Coherent finds the AI bottleneck

Dear WIMM Supporter,

Today, I look at Coherent Corp - a company from my AI portfolio.

Onto the update:

TL;DR

Coherent is a photonics company. That sounds boring, which is usually where interesting things hide. It makes the optical components, lasers, transceivers, subsystems, materials, and manufacturing platforms that allow data to move faster, farther, and with less power. Don’t you think that these will be even more important in the years to come?

The company now reports through two main market segments. Datacenter & Communications is the engine. It represented 75% of Q3 revenue and is growing fast because AI datacenters need more bandwidth, lower latency, and better energy efficiency. Industrial is the legacy and diversification layer. It includes applications in manufacturing, semiconductor capital equipment, materials processing, sensing, and now increasingly thermal and power efficiency solutions for AI datacenters.

The moat starts with scale. Coherent is manufacturing at scale in a part of the supply chain where capacity is scarce. The most important phrase in the quarter was not “AI datacenter demand”. Everyone has that. The important phrase was “indium phosphide capacity.” If AI compute is the new oil, optical networking is the pipeline, and indium phosphide is one of the constrained materials inside the pipe.

The second moat lever is vertical integration. Coherent is positioned across materials, devices, components, modules, and systems. This matters because the AI datacenter is moving from buying isolated parts toward buying complete performance envelopes. Customers care about bandwidth, power efficiency, thermal management, reliability, and supply assurance. Coherent can serve more of that stack than a point-product supplier.

The third moat lever is customer embeddedness. Once a hyperscaler, networking company, or NVIDIA-like platform partner qualifies a supplier for critical optical infrastructure, the relationship becomes sticky. Qualification cycles are long. Failure costs are high. Supply commitments matter. The switching cost is not software lock-in. It is operational risk. Nobody wants to save pennies on optics and then discover that a billion-dollar AI cluster is bandwidth-starved.

The fourth moat lever is manufacturing know-how. Coherent’s 6-inch indium phosphide ramp matters because it improves both capacity and yield. Higher yield is not just an engineering achievement. It is margin expansion disguised as process improvement. When the same wafer platform produces EMLs, CW lasers, and photodiodes with better yield than older 3-inch lines, the company gets more units, better cost, and more strategic relevance at the same time.

The brand matters, but less like Apple and more like TSMC. The customer does not buy Coherent because it is emotionally attached to the logo. The customer buys Coherent because it needs the product to work, scale, ship, and survive the qualification process. In infrastructure, trust is brand.

The non-obvious point is that Coherent is not simply benefiting from AI. It is benefiting from the physical limits of AI. GPUs created the first bottleneck. Power, cooling, memory, and networking are creating the next bottlenecks. Coherent sits in several of those constraints.

Concluding, Coherent is moving from cyclical supplier to capacity gatekeeper. That is the difference between selling components and selling scarcity.

The 3rd quarter was clean. Revenue was $1.81 billion, up 7% sequentially and 21% year over year. On a pro forma basis, excluding divested businesses, revenue grew 27% year over year. Non-GAAP gross margin reached 39.6%, up 105 basis points year over year. Non-GAAP operating margin reached 20.3%, and non-GAAP EPS rose 55% year over year to $1.41.

The structural point is that profit grew faster than revenue. This is what investors should care about. Revenue growth can be cyclical. Margin expansion plus operating leverage means the company is gaining efficiency from scale, yield, pricing, and mix. In Q3, Coherent showed all four.

Management emphasized demand visibility. Orders now reach into calendar 2028, and customer LTAs extend to the end of the decade. That changes the character of the business. This is no longer a quarter-to-quarter optical cycle story. It is becoming a capacity allocation story. When customers commit years ahead, the supplier can invest with more confidence. That does not remove risk, but it improves the quality of the risk.

Datacenter & Communications is the center of gravity. Segment revenue increased more than 40% year over year and reached 75% of total company revenue. The data center business grew 13% sequentially and 37% year over year, while Communications grew 16% sequentially and 60% year over year. This is not one product line doing all the work. The growth is coming from transceivers, DCI, transport, OCS, and early CPO positioning.

The 6-inch indium phosphide ramp was the most important operational update. Coherent expects to double internal InP output by the end of the calendar year, one quarter earlier than planned, and more than double it again by the end of calendar 2027. The company also said its 6-inch yields exceed 3-inch yields across EMLs, CW lasers, and photodiodes. This is where the moat becomes visible in the P&L. Better yields mean more supply, lower cost, and higher gross margin.

OCS is now more important than the market probably appreciated. Coherent increased its OCS market opportunity view to more than $4 billion and said it resolved a production bottleneck. That matters because OCS moves the company up the value chain. It is one thing to sell optical components. It is better to sell systems that sit deeper inside AI networking architecture.

CPO is the longer-term prize. Management described co-packaged optics as a transformational growth opportunity with more than $15 billion of incremental addressable market. The NVIDIA partnership is the obvious validation point. The $2 billion equity investment gives Coherent capital, credibility, and demand visibility. The supply agreement runs through the end of the decade and includes multiple CPO-related products, including high-power CW lasers.

Industrial remains soft, but strategically useful. Revenue declined, reflecting broader industrial weakness, but management highlighted signs of improvement in semiconductor capital equipment. More interestingly, Industrial may become an AI-adjacent business through thermal materials, XPU cooling, and thermoelectric generation. That could turn a slower segment into a portfolio of energy efficiency options for large AI datacenters.

The vague part of the call was customer concentration and economics. We know the demand is strong. We know NVIDIA is important. We know hyperscalers are engaged. What we do not fully know is how much pricing power Coherent keeps when the largest customers also help fund the capacity.

What this call told us about the moat that the numbers alone do not: the moat is shifting from product breadth to supply-chain control. In AI infrastructure, scarce, qualified, high-yield capacity is a moat.

This 3rd quarter was more about proving that Coherent can convert AI optical demand into capacity, margin, and backlog.

The most important initiative is expanding internal indium phosphide capacity. The goal is to double output by the end of calendar 2026, now expected one quarter earlier than originally planned. This is the engine behind transceivers, CW lasers, and CPO products.

The second initiative is the continued ramp of 800G and 1.6T transceivers. 800G should continue growing in calendar 2026, while 1.6T ramps rapidly through the rest of the year and into 2027. This matters because each speed transition reshuffles supplier positions. The winners are the companies that can ship volume, not just demo technology.

The third initiative is OCS production. Coherent said it resolved a bottleneck and is ramping across two production facilities. If OCS grows quickly over the next few quarters, it will support the idea that Coherent is becoming a systems supplier, not just a component vendor.

The fourth initiative is initial CPO revenue in the second half of calendar 2026. This will likely start in scale-out architectures, with scale-up CPO expected to begin in the second half of calendar 2027. CPO is important because it addresses the limits of copper as AI networks demand more bandwidth and lower power.

The fifth initiative is internal simplification. The ERP consolidation and shared services work sound unglamorous, which is exactly why they matter. A company growing this quickly can destroy value through operational complexity. Better systems and lower SG&A intensity help convert growth into earnings.

Near-term catalysts:

Over the next 1–3 years, Coherent wants to become one of the default optical infrastructure suppliers for AI datacenters. The model is moving from broad photonics exposure to a more concentrated, higher-value, AI infrastructure platform.

The largest strategic bet is capacity. CapEx increased sharply in Q3 and should rise again in Q4. This is the correct move if demand visibility is real. It is also the main risk. Hardware companies usually get into trouble when they build capacity into a temporary demand spike. The difference here is that AI networking demand is being supported by backlog, LTAs, NVIDIA’s investment, and orders extending years forward.

The second major bet is CPO. If CPO becomes a core architecture for scale-out and scale-up AI networks, Coherent’s role expands. The company can supply lasers, external laser source modules, and other CPO-related solutions. In plain English, it moves closer to the heart of the AI cluster.

The third bet is systems. OCS, multi-rail, and transport solutions pull Coherent away from pure component economics and toward higher-value networking layers. This could improve margins and customer stickiness. It also increases execution difficulty.

The fourth bet is thermal and power efficiency. AI datacenters are increasingly constrained by electricity and heat. Coherent’s thermal materials and thermoelectric generation solutions are still early, with revenue expected to ramp in the second half of calendar 2027. But strategically, this is interesting. The company is looking beyond “move the data” toward “make the AI factory physically possible.”

Verdict: Buy.

“Where is my moat?” is designed for individual readers, though the occasional forward is absolutely fine. If you’d like to set up multiple subscriptions for your team with a group discount (minimum 5 seats), reach out to me directly.

Thanks for your support & have a wonderful day!

Thank you for being a supporter of WIMM!