Photo by Steve Johnson on Unsplash

If your mental picture of China is still stuck on thick smog, cheap goods, and crumbling ghost cities, Rich Turrin is here to bring you up to speed on what’s happening in China. Speaking from where he now lives in Shanghai, Turrin says, “What I see in China…is a nation with a full-on, 100% focus on technology…from AI to thorium reactors to constellations of supercomputers shot into space.”

His central claim is that AI leadership won’t be decided by the flashiest model, but by who can deploy it at scale with the lowest real-world cost. In other words: kilowatts, not code—and China has spent years building the infrastructure and energy base to capitalize.

A “Whole-Nation Approach” (That America Once Mastered)

Turrin sees China operating with what he calls a “whole-nation approach,” echoing the industrial-policy mindset the U.S. embraced during its peak technology surges. The difference today, he suggests, is not capability but coordination and persistence. China is trying everything, and it is doing so at scale—AI, advanced manufacturing, new energy systems, and even space-based computing concepts.

That matters because technological leadership is rarely just about having the best idea. It’s about sustained follow-through: training people, deploying systems, iterating quickly—and crucially, building the physical capacity to run those systems. The more AI becomes compute-intensive, the more it becomes an energy-and-infrastructure story.

China Has the Power: Hydro, Nuclear, Renewables

If “kilowatts, not code” sounds abstract, China’s power buildout makes it concrete. China has expanded electricity supply on multiple fronts at once—mega-hydro, nuclear, and record wind/solar—while simultaneously building the long-distance transmission needed to deliver that power to coastal industrial centers.

A few indicators underscore the scale:

• Total capacity: China’s total installed power capacity has surpassed roughly 3,400 GW (2025)—a foundation that supports energy-intensive growth sectors, including AI data centers.
• Total generation: China now produces roughly twice as much electricity annually as the United States, giving it a clear edge in overall output—an advantage that matters when training and running models becomes a persistent, economy-wide load.
• Mega-hydro and storage: China continues to push large hydro and pumped storage in the west. One emblematic example is the planned Medog hydropower project in Tibet (~60 GW), which—if completed as described—would be the largest hydropower station in the world, several times larger than Three Gorges. Projects like these are not just about cheap electrons; they also provide grid balancing for variable wind and solar.
• Nuclear as baseload: China is the world’s most active builder of new nuclear plants, with dozens operating and more than 20 under construction, and policy documents still frame nuclear as a “pillar” of low-carbon baseload power toward long-term decarbonization goals. New builds span large coastal reactors and advanced designs (including high-temperature gas-cooled and small modular concepts).
• The grid to connect it all: China’s ultra-high-voltage (UHV) transmission buildout is designed to move massive amounts of hydro, wind, and solar power from regions like Tibet and Inner Mongolia to eastern load centers—exactly the kind of backbone you need when electricity demand grows in big, lumpy increments (think: industrial parks, EV charging, and data-center clusters).

For AI, this is the unglamorous competitive edge: it’s not just about having GPUs. It’s about having enough reliable power—and a grid capable of delivering it—so compute can scale without becoming a national bottleneck.

Chip Bans Haven’t Stopped China’s AI

On the U.S.–China AI rivalry, Turrin rejects the idea that chip restrictions will “kneecap” China. He points to a quote from Nvidia CEO Jensen Huang: “Nvidia went from 95% market share in China to zero. I can’t imagine any policymaker thinking that’s a good idea.”

But the deeper issue isn’t Nvidia’s revenue. It’s the mistaken belief that blocking top-tier chips blocks model progress. “Chip bans do not stop or prevent China from developing large language models,” Turrin says. They force China to spend more—because training takes longer on weaker chips and consumes more electricity.

This is where the “kilowatts, not code” framing bites: if you can’t buy the most efficient compute, you can still get there by brute force—more machines running longer, drawing more power. And China, he argues, has planned for that. While many Western countries worry about grid constraints, China has expanded generation capacity for decades. So when AI becomes more power-hungry, China treats it as “a minor inconvenience” because “they have tons of electricity to burn.”

In other words: the ban changes efficiency, not inevitability—and it shifts the competition toward energy capacity and infrastructure resilience.

Open Source: China’s Quiet Attack on the Subscription Assumption

Even more disruptive than model rankings is distribution. Turrin argues the West implicitly assumed global AI would look like global software: dominant U.S. platforms selling subscriptions to everyone else. But affordability breaks that story.

He cites surveys indicating that across the developing world, many leaders admire U.S. AI but can’t justify the price tag. “When surveyed,” he says, “artificial intelligence managers across the developing world—70% said, ‘Hey, we love America’s AI, but we can’t afford to pay for it.’”

China’s counter is open source. “What it did was to make DeepSeek models, Qwen AI models, and countless others open-source and available to anyone for free,” Turrin says. “If that doesn’t sound like a disruptor, nothing is.”

And he pushes back on a common fear: using these models doesn’t necessarily mean sending data to China. “You can run them on your own cloud instance,” he notes, meaning organizations can keep data local while still benefiting from the model.

If open source becomes the default, the question shifts from “Who has the best AI?” to “Who captures the value?” And it raises a parallel, very physical question: who can afford to run AI widely? Free models still require compute—and compute still runs on kilowatts.

The AI Bubble: Financing Hype Meets a Productivity Reality Check

Turrin is not anti-AI—he’s anti-hype. He argues a bubble is forming because investment expectations are running far ahead of enterprise reality. He points to an MIT paper: “95% of AI implementations are failing.”

He’s careful with the implication: high failure rates don’t prove AI is useless. They prove that meaningful productivity gains require workflow redesign, training, governance, and business-model change. “You’re not just bolting on AI,” he says, and suddenly getting “20–30% lifts in productivity.”

The real mismatch is speed. “The problem is not that AI won’t deliver the gains in productivity—it will,” he says. “But it won’t deliver them at the speed with which the AI hype is predicting.”

This is also where energy and infrastructure creep back in. Many of the biggest bets in AI aren’t really software bets—they’re data-center, grid, and generation bets dressed up in tech language. “Does Sam Altman deserve a $7 trillion investment in AI infrastructure?” Turrin asks, contrasting that sum with what it could do for other global priorities. Even if AI is transformative, he suggests, the ROI timetable may not justify the current frenzy—especially when the bottleneck may be power delivery, not algorithmic novelty.

Talent and Adoption: The Hidden Flywheel Behind China’s Applied AI

For Turrin, the long game isn’t just chips or models—it’s people and adoption. He points to China’s growing advantage in technical education: “By 2022, China awarded more than 50,000 or 51,000 STEM doctorates—50% more than the 33,000 from the United States.” And at the broader level, he notes China produces “something like seven times the number” of STEM graduates.

Just as important, he argues the old U.S. advantage—attracting the world’s best talent—is weakening. “Those Chinese young people are not coming to the United States anymore,” he says, describing a shift toward staying home or choosing Europe.

Then there’s public sentiment. Turrin highlights survey work showing the U.S. as unusually anxious about AI, while countries such as India, China, and South Korea are more optimistic. In his view, enthusiasm matters because it accelerates testing and adoption. China isn’t only chasing AGI; it’s building an ecosystem where AI is applied—traffic systems, industrial robotics, “dark factories” without the need for lights—and people are willing to use it.

That willingness becomes a competitive advantage: faster deployment, more feedback, quicker iteration. And as AI spreads from the cloud into factories, logistics, and cities, adoption also means more devices, more compute at the edge, and more steady demand for electricity. The “people” flywheel ultimately feeds the “kilowatts” flywheel.

The Bigger Race Might Be Energy

Turrin’s final point is clear: AI may be the headline, but energy is the foundation. He points to breakthroughs like China’s “longest and hottest fusion energy burn ever” and its push into thorium reactors—“the world’s only production thorium reactor,” he says, with plans for more. In his view, abundant, clean energy could matter even more than who has the best chatbot.

Because when AI becomes embedded everywhere, the question is no longer “Can you build the model?” It becomes: Can you power the model, scale the compute, and keep the system running cheaply and reliably? That’s where kilowatts beat code.

He closes with a challenge to zero-sum thinking: “My hope is that your listeners see China as a competitor, not as the enemy.”

So here’s the question worth sitting with: if AI hype deflates in the West, will it slow progress—or will it force a pivot toward the applied, infrastructure-and-energy-driven approach that China is betting on?

More By This Author: