The Second Half After AI Computing Power: Why 'Optical Interconnect' is the Core Battlefield of the Future?

Author: qinbafrank

Let’s discuss the logic behind the optimism for optical interconnects. In mid and late July, I talked about the sectors I’m optimistic about in the U.S. stock market, which have already exploded (CSP, software, optical interconnect). The rationale for CSP/software was discussed thoroughly in a long article in mid-July. Today, let’s talk about why I’m optimistic about optical interconnects. What are the catalysts?

Looking back, the optical interconnect sector was the first to adjust in this wave of semiconductor corrections and has been adjusting for the longest time. From a personal perspective, a report from Bernstein in mid-May stated that the CPO realization wouldn’t come so quickly, and this year still focuses on optical modules and connections. After that, the optical sector began to adjust, while the U.S. stock market only saw decent performance in connections.

Recently, I see two catalysts:

1. "Nvidia's Senior Vice President Gilad Shainer announced at a recent technology forum that CPO has begun mass production. The switches developed in collaboration with its supply chain have been delivered to close customers and are being deployed in Nvidia's own facilities. A large number of switches adopting CPO technology are expected to be introduced to AI factories globally this year;

2. Lumentum CEO Michael Hurlston stated that the shortage of AI indium phosphide lasers will be more severe than the memory crisis.

How to view this logic:

1. The delivery of CPO means that the market should start to expect that with mass production and delivery, CPO should gradually ramp up, and business realization will begin to materialize;

2. The shortage of indium phosphide, and the market still has muscle memory regarding shortages;

3. Optical adjustments have been early and have lasted long enough, while semiconductors have concentrated on storage. Recently, there has been little discussion about optical interconnects, which means there is a difference in expectations;

4. The preliminary financial reports of super-large CSPs have proven the capital expenditure ROIC, and the construction of data centers will accelerate. The demand for large-scale cluster interconnects and data center interconnects is actually further exploding.

In summary:

The demand for optical interconnects in AI data centers is real and accelerating, and the entire industry chain has moved from "demand validation" to the "supply bottleneck" stage; NVIDIA's CPO officially entering mass production is a historic signal, coexisting in the short term (at least until 2027) with pluggable optical modules, with the ultimate bottleneck locked in the production capacity of indium phosphide (InP) laser chips.

Technical Roadmap Timeline:

1. Scale-up (within racks): Mainly copper cables/pluggable.

2. Scale-out (cross-rack/cross-data center): CPO will penetrate first (power consumption from 30W to 9W/port, significantly increasing bandwidth).

3. Coexistence window: Pluggable will dominate for the next 18-24 months; CPO and pluggable are non-zero-sum, at least until 2027.

"Demand has been validated across the entire chain, and the bottleneck is in InP lasers; whoever can self-produce/expand production the fastest and has the safest supply chain will gain the greatest flexibility."

The next two weeks will see a concentration of earnings reports from companies in the optical interconnect field. If the earnings reports are further validated, it could be a catalyst window.