The AI Productivity Story Is Starting to Sound Familiar

Concentration Risk Is the Real Story

Bloomberg's other segment this week touched on something that concerns me more than the hype, actually, let me be precise, worries me considerably more. AI concentration risk.

I called my old colleague Heinrich at Siemens last month, we were talking about supply chain dependencies in industrial automation. He made a point that stuck with me: in the 90s and 2000s, if one PLC vendor had problems, you could switch. It was painful, sure, but possible. The ecosystem had enough diversity that single points of failure were manageable.

The AI infrastructure doesn't look like that. A handful of companies control the foundation models. A handful of chip manufacturers supply the hardware. The cloud providers running inference at scale? You can count them on one hand. This is the kind of concentration that makes supply chain managers lose sleep.

And it's not just theoretical. We've seen what happens when critical infrastructure gets concentrated. I remember when a single Fanuc controller shortage in 2011 basically paralysed half the automotive industry for months. Now imagine that kind of dependency, but for the AI systems that are supposedly going to run everything.

What I'm Actually Seeing

I'll be honest, I don't have comprehensive data on this. Nobody does, which is sort of the point. The productivity claims are based on surveys and self-reported improvements. The hard numbers, the kind you'd need to justify the market valuations we're seeing, those remain unclear.

What I am seeing, from talking to people still in the industry, is a lot of pilot programs. A lot of experimentation. Some genuine wins in specific applications. But the broad-based productivity revolution? It's too early to say whether that's coming or whether we're in another hype cycle that'll take years to mature into something real.

The cobots eventually did transform certain applications. Palletising, machine tending, some assembly tasks. It just took a decade longer than the 2015 projections suggested, and the scope was narrower than anyone predicted. I suspect AI will follow a similar trajectory.

The Investment Question

I'm not a financial analyst. I fix robots, or I used to. But even I can see that the money flowing into AI infrastructure assumes productivity gains that haven't materialised yet. Maybe they will. Maybe the lag is just longer than expected.

Or maybe we're watching another technology that's genuinely useful get inflated into something it can't actually be. I've seen that movie before. The ending is usually a bunch of useful tools that get adopted slowly, a lot of writedowns, and a decade of people saying "remember when everyone thought AI would...?"

I hope I'm wrong. Some of this technology is legitimately impressive, and I'd like to see it actually help people work better rather than just generate hype. But the pattern recognition part of my brain, the part trained by watching industrial automation hype cycles for 30 years, is sending up warning signals.

We'll see. That's basically all anyone can say right now. We'll see.

Aisha Patel · 25 Jun · 8 min