Why the Skeptics Are Right — and Why I'm Investing Anyway

1Flourish has two new investments in advanced manufacturing companies building robotics systems. We made those bets because we believe AI has already changed the economics of precision manufacturing in America — not in 10 years, in the 0-to-5-year window we're in right now.

It's a narrow thesis, not a triumphalist one. It doesn't involve reshoring t-shirts or washing machines. It's specifically about quality-critical, IP-sensitive sectors — aerospace components, defense hardware, specialized industrial equipment — where the decisive variable was never labor cost. It was reliability. And AI is transforming reliability faster than the skeptics expected.

But before I argue it, I want to make the strongest case against it. Because the naysayers are serious people, and some of them are right — just not about the part I'm betting on.

The Bear Case Is Real

Three criticisms deserve real engagement.

The productivity math is contested, and the wage gap is real. Daron Acemoglu — who won the Nobel Prize in Economics in 2024 — has modeled AI's economic impact at roughly 1.1 to 1.6 percent of GDP over a full decade. Goldman Sachs puts it at 7%. McKinsey says 0.5 to 3.4 percent annually. The variance is telling: nobody actually knows. Meanwhile, U.S. manufacturing labor averages $25 to $30 per hour against $6 to $7 in China. Even with 15 to 30% Industry 4.0 productivity gains, that gap doesn't close for most product categories. Boston Consulting Group estimates reshoring still adds 10 to 30% to production costs even with modern automation.

Then there's the workforce paradox. The Reshoring Initiative's 2024 annual report records 244,000 new U.S. manufacturing jobs announced — a remarkable number. But right now 409,000 manufacturing positions are unfilled. Deloitte projects that by 2033, the industry will need 3.8 million new workers, with 1.9 million at risk of going unfilled. You can't reshore what you can't staff. And modern factories need a different worker: someone who can maintain a robot arm, read a sensor dashboard, and debug a CNC program. That person barely exists at scale yet.

This is a serious bear case. Anyone who dismisses it is selling something. I take it seriously — and I'm still investing. Here's the distinction that matters.

The Bear Case Hits the Wrong Target

All three objections are correct about commodity manufacturing. T-shirts, furniture, consumer electronics at scale — probably not coming back to U.S. factories in any meaningful volume. That's not the bet I'm making, and conflating it with my bet is the critical error in most skeptical takes.

The category is precision, quality-critical, IP-sensitive manufacturing — aerospace components, defense hardware, advanced semiconductors, medical devices, and the specialized tooling that produces all of them. These sectors never fully left, because a machined titanium aircraft component where a 0.001-inch error grounds a fleet isn't price-sensitive — it's reliability-sensitive. The calculus was never about labor cost. And AI is transforming reliability.

Acemoglu is modeling economy-wide productivity. He's probably right in aggregate. But investment returns don't need economy-wide transformation — they need transformation in the right sector. In precision manufacturing, the decisive variable was never labor cost. It was the ability to design, simulate, fabricate, and inspect to extreme tolerances, reliably, at scale. That's exactly what AI is now improving — faster than the skeptics expected three years ago.

Three Possible Futures

I see three distinct visions for how advanced manufacturing in America unfolds — each internally coherent, each with strong arguments for and against it.

Vision 1: Full-Stack AI-Native Manufacturing (7–12 year scaling horizon). Software redefines the entire industrial stack, from AI-designed parts to autonomous factories. The case for it: convergence of generative design, generalist robotics, and materials AI is real and accelerating. The case against: each component matures on its own timeline — simultaneous convergence before 2030 is unlikely.

Vision 2: Precision Verticals Win First (0–5 year scaling horizon). AI tips the economics in high-value, quality-gated sectors. Already happening: 88% of reshored jobs in 2024 were in high or medium-high tech. The economics work without AI solving everything simultaneously. The ceiling risk: defense and aerospace are roughly $900B, not $2.3T — a strong investment story, not yet a macro reshoring story.

Vision 3: New Industrial Geography (1–7 year scaling horizon). The US designs and integrates; allied nations manufacture. Already the current trajectory — tariffs are accelerating bifurcation from China, not full reshoring. The long-term risk: tacit knowledge erodes when engineers don't make things.

The honest answer is that all three are partially true, playing out in different sectors on different timelines. Vision 3 describes where most of manufacturing goes in the near term. Vision 2 is where the investment returns are. Vision 1 is where this ends up if the technology matures as proponents believe.

My thesis is that Vision 2 is the beachhead for Vision 1. Precision verticals aren't a ceiling — they're a training ground. Every aerospace component an AI system learns to machine builds generalizable knowledge. Every validated material shortens the path for the next one. American technological leadership has always followed this pattern: win where quality beats cost first, then expand. Semiconductors did it. Defense electronics did it. Advanced manufacturing is doing it now.

What I'm Seeing From the Inside

Part of what gives me confidence is proximity to the underlying research. Through my work with Stanford's Intelligent Systems Laboratory, I see how decision-making algorithms originally built for autonomous aircraft — systems that must act in complex, dynamic environments where errors have physical consequences — are being adapted for exactly the kind of coordination challenges a modern AI-enabled factory faces. A fleet of collaborative robots on a production floor is, at the algorithmic level, a close cousin to air traffic management. That research is maturing. a16z's Erin Price-Wright put it well: AI systems are "purpose-built to deal with complexity that's extremely difficult to program deterministically." That's not automation replacing judgment. It's augmentation — applied to atoms instead of bits.

The optimists aren't wrong about the destination. They're wrong about the speed — and about which sectors get there first.

Where This Series Goes

The next piece looks at what's actually reshoring right now — and why 88% of it is already in high-tech sectors. The precision verticals thesis isn't a prediction; it's a description of something happening today. The third piece gets specific on where I see the investable opportunities in the 2025–2030 window, and why the most important companies may not be manufacturers at all.

Acemoglu is right that the triumphalist version of this story is probably wrong. But the modest, specific, beachhead version — the one grounded in sectors where quality already beats cost, where national security provides tailwinds, and where AI is improving capabilities faster than skeptics expected — that version I find compelling enough to invest in. Twice, so far.