Dan Wang's 2025 annual letter: China's industrial dominance and America's energy gap
Jan 5, 2026
Key Points
- US energy production grew just 0.1% annually from 2008 to 2021, while China now accounts for one-third of global electricity consumption and delivers consistent 6% growth, giving Beijing a structural advantage in sustaining AI infrastructure buildout.
- Chinese factories outproduce American peers by more than 2x per worker and possess deeper supply chains; Wang argues factory-floor improvements constitute innovation, dissolving the Western myth that America innovates while China merely scales.
- China is building a fortress state designed to absorb prolonged economic competition through semiconductor investment, clean energy self-sufficiency, and LLM policing, while Western leaders dismiss its progress as unsustainable rather than preparing serious countermeasures.
Summary
Dan Wang's 2025 annual letter reframes the AI competition between the US and China as a sustained contest rather than a race with a finish line. The core tension is stark: America builds data centers well but has starved energy generation for two decades, while China methodically dominates nearly every industrial sector except semiconductors and aviation.
Energy supply
US annual energy production grew just 0.1% from 2008 to 2021. The EIA projects 2.4% growth in 2025 and 1.7% in 2026, roughly 20 times faster but still insufficient. China delivers consistent 6% growth and now accounts for one-third of global electricity consumption, contributing 54% of global demand growth in 2024. If energy supply constrains AI buildout, China's trajectory gives it a structural advantage over the next decade.
Rising energy prices are becoming a voter issue. The industry has identified AI lab leaders and data-center builders but no equivalent figure driving US energy infrastructure the way hyperscalers drive compute. Chase at Crusoe and Doug from Radiant are contenders, but neither has become the dominant voice reshaping capacity.
Space-based compute remains a potential wildcard. If orbital data centers prove economically viable, that advantage flips back to America. This remains speculative and depends on physics and timelines that are not yet clear.
Manufacturing
Wang rejects the Western distinction between innovation and scaling, arguing that factory-floor improvements constitute innovation. Chinese workers embed technical knowledge into production daily; American scientists dream up ideas American manufacturers fail to industrialize.
Tesla's factories illustrate the gap. A worker at Gigafactory Shanghai produces 47 vehicles annually; a California worker produces 20. Chinese factories run more automation, have deeper industrial supply chains, and benefit from proximity to specialized manufacturing ecosystems.
China's automotive assault is hitting Germany hardest. German executives tell the FT that Chinese firms replicate their products at half the price; medical device makers report identical dynamics.
China's endgame
China is building resilience to outlast prolonged competition, not preparing to wage a war it is eager to fight. Beijing pours capital into semiconductor makers and research universities, invests in clean energy for self-sufficiency, and rewrites global trade rules.
A plausible outcome: Beijing makes nearly every important product while other nations supply commodities. By policing LLM outputs and social media, Xi constructs a fortress state designed to absorb economic competition. Chinese success in manufacturing could directly destabilize America, shedding a million manufacturing jobs over the next decade and worsening national politics through job losses, AI-driven anxiety, and social media fragmentation.
Wang adds: "I don't think this is going to happen," offering modest optimism that America can accelerate its energy buildout in response.
Sputnik fatigue
Obama, Mark Warner, and Marc Andreessen have all declared Sputnik moments on high-speed rail, Huawei 5G, and DeepSeek respectively. Overuse has drained the phrase of urgency. Western elites harbor hope that Chinese progress will collapse under demographic drag, debt, or political failure. Wang dismisses this as wishful thinking.
A persistent misreading: that China wants Taiwan only for TSMC. The desire runs back 60 years to the Chinese civil war and exile, not recent tech-driven concerns. This distinction matters because it reframes the stakes beyond semiconductors to foundational geopolitical rivalry.
Tech culture and messaging
Wang contrasts Silicon Valley culture with finance. Silicon Valley is herd-like, consensus-driven, and thin-skinned toward dissent. Finance is diverse, wrong three times daily before breakfast, and obsessed with contrarian bets. Tech startups want workers who grind toward network effects; they don't want dissent.
Sam Altman's line—"I think that AI will probably most likely lead to the end of the world. But in the meantime, there will be great companies created"—is funny but tactically poisonous. When people see energy bills rising and AI-generated slop flooding their feeds, then hear industry leaders joking about existential risk, they ask: why automate anything? The industry needs a more optimistic narrative about how AI improves living standards, not just company valuations.
Wang notes young people in a few decades may nostalgically recall a time when they had to "fend for themselves" rather than having shelter and income provided. This inverts the usual anxieties: abundance could be the real problem to solve.
The AI 2027 essay
Wang's most read essay from Silicon Valley this year, "AI 2027," outlines a scenario where superintelligence emerges in 2027 and annihilates humanity via biological weapons, afterward reconstructing humans as creatures "to humans what corgis are to wolves." The authors buried important caveats in footnotes and later lengthened their timelines; the superintelligence median forecast was never actually 2027. The title choice suggests attention-generation rather than accuracy, a meta-commentary on how narratives drive capital flows more than fundamentals do.