To The Moon

If I had to choose one for the next 2–3 years, I would pick Marvell (MRVL).

The key distinction is that CPO is not simply an “optics boom”. It changes where the value accrues.

1. Marvell: best overall CPO exposure

Marvell is positioned across the interconnect stack rather than relying solely on optical modules. Its Celestial AI acquisition gives it Photonic Fabric for scale-up CPO, with management targeting a US$500m annualised run-rate by FY2028 Q4 and US$1bn by FY2029 Q4. 

That is potentially a much larger incremental opportunity than merely selling more transceivers.

2. AXT: my second choice, but potentially the biggest near-term torque

AXT is becoming a critical upstream bottleneck. Q2 InP revenue hit a record US$30.7m, versus US$13.6m in Q1, driven by AI optical demand. 

The problem is concentration and China export-permit risk. Its shares have already rerated dramatically.

3. AAOI: high upside, higher execution risk

AAOI has direct exposure to high-speed optical modules and is developing NPO/OBO and ELSFP products. Its roadmap makes it an interesting CPO beneficiary, but I would want to see sustained margin and volume improvement before making it my top pick.

4. HTGD: I would verify this ticker

If you mean SGX:HTGD, that is Trip.com, not an optics company. If you meant a different optical name, tell me the company name and I can rank it properly.

My ranking

MRVL > AXT > AAOI > HTGD

The interesting part is your observation about Coherent falling 7.99% despite beating and raising. That is exactly the warning sign I would focus on. The market is beginning to distinguish between optics demand and which optical architecture actually captures the economics.

Nvidia's Spectrum-X Photonics entering mass production is meaningful because CPO is moving beyond demonstrations into actual deployment. Nvidia says the architecture can deliver 5x lower network power and 10x better MTBF versus conventional approaches. 

**For me, Marvell is the cleaner long-term bet because it can benefit from the architectural transition rather than merely hoping CPO increases total optical-module volumes.**