Interconnect, not compute, is now the binding constraint on AI clusters. Copper, pluggable optics and CPO will coexist for years, so the strongest names span both paths or own the tightest bottleneck. Top five: Credo, the fastest-growing full stack; Applied Optoelectronics, the 800G/1.6T ramp; Marvell, broadest but richly priced; Lumentum, the laser chokepoint; Tower, the foundry bet. Coherent, Amphenol, Fabrinet, GlobalFoundries, TSMC and Broadcom also gain. Astera Labs is pricier and narrower. … Show more

Broadcom's diverse networking exposure makes it one of the best positioned in the theme. The market has overlooked it year to date. Optical DSPs, Tomahawk and Jericho underpin the view; Tomahawk 6/7 ramps in 2027, and Tomahawk Ultra and optical DSPs should accelerate growth. New XPU customers and extended hyperscaler deals add to earnings upside and an eventual re-rating. Risks remain, and weak semiconductor flows warrant caution about calling a bottom. x.com 

ASMPT's AMICRA bonders look set to fly off the shelves soon, while its thermo-compression bonding and advanced packaging business is growing well. x.com 

Coherent is the most potentially undervalued photonics play because the market has not caught up to what it has become. Its March 23 demonstration with Tower Semiconductor of 400 Gbps per lane on a silicon modulator, aimed at 3.2T transceivers, is not priced in. Revenue rose from $1.58B to $1.69B with $1.78B guided, and it was first to mass-produce 1.6T transceivers. Despite running from $61 to $336, it trades at a fraction of Arista's and Broadcom's valuations. x.com  x.com 

Amkor looks cheap at roughly 17x projected 2027 earnings of $2.83 a share, with its Arizona advanced-packaging capacity near TSM not expected to run near full utilization until 2028. The buildout has incurred substantial debt; the market may begin looking toward 2028 next year. The ramp remains prospective. x.com  x.com 

ASMPT is likely to become CXMT's alternative thermocompression bonding supplier after Korea's Hanmi suspended shipments to China. Its main TCB customer today is SK hynix, while Micron has excluded it on geopolitical grounds in favor of BESI, and reports say ASMPT has been in contact with Chinese HBM customers, most likely CXMT. x.com 

Lumentum and Coherent are investments driven by product iteration, while AXT is a bottleneck trade best treated as speculation. Shortage-driven price hikes on the same parts define a cyclical stock, prices quickly discount shortage expectations, and once profits truly climb it is time to exit. Bottlenecks also push customers to design around them and invite new entrants. Investable companies keep lifting margins through yield gains and new products, driving token deflation while growing revenue. … Show more

Credo is well placed if the US restricts Chinese optical transceivers from 3.2T, because it now has a full non-Chinese optical stack. Its $750M DustPhotonics acquisition brought in-house silicon photonics for 800G, 1.6T and 3.2T, including co-packaged optics, alongside its own DSPs and SerDes. If optics supply tightens, more short links stay on copper, which is Credo's AEC business. Management already expected optical lines to top $600M next year before this. x.com 

Himax's microlens monopoly is real but narrower than the market thinks. As the Hunterbrook and Citrini report claimed, and Tianfeng Securities confirms, it is the sole microlens supplier for TSMC's COUPE Gen 1-2. That is not a monopoly on Nvidia's co-packaged optics: Himax is not among Nvidia's 11 official CPO partners, TFC, Corning and Orbray build CPO-grade fiber array units without it, and 80% of revenue is still display drivers. That leaves it a small, low-conviction long. x.com  x.com 

Soitec now looks overextended after a rapid run, and a pullback is likely. The AI photonics bottlenecks behind the thesis are real, but it is hard to judge what premium each name deserves, which makes the stock hard to value at these levels. x.com