The Scarcity Cycle: Why the Real AI Trade is Moving Upstream

For about two years now, the logic for investing in AI infrastructure has been suspiciously straightforward: more GPUs mean you need more bandwidth, which forces more optical connections, which in turn spikes module revenue. The market didn't fight it. It priced in 400G, then 800G, and now it’s staring at 1.6T optics as a simple volume play—basically a headcount game where the winner is whoever can shove the most units out the door.

But that view is missing the real story.

If you look at the actual mechanics of the optical supply chain, there's a much bigger shift happening. AI networking is turning into a scarcity economy, not a volume one. In these kinds of markets, the winners aren’t usually the people at the tail end of the line just putting the final pieces together.

The Demand Signal is Backdrop, Not the Lead

The numbers are still impressive, sure. Shipments for 800G modules just about doubled this year, jumping from 20 million units to somewhere in the mid-40s. The next leap to 1.6T looks even steeper, with demand potentially hitting 75 million units. Even if you take those forecasts with a grain of salt, the trend is undeniable: there’s more optical hardware packed into every AI cluster than there used to be, and that shift toward 1.6T is very real.

This growth can easily outpace the total money being thrown at datacenters. Optics only accounts for about 4% of infrastructure spending right now. If that moves to 5% or 6%—which isn't a stretch—it creates a massive dollar opportunity without needing some impossible surge in overall budgets. People who doubt the optics forecasts because they’re tracking module demand one-for-one against total capex are fundamentally looking at the wrong thing.

But demand is just the setting.

The Real Bottlenecks Are Upstream

The thing holding back the next phase of AI optics isn't how much customers are willing to spend. It’s the supply of high-power lasers, modulated laser chips, and the indium phosphide substrates they’re built on. It’s the MOCVD tools and the manufacturing intuition you need to run those processes at scale.

This cycle feels different because these bottlenecks aren't neatly lined up. They’re hitting all at once. Usually, one part of a supply chain clears and the whole thing breathes. Here, the pressure is jammed across equipment, materials, and production yields. It makes "getting back to normal" a much slower, more painful process, which means the premium for actually being able to deliver stays high for a lot longer than people expect.

Optical chip yields are still hovering around 60% or 70% in some key areas. Just because a company announces a certain capacity doesn't mean they can actually ship it. Real volume depends on yield and fighting for equipment that is still incredibly hard to get. If you’re treating a company’s expansion plans as bankable revenue, you’re counting the wrong units.

The Nvidia Orbit

Nvidia is doing more than just generating traffic; they’re actively rebuilding the supply chain. They’re funding capacity for lasers and optical engines, and they clearly want strategically sensitive parts made in North America. A lot of this is political. Governments want domestic control over the guts of AI infrastructure, and Nvidia has the leverage to push production into those local hubs.

This doesn't mean we’re going to see a total return of module assembly to the US—the labor costs are still a nightmare to scale. But it does mean the most sensitive, high-value parts of the optical stack are being pulled into orbits that favor domestic suppliers. There’s a lasting valuation premium there that the market hasn't quite priced in yet.

Following the Profit

The investment takeaway comes right from that supply structure. The real profit is shifting to the people who control the bottlenecks—the laser makers, the chip designers, and the substrate producers. Module leaders only look good if they own their own components. The companies that are just assembling parts are going to feel their margins get squeezed, because in this world, the power belongs to the fulfillers, not just the volume shippers.

There’s also a second-order win for equipment suppliers in things like MOCVD and inspection. When yield matters as much as demand, the tools that actually make the yield possible gain an incredible amount of pricing power.

Then there’s the stuff like optical circuit switching and waveguide architectures. They might feel like science projects, but the big cloud providers are already treating them as strategic priorities. They aren’t a revenue story for next quarter, but they are the architecture the "smart money" is already funding. That’s how these shifts always start—quietly, and well before the consensus arrives.

The reality is blunt: demand makes people excited, but controlling the supply chain is what makes them money. In the AI optical market, the companies that own the scarce lasers, the substrates, and the relationships that make 1.6T actually shippable are the ones who will win this cycle. The ones just putting the parts together will be left scrap-hunting.

This is a controlled scarcity cycle. Invest like it.

not investment advice