Falling behind China on deep tech and chips

The contest between the United States and China is increasingly a contest over technology, and the U.S. cannot assume it will stay ahead. Three very different conversations treat technological leadership as a national security question. But the risk they describe is less that America has already lost and more that its lead rests on fragile foundations.

Semiconductor analyst Dylan Patel and AI researcher Nathan Lambert give the most detail. America's AI lead depends on Nvidia's designs and on TSMC's factories in Taiwan. The Chinese lab DeepSeek made real engineering breakthroughs on weaker, export-approved chips, and as Patel puts it, "necessity is the mother of innovation". But its famous low cost figure covered only a final training run, not research. Lambert reframes what export controls do. China will always be able to train models, but controls limit how much AI it can run. Patel calls the controls a bet on timing. If AI is not transformative within five to ten years, they help China win in the long run, because China can outbuild the U.S. and will subsidize its own chips. He estimates bringing chipmaking fully home would take a decade and a trillion dollars, against a CHIPS Act of about 50 billion dollars and roughly 200 billion a year in Chinese subsidies. His biggest fear is that cutting China off pushes it toward a blockade or takeover of Taiwan.

Elon Musk puts the rivalry in history, saying "the foundation of war is economics". He thinks China's economy may grow to two or three times America's, and he calls the incorporation of Taiwan certain from China's standpoint. He also predicts physical bottlenecks: chips now, then voltage transformers, then electricity. Yet he offers reassurance too, noting China has historically been inward-focused.

Former Navy pilot Ryan Graves argues that China's state money can fund long-shot science that American venture capital avoids. He says secrecy made sense when the U.S. had no peer, and that the answer now is to activate America's scientists in public.

They agree China is a peer competitor, that state money is competing with market money, that Taiwan is the flashpoint and that physical capacity matters. They differ on the response. Patel backs controls only if AI changes society soon. Graves wants openness. Musk stresses building capacity and some oversight of AI. Lex Fridman adds that importing the world's best talent should not be controversial. They also differ on urgency, with Lambert expecting transformative AI after 2030 and Patel around 2027 or 2028.

The article ranks the evidence. The chip details are strongest. Graves's examples, such as Chinese interest in gravity manipulation, come from conversations and are the most speculative. Patel and Lambert say U.S. firms still lead in total compute, and neither thinks the Chinese government subsidized DeepSeek.

What to do with it: when a headline claims a cheap AI breakthrough, ask what the cost figure includes. Follow export controls and their loopholes. Treat Taiwan and electricity as technology issues. Compare investment, not announcements. Keep both risks in view: falling behind, and provoking a crisis.