ilster

🚨 Where Is The Photonics Trade Headed Next? $AAOI $POET $GLW $MRVL $CIEN $LITE & $COHR earnings came out last week. So let’s talk about what they actually told us and how that signal flows through the rest of the ecosystem. When companies at this layer post strong quarters at the same time, guide higher, and openly talk about being sold out, it means real infrastructure is being built in a very significant way. Hardware is being ordered, systems are being installed, and capacity is being scaled. That’s what makes their results so useful as signals. What Lumentum Holdings and Coherent Corp just showed is that photonics is moving beyond a simple upgrade cycle and into something broader. This is starting to look like the early stages of an infrastructure build, where light becomes core plumbing for how AI systems operate. Both calls reinforced four things: • Demand visibility is moving outward (multi-year commitments, customers ordering further ahead) • Supply is still the bottleneck (InP fully allocated, demand exceeding capacity by ~25–30%) • The architecture shift is very real (800G now, 1.6T coming, plus OCS and CPO ramps) • Manufacturing execution is becoming a moat (yield, throughput, six-inch InP, assembly scale) These are the keys. Now the key question becomes: how does that signal flow through the rest of the stack? Let’s comb through it. $AXTI Before we even get to transceivers and optical engines, the photonics stack starts upstream with raw materials. A huge portion of high-speed datacenter optics relies on compound semiconductors, especially Indium Phosphide. This is where AXT fits. AXT supplies InP substrates that sit at the very beginning of the chain. Think of it as feedstock that downstream players turn into lasers, detectors, and optical components. This matters because Lumentum and Coherent both confirmed they are still supply constrained at the laser level. That constraint starts upstream. So AXTI becomes an early tell. For AXT I’m watching: • InP availability and lead times • Pricing dynamics • Any commentary around export permits or shipment friction • Mix shift toward InP vs other substrates What we know is that when substrates get tight, everyone downstream feels it. $AAOI Once light is generated, it has to be turned into usable data signals. That happens inside transceivers, sitting at the edge of every fiber connection. That’s where Applied Optoelectronics operates. As networks move from 400G to 800G and toward 1.6T, transceivers become harder to manufacture. Execution starts to matter as much as design. Yield, reliability, and production scale become everything. Lumentum just told us their EML lasers are sold out for six quarters and demand keeps widening. That typically means transceiver makers are seeing real orders forming behind the scenes. For AAOI, the questions are simple: Can they qualify fast enough? Can they manufacture consistently at higher speeds? On AAOI’s call I’m watching: • 800G shipment scale and 1.6T timelines • Customer concentration vs diversification • Yield and manufacturing commentary • Gross margin behavior as volume ramps That determines whether demand turns into durable revenue. $POET $MRVL As systems get denser, customers increasingly want integrated solutions instead of assembling everything piece by piece. That shift pulls architecture players into focus. Marvell is building toward owning more of the data-movement stack, spanning switching silicon, optical DSP, silicon photonics adjacency, and now scale-up fabrics. They’re trying to become connective tissue for massive AI clusters. Alongside that sits POET Tech POET lives in the optical engine layer, integrating lasers, detectors, and electronics into compact photonic assemblies. As optics moves closer to compute, this layer becomes more valuable because it reduces system complexity and can lower manufacturing cost. For POET, milestones are very sequential: Design wins → qualifications → repeat orders → volume shipments. On POET’s call, I’m watching: • Language shifting from development units to production • Qualification progress • Evidence of automated test and packaging scale • Revenue mix moving toward repeat customers That progression is what turns promising tech into real business. $TSEM Design alone doesn’t build data centers. Everything still has to be fabricated, packaged, and tested. That’s where Tower Semiconductor enters. Tower provides specialty manufacturing, including silicon photonics processes. As optical integration rises, foundries become strategic assets. On TSEM’s call, I’m watching: • Silicon photonics utilization • New program ramps • Packaging/test investments • Specialty process demand tied to optical I/O Capacity and consistency directly determine how fast photonics can scale. $CIEN AI infrastructure doesn’t stop inside one building. Data has to move between campuses and regions. That’s handled by optical transport, which is the domain of Ciena. When cloud providers spend on transport, it confirms clusters are being connected at scale. On CIEN’s call, I’m watching: • DCI backlog and order growth • ZR/ZR+ coherent optics mix • Lead times • Explicit AI-driven demand commentary Strong transport demand validates a broad network build. $GLW Every optical link ultimately relies on fiber. That’s where Corning fits. No fiber means no optics. As racks get denser and architectures get more leaf-spine-heavy, you add higher fiber counts, tighter bend tolerances, more pre-terminated assemblies, and you try to install it faster with less labor. On GLW’s call, I’m watching: • Datacenter fiber volumes • High-density cable mix • Capacity expansion • Orders tied to new builds vs upgrades $FN One of the clearest real-world indicators of activity is Fabrinet. Fabrinet assembles and tests optical hardware for Lumentum and Coherent. When FN ramps, products are already moving through factories. On FN’s call, I’m watching: • Backlog growth • Optical line utilization • Hiring and capex • Margin leverage vs overtime pressure $SMTC $MTSI Behind every optical lane sit analog components that keep signals clean at extreme speeds. That’s Semtech and MACOM Technology Solutions. As speeds rise, their content per system increases. On their calls, I’m watching: • 800G → 1.6T attach rates • Design wins converting to production • Margin expansion from higher-performance parts $AVGO $ANET $NVDA Finally, there are the big dogs. Broadcom defines networking sockets. Arista Networks builds the Ethernet fabric. NVIDIA sets cluster architecture and deployment scale. When these accelerate, photonics follows. On their calls, I’m watching: • AI capex cadence • Cluster architecture changes • Optical I/O, CPO, and scale-up fabric language They determine the future and the pace of the entire build. ... Lumentum and Coherent just confirmed we’re entering a broader photonics build phase. Photonics is becoming core infrastructure for AI. No doubt about it. Now it’s on us to find the players that capture the CAPEX over the next few years. If you found this insightful, please do save and repost.