U.S. optical earnings map: where the AI interconnect bottleneck sits

The demand question has mostly been answered

From first principles, AI optics is not a theme label. As AI clusters grow denser, the cost of moving data across GPUs, racks, data halls and data centers rises sharply. GPUs are the compute bottleneck; optical interconnect is the data-movement bottleneck.

The AAOI, COHR, LITE and AXTI earnings set says demand has entered the income statement, but not equally into profit. LITE has translated scarcity into high margins; COHR is scaling a broad photonics platform; AAOI is proving 800G volume but still has margin work; AXTI has upstream InP visibility, gated by export permits.

Where the real bottleneck sits

The hardest bottlenecks are slow to expand, difficult to replace, and valuable enough for customers to prepay or lock capacity. By that standard, LITE's lasers/OCS/CPO and AXTI's InP substrates look closer to hard constraints than generic module assembly.

NVIDIA's $2B investments in LITE and COHR are supply-chain options: capital in exchange for future access to advanced lasers, optical networking products and U.S.-based capacity. That validates optics as strategic AI infrastructure.

The earnings debates

LITE's debate is margin. The market already knows growth is strong; the important part is non-GAAP gross margin at 47.9%, operating margin at 32.2%, and a Q4 guide for 35%-36% non-GAAP operating margin. If EML, pump laser, OCS and CPO scarcity persists, margin revisions matter more than a small revenue beat.

COHR's debate is platform durability. Datacenter & Communications revenue reached $1.36B in FY26 Q3. It is not the highest beta, but NVIDIA commitments, 6-inch InP, transceivers, transport, OCS and CPO make it a broad photonics platform. The next test is whether Q4 gross margin can approach or exceed the 41% guide ceiling.

AAOI's debate is Q3 guidance. Q1 included the first volume 800G shipment to a large hyperscale customer, and Q2 revenue guidance of $180M-$198M is strong. But non-GAAP EPS guidance of -$0.03 to +$0.03 means revenue is running ahead of profit conversion. The thesis needs Q3 acceleration plus a margin inflection.

AXTI's debate is permit timing. InP backlog above $100M, roughly $34M of Q2 recognizable revenue, and Tongmei's $632.5M financing all point to strong demand. If permits clear, revenue can be nonlinear; if they lag, backlog cannot become earnings.

Failure conditions

The main risk is not immediate demand disappearance. It is supply expansion outrunning 1.6T, 3.2T, OCS and CPO adoption. LITE, COHR, AAOI and AXTI are all expanding capacity, while NVIDIA is also securing LITE, COHR and other strategic optical layers.

Customer lock-in is the second risk. Capital and purchase commitments can secure demand, but they can also secure pricing, priority access and delivery constraints for the buyer. Supplier quality must therefore be judged by revenue, gross margin and operating cash flow together.

Conclusion

My review order is: LITE margin and OCS/CPO supply gap; AAOI Q3 guidance and margin inflection; AXTI permits and InP backlog conversion; COHR D&C mix and the high end of Q4 gross margin. LITE is profit quality, AAOI is execution beta, AXTI is policy/material nonlinearity, and COHR is platform durability.

This is not a trading recommendation. It is a review framework: as long as scarcity keeps entering margins and cash flow, the optics chain remains investable as infrastructure. If only revenue grows while margin and cash flow lag, the upside narrative becomes valuation pressure.