Dear Aventine Readers, 

This week we dive into an evolving Substack discussion over China's role in the world's energy future. When it comes to green energy, China is a contradiction: In recent years it has installed both more renewable power capacity than the rest of the world combined, and also more coal plants. And, as a few Substackers point out, China's motives are driven not by green energy targets but by energy security and strategic control. So while the climate benefits of China's green energy push are substantial, they come with downstream consequences. 

Other Substack highlights: 

Thank you as always for reading. 

Sincerely, 

Danielle Mattoon
Executive Director, Aventine

Subscribe

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

Occasionally when tracking a topic on Substack, what starts with a take or two evolves into a fully developed story, told from multiple angles. This was the case when people started debating China's role in the future of energy in the fall. A handful of well-known commentators held up China as an example of what other nations should aspire to, pointing out that in recent years it has installed more renewable power capacity than the rest of the world combined. Noah Smith wrote that the country was “quietly saving the world from climate change.” Paul Krugman wrote that it would “rule renewable energy.” Mainstream outlets soon followed. The Financial Times considered how Chinese companies now “dominate many clean technology industries”; The Economist called it “the world’s renewable-energy superpower” and asked whether it could become a climate superpower too. 

The backlash came quickly. This was “greenwashing with Chinese characteristics,” according to Seaver Wang, Ted Nordhaus and Vijaya Ramachandran writing at The EcoModernist, a Substack published by the Breakthrough Institute, an environmental research center. Robin Brooks, a senior fellow at Brookings, dismissed the narrative as an “electrification mirage.”

The root of this particular pushback is coal. Because while yes, China has been a leader in building renewable energy capacity, it has also spent the last decade building more coal power capacity than the rest of the world combined. Brooks argued that, at the highest level, “what matters is where the electricity comes from.” Based on his analysis of the current numbers, he writes, “It’s true that solar and wind are growing in importance, but … their shares in total electricity generation [in China] were five and nine percent, respectively, in October 2025 versus 67 percent from thermal power.” (This is somewhat misleading as a choice of statistic given that “thermal” includes oil, gas and renewables that get burned, such as biomass. In response, Chinese journalist Yuzhe He and energy investment adviser David Fishman noted on He’s Substack that coal accounted for more like 54.8 percent of electricity generation in China at the end of 2024.)

Nordhaus went on to point out that much of China’s industrial complex remains fundamentally coal-powered. Aluminium, silicon, magnesium, nickel and steel — many of them building blocks of the clean energy economy — are still produced largely using fossil energy. While cleaner production pathways exist, Nordhaus argued, China has been slow to adopt them, and in some cases has simply ignored them, undercutting higher-cost green producers in Europe and North America.

“The reality is that coal remains the ‘ballast rock’ of China’s energy supply,” the authors at The EcoModernist concluded. 

What do the numbers say? 

It turns out, according to data, that both narratives can be true at once. On By the Numbers, a Substack that focuses on the data for environmental and climate issues, Hannah Ritchie, a researcher at the University of Oxford, took a look at the various factors contributing to China’s energy profile. The data was collected before the end of 2025 and her findings showed that even with an increase in the number of coal plants, total CO2 emissions from China seemed likely to flatline last year in large part due to the expansion of clean power. 

Thanks to the addition of roughly 430 terawatt-hours of new non-fossil electricity last year (almost the entire annual electrical output of Germany), emissions from electricity generation fell by around 2 percent by the end of the third quarter of 2025, despite demand rising by more than 6 percent. At the same time, many of China’s coal plants operate as so-called peaker plants, which are used only when extra supply is needed. Emissions from transportation also declined, thanks to the rapid adoption of electric vehicles. More than half of new cars sold in China are now electric, as are around 10 percent of all vehicles on the road. In the United States EVs account for roughly 10 percent of new sales and about 2 percent of the total fleet.

The catch, as Nordhaus also pointed out, is industry. Emissions from heavy manufacturing have risen enough to offset gains, leaving overall emissions roughly flat rather than falling.

How does China stack up against the rest of the world? Michael Hill, a policy researcher in the UK, dug into the many and varied ways it’s possible to draw those comparisons in Notes on Growth, a policy and economic Substack. In absolute terms, China is the world’s largest emitter. On a per capita basis, it sits below the United States and roughly in line with parts of Europe. In terms of renewable deployment, China has installed about twice as much wind and solar capacity as the rest of the world combined over the past two years, but that reflects its enormous scale as much as its ambition. Look instead at carbon intensity — emissions per unit of electricity generated — and China performs worse than most developed economies as well as the global average.

If there is a pattern here, it’s this: China’s growth is increasingly powered by clean energy, but at its core, its energy system is overly reliant on fossil fuels.

Behind China’s push for green technologies 

To understand what’s happening here, Hill contended, you have to understand China’s push to secure its own power supply. The country is the world’s fifth-largest oil producer, but it consumes roughly twice as much as it produces — a strategic vulnerability that, argues Hill, helps explain the ferocity of China’s push to manufacture electric vehicles. Electrification reduces exposure to imported oil and enhances energy sovereignty. Similar logic applies across the energy system. China sits on vast coal reserves and produces around 90 percent of the coal it consumes. When energy security and climate action conflict, China chooses energy security. Coal remains central because it is domestically available, reliable and protected from geopolitical turbulence. 

In another post, Hill challenged the assumption that China’s dominance of green supply chains reflects a masterful climate-industrial strategy. The reality, he suggests, is that China was largely lucky. The country’s industrial growth coincided with the rise of clean tech; it identified priority industries that were in the ascendance and scaled them aggressively; and it had a workforce that included both highly skilled workers and cheap labor. 

One result of all this is that much of the world’s energy transition now runs through Chinese supply chains. Solar panels, wind turbine components, batteries and grid equipment are all heavily dependent on Chinese manufacturing. Nuclear power is one of the few major low-carbon technologies that can be built without Chinese parts. 

But there’s another dimension to this story. Bruno Maçães, a political scientist writing on the Substack World Game, argued that the next great industrial push will be building infrastructure for artificial intelligence, and countries that can supply vast amounts of reliable power to train and run large language models will end up in the strongest geopolitical position. 

China has huge volumes of renewable capacity coming online, along with a vast fleet of coal plants operating below their full potential that can be ramped up quickly. The US, by contrast, faces grid constraints, slow permitting and political resistance to new forms of power generation. Maçães warns that America could be heading toward energy scarcity, just as demand from AI accelerates. 

Gerard Reid, a partner at Alexa Capital, is even more direct on his Substack. “Energy trade shifts from fuels bought on spot markets to capital equipment financed over decades, locking in standards, supply chains, and influence,” he argued. “This is not climate policy dressed up as strategy. It is long-term geopolitics executed through energy.” In the 20th century, power belonged to those who controlled oil. In the 21st, Reid suggests, it will belong to those who build the cheapest, most resilient, most scalable electrical systems. 

Demis & Dario go to Davos, from The Change Constant

At the World Economic Forum in Davos last week, Demis Hassabis, CEO of Google DeepMind, and Dario Amodei, CEO of Anthropic, sat down for a public conversation about progress toward AGI, AI’s impact on jobs, how the AI industry might sustain itself financially and much else besides. The full discussion is worth watching, but this write-up from The Change Constant is a useful shortcut. One of the biggest takeaways is the tension both men openly acknowledge between the pace of AI development and the ability of society to adapt. For commercial and geopolitical reasons, both men believe they have little choice but to keep pushing AI forward at an exponential pace. They also seem to believe that the world would be better off if progress were slower, giving societies more time to adjust to the disruption AI is already creating. It’s at once both comforting and unsettling to hear two of the people most responsible for shaping the future of AI articulate this tension so plainly.

The Internet is Turning Against OpenAI, from AI Supremacy

Meanwhile, OpenAI, another giant of the AI boom, seems to be having a rough start to the year. In this post, Michael Spencer walks through the company’s increasingly delicate market position, troubling finances and questionable product strategy. A few points stand out. OpenAI’s share of enterprise spending on large language models has reportedly fallen from about 50 percent in 2023 to roughly 27 percent today, while Anthropic’s has surged from 12 percent to around 40 percent over the same period. One analyst estimates that OpenAI is now spending $3.30 for every $1.00 in revenue it generates. OpenAI is said to be doing around $20 billion in annual revenue, compared with roughly $9 billion at Anthropic, although the former is six years older. Spencer also points to mounting external pressures like Elon Musk’s potentially ruinous lawsuit against the company and a string of product moves — from leaning into “adult mode” to ads and consumer hardware — that look, to him, a little desperate. Taken together, Spencer argues, users and enterprise buyers are starting to view OpenAI more skeptically, and to look more seriously at alternatives from Google and Anthropic. 

What’s Wrong with NIH Grants?, from Statecraft

A lot, according to Mike Lauer, the former deputy director for extramural research at the National Institutes of Health, in this Q&A. The core problem, he argues, is hyper-competition. Over time, the rate of successful grant success has collapsed, from around 60 percent to as low as 10 percent, turning funding into a painful lottery. At the same time, there has been a ballooning of bureaucracy: Proposals now run to hundreds — sometimes thousands — of pages, consuming enormous amounts of time for both applicants and reviewers. Lauer sketches several ways out: simpler applications; lighter reporting requirements; fewer small grants; more ambitious “block grants” to fund high-risk, high-reward science. The problem is that reforming a system this entrenched will be far harder than acknowledging that it’s broken.

Capital in the 22nd Century from Philosopher Count

In this essay, Stanford economics researcher Philip Trammell and podcaster Dwarkesh Patel explore a future in which AI and robotics become so capable that human labor is no longer needed. In that world, capital replaces labor rather than complementing it, and the returns from owning machines soar, while wages fall toward zero. To frame the argument, Trammell and Patel revisit the work of Thomas Piketty, who famously argued a decade ago that economic inequality tends to increase indefinitely through the generations. At the time, many economists dismissed his work. In a world of automation, Trammell and Patel suggest, Piketty may turn out to be right.

AI isn’t “just predicting the next word” anymore, from Clear-Eyed AI

If the way you think about large language models still boils down to “they just predict the next word,” this post is worth a read. That description has become increasingly outdated as AI has progressed over the last couple of years. Steven Adler, a former OpenAI safety researcher, explains how leading models have moved beyond simple next-token prediction toward something closer to path-finding. Modern “reasoning models” are trained heavily using reinforcement learning, which encourages them not just to produce plausible text but to explore many possible solutions through trial and error before selecting one. The result is a system that doesn’t really autocomplete sentences so much as develop a more general capacity for problem-solving that it can transfer across tasks. And it’s an important distinction, because it emphasizes how these systems have become more complex than previous metaphors may have us believe.

How Did TVs Get So Cheap? from Construction Physics

In a world where almost everything seems to be getting more expensive, there is one particularly strong exception: the television. Adjusted for screen size, resolution and price, TVs are now more than 90 percent cheaper than they were 25 years ago. This post unpacks how that happened. The biggest driver, it turns out, is scale. The transistor-filled glass sheets used to make display panels have grown dramatically in size, and the cost of the machinery required to manufacture them has risen only modestly in comparison. The result: far more screen per dollar. But that’s only part of the story. Manufacturing equipment has become more efficient. Yields have improved as processes stabilized. Materials have advanced, reducing defects and waste. This post is a reminder of what sustained industrial improvement can achieve, at least some of the time.

Not For Human Consumption, from Vector Culture

In December, at a San Francisco venue called Frontier Tower, guests gathered for what was billed as a “Chinese Peptide Rave.” Between drinks and techno sets, attendees took part in mix-your-own peptide workshops and learned injection techniques. Yes, really. This post dives into the fast-growing gray market for peptides: experimental compounds sold online with labels that insist they are “not for human consumption,” even as thousands of people inject them in pursuit of longevity, weight loss, muscle gain, cognitive enhancement, or some combination of all of those. The piece explains how this market can exist at all, tracing the regulatory loopholes that allow billions of doses to flow in from overseas manufacturers while staying just outside formal drug approval pathways. The FDA has begun sending warning letters and stepping up enforcement, but demand continues to surge, driven by communities of self-experimenters convinced they’re on the frontier of medicine. It’s a fascinating, if troubling, look at what happens when biotech innovation, lagging regulation and Silicon Valley self-optimization collide.

How to party like an AI researcher, from Jasmine’s substack

When NeurIPS — the Conference on Neural Information Processing Systems — launched in 1987, it drew about 600 attendees. This year, more than 26,000 people registered. Along the way, the world’s premier AI research meeting has transformed into something closer to a tech festival. In this dispatch from the event in December, Jasmine Sun moves between earnest conference debates about post-AGI politics, expo halls where quant firms try to recruit PhDs before the AI labs grab them and yacht parties stocked with champagne and canapés. The academic conference, she suggests, has become a full-blown social circuit. “NeurIPS feels like one long holiday party,” she writes, “where grad students from around the world break from tuning their hyperparameters to drink champagne on some tech company’s dime.”