machines Yet Tolerate Human Carnage, from Noema
3,500 words, or about 14 minutes

If there's a sentence that should stick with you from this piece, it's this: "Data doesn't cure fear." It's something that the self-driving car industry is grappling with. Data shows that autonomous vehicles are dramatically safer than those driven by humans: Waymo reports that its vehicles have been involved in 94 percent fewer crashes that cause serious injury or worse than human drivers on the same roads. But consumers still don't trust them. Each and every failure, even when no human is injured, is highly scrutinized in the media. In this piece, Jonathan Slotkin, a neurosurgeon, argues that this asymmetry is partly due to illusory superiority (nearly everyone thinks they're a better than average driver, so machines are judged against an idealized self) and algorithm aversion (we forgive human error but not machines). He also points out that the healthcare industry has been on a decades-long journey to combat a similar challenge. Clinical safety and patient experience are correlated but distinct problems, and one of the best ways of maintaining people's trust has proven to be radical transparency. In 2001, the University of Michigan Health System stopped routinely fighting malpractice claims and began proactively identifying medical errors, explaining them openly to patients, apologizing when appropriate and offering compensation when the system was at fault. As a result, claims and lawsuits fell. The article argues that similar transparency might be what the self-driving industry needs to be fully accepted.