The Next AI Bottleneck Theme

This Fourth AI bottleneck could create the next generation of winners

Every AI cycle has a bottleneck.

The companies sitting inside that bottleneck capture the value of the cycle.

It has happened three times already.

Each time, the bottleneck was hidden in plain sight until the move was almost over.

The fourth is forming right now.

---

Bottleneck #1: Chips

From 2020 through 2022, compute was the constraint.

GPU supply was rationed. Hyperscaler allocations were negotiated quarterly. Inference workloads ran on the largest pool of chips any operator could secure.

One company sat at the center of that bottleneck.

Nvidia went from a $300 billion market cap at the end of 2020 to over $4 trillion by mid-2025.

Most retail investors held $NVDA through part of that move. Few held the full move.

---

Bottleneck #2: Memory

By 2023, GPU supply eased enough that the constraint moved one layer deeper.

The chips needed High Bandwidth Memory to function.

HBM3 became the new bottleneck. SK Hynix dominated. Micron caught up. Samsung stumbled on qualification.

Memory stocks that traded sideways for years suddenly doubled.

Different bottleneck. Same pattern. Same winners-take-most dynamic.

---

Bottleneck #3: Photonics

By 2024, memory caught up.

The chips had memory. The memory had chips. Neither could talk to each other fast enough.

Optical interconnects became the new bottleneck. $COHR. $LITE. $AAOI.

The photonics super cycle began. The names re-rated 3 to 5 times.

Subscribers who positioned early through my Photonics Watchlist caught a real piece of it.

---

Bottleneck #4?

The first bottleneck created Nvidia ($NVDA).

The second rewarded Micron ($MU) and sandisk ($SNDK).

The third rewarded Lumentum ($LITE), Coherent ($COHR), and Applied Optoelectronics ($AAOI).

The fourth could create an entirely new group of winners.