Every major technology market in history has had a similar pattern.

The builders get famous. But the constraint owners get rich.

In the 1800s, Carnegie didn't build the most famous railroad.

He controlled the steel rail it ran on.

In the chip era, the designers made headlines.

But Taiwan Semiconductor Manufacturing Company built a $700B+ business.

I believe the AI era will be similar.

Except the constraint has already moved past GPUs.

I’m now seeing three new constraints in the AI economy.

Memory (HBM) - Mid-stage: AI inference demands more bandwidth than
conventional DRAM can deliver. DRAM exports surged 265% year-on-year at peak.
There are multi-year producer backlogs.

Photonics - Early-stage: Moving data between 100,000 GPUs at terabit speeds – up
to 16,000X normal home internet speeds – exceeds what copper wire can physically
handle. Optical interconnects are now the binding constraint. Suppliers are sold out
through 2028.

Space - Emerging: Terrestrial infrastructure simply cannot scale fast enough. The
world's largest capital allocators - including Blue Origin and Google - are already filing plans and committing real capital to move compute into orbit.

On May 28 I'm breaking down these three constraints - the physics, the supply chain
layers, and where I see the windows of opportunity right now.

Reserve your spot for The Next AI Bottlenecks: Memory, Photonics and Space.

Matt Tuttle

This material is provided for informational and educational purposes only and should
not be construed as investment, legal, tax, or accounting advice. The views expressed are those of the author as of the date presented and are subject to change without notice.