Atoms, bits, and the trillion-dollar manufacturing race

She starts from history, on the principle that it rhymes. The Gilded Age is her first case because regulatory, technical and social change converged: Andrew Carnegie industrialized a steelmaking process he had taken from scientists rather than invented, turning pig iron into steel roughly ten times more cheaply, which let him lay rail track cheaply enough to become the figure we remember, while J.P. Morgan did something structurally similar in banking. Her second case is the 1940s heavy press, underappreciated beside the atomic bomb, which let the United States stamp out aircraft housings and bodies fast enough to matter, alongside the New Deal and the highway system. The pattern is that new tools applied to new kinds of projects produce order-of-magnitude gains in productivity, and that the sharpest version of the cycle comes from existential competition, which is to say war. Then the reversal. The host puts the end of the productivity run at around 2010; she pushes the date back, locating the turn in the 1970s and 1980s as the country moved from a builder economy to one innovating through services and financial structures, with NAFTA and the free trade agreements of the 1990s making it economical to manufacture elsewhere and beginning a multi-decade hollowing out.

Her account of China is not a story about copying. She has been going there for 23 years and describes three successive lenses: a socioeconomic one from not-for-profit work in the early 2000s, a political-economic one around the transition to the Xi era, and a techno-economic one once she was deep in the manufacturing sector. What compounds, on her account, is the ability to automate production lines with homegrown software, hardware design and tooling, which carries across from an iPhone to a car. The visible results are BYD and Xiaomi, which she insists are global leading innovators rather than me-too firms, and battery chemistry, where she credits China with bringing the chemistry she calls LFP into wide use and then moving to sodium ion, which she calls cheaper, safer and less dense but very good for uses like grid-scale storage. She does not spell out the first acronym or say what the density comparison is of. She also treats the arrangement as having been mutually beneficial: the average American, she says, has no idea how much value was created here on the backs of Chinese workers, and roughly fifteen years of broadly cheaper goods is a real thing lost from the telling, even though those dynamics certainly took a toll on the competitiveness of US manufacturing. The dependence also cuts against American leverage. Her reading of the critical minerals episode of the previous year is that the United States does not have the cards to push hard, because cutting itself off would devastate its own economy. In some ways, she says, globalization is the greatest deterrent to war, and she notes the tone toward China is more docile than the tone toward Europe.

The investment thesis follows from both halves. Most of global output is still in making, powering and moving things, the American economy is almost wholly propped up by services, and so the only route back is to leapfrog in how those things are done. The moat argument is the interesting move, because it converts the obvious objection into the asset. Physical businesses have to stand up plants, work out new ways to cut and create and scale real things, and absorb the long stretch of upfront loss that goes with it, all of which costs more money and time than shipping software and iterating. That cost is exactly why the resulting positions are defensible enough to run away with an industry, and why she names Tesla, SpaceX and NextEra, companies to some extent like Apple, and historically even Nvidia, as the shape of the answer. Her firm’s job, she says, is not doubles and triples but companies worth a hundred times more than at entry, and she adds a second-order claim: you want board members who will not lose their nerve at the first factory failure. Energy is where she thinks this runs into a wall. Today’s models are heavy-handed: a human brain runs on about twenty watts, and operation for operation what costs a person a few watts can cost an AI system kilowatts depending on the model and the silicon. She wants a better way to do AI and names a field she calls neuromorphic chips and neuromorphic models, different approaches to how AI is done, then immediately brackets it as a seven to ten year thesis nowhere close to daily life. So in the near term, demand for today’s AI outstrips US generation and transmission by orders of magnitude, the infrastructure is ossified, and transformers carry lead times she puts at a minimum of three, sometimes five, sometimes longer, without stating the unit. Her preferred answer is to co-locate large solar with batteries and smart controls next to data centers, and she adds gas turbines into the recipe mid-sentence, on the grounds that this gives a data center durable firm load without driving costs up and without the climate impacts of a massive series of gas turbines. She calls the national pullback from behind-the-meter development tragic, and notes a recent move by the federal power regulator, which she names only by its initials, that stops such projects being blocked as though they were utilities.

Asked where the next frontier is, she does not answer with a technology. She says the themes that matter behave like pendulums, that knowing where you sit on the curve is where the good bets are, and that her contrarian call is that deglobalization is near its inflection and a resurgence of the desire to collaborate is coming over the next few years. The mechanism she gestures at is trust infrastructure: if some entity could assure one country that another country’s technology was clean, auditable and free of spyware, the productive consequences would be large. Her worked example is that a Chinese autonomy company will almost certainly never operate on US streets in the current climate, and that this is a loss, because real competition would sharpen the American incumbent and benefit the consumer. She is explicit that she cannot show her work: the thousand data points behind the intuition, she says, are a black box.