Earlier this month, Lumentum reported $665.5 million in quarterly revenue — a company record — with 65% year-over-year growth. Non-GAAP operating margins expanded by over 1,700 basis points. The stock has since surged past $700, crossing a $48 billion market cap — overtaking Coherent for the first time. One day after Lumentum’s report, Coherent posted $1.69 billion in revenue, beat estimates on both lines, and the stock dropped as much as 20% intraday before partially recovering.
Both companies make lasers and optical components for AI datacenters. Both beat expectations. When I mapped the optical interconnect technology stack in December, I placed Lumentum and Coherent together in Tier 1 — “pure-play optical” with the highest leverage to the bandwidth buildout. The market has since made it clear that not all Tier 1 positions are created equal — in an AI transceiver market that Jefferies projects will reach nearly $50 billion by 2028.
Then this week delivered two more data points. On Monday, Keysight Technologies reported record results — $1.6 billion in revenue, 30% order growth, and a stock move of +23% — with management calling out 800G/1.6T optical testing as a primary growth driver. And last night, Applied Optoelectronics dropped an earnings call that adds a new wrinkle to the laser supply chain story. More on both below.
Coherent is the larger company by revenue but now trails Lumentum in market cap. The reason is visible in one number: Lumentum is growing nearly 4x faster and expanding operating margins at 12x the rate — 1,730 basis points year-over-year versus Coherent’s 140. That’s not a valuation gap driven by sentiment. It’s the market pricing in the quality of the growth.
The Laser Physics That Wall Street Misses
To understand why this divergence is structural and not cyclical, you need to understand what’s happening at the component level inside AI datacenter networks.
Every 1.6T optical transceiver needs a laser source. I’ve written before about how the entire AI networking stack begins with indium phosphide laser chips at Layer 1. As the industry transitions from 100G per lane to 200G per lane, the physics requirements change dramatically. This isn’t like going from a 4-cylinder to a 6-cylinder engine. It’s more like switching from propeller aircraft to jet propulsion.
At 200G per lane, Electro-absorption Modulated Lasers (EMLs) become the required technology. VCSELs — the cheaper, easier-to-manufacture vertical cavity lasers that dominated at 100G — face fundamental reliability problems at these speeds. Failure rates scale dramatically with bandwidth — some reliability analyses suggest degradation follows approximately the fourth power of bandwidth. That’s not a manufacturing challenge you can optimize away. It’s physics.
When I was working in Professor Fainman’s ultrafast nanoscale optics lab at UC San Diego, one of the things you learn quickly is that semiconductor laser performance at high power and high speed is governed by material science that took decades to develop. Narrow linewidth — the spectral purity of the light source — gets exponentially harder to maintain as you push power higher and temperatures up. Lumentum’s published specifications show <500 kHz linewidth at +24 dBm (~250mW) at 50°C for O-band applications. To my knowledge, no competitor has publicly matched these specifications, credit to Irrational Analysis for pointing this out to me.
Based on industry pricing data, Lumentum’s 200G EMLs command an estimated ~2x the ASP of their 100G counterparts — confirmed by Jefferies’ optical model, which notes “200G laser pricing nearly 2x 100G at only 15% increase in COGS.” That cost-to-price spread is the margin story in a single data point. Despite representing only about 5% of unit volume, we estimate 200G EMLs are already contributing approximately 10% of datacenter laser revenue. As 1.6T deployments ramp through 2026 and 2027, this mix shift becomes a powerful margin driver at a time when their San Jose high-power laser fab is sold out.
The lane transition also explains the divergence at the company level. Lumentum had limited share in VCSELs, so the InP business is nearly entirely new revenue. Coherent, by contrast, is partly cannibalizing their own VCSEL business as the transition unfolds.
The 6-Inch Indium Phosphide Question
Coherent has staked their competitive positioning on scaling to 6-inch indium phosphide wafers — up from the industry standard 3-inch. The pitch: more than four times as many chips per wafer at less than half the cost, similar to how the silicon industry moved from 200mm to 300mm wafers decades ago.
The program is real and progressing. On their Q2 earnings call, CEO Jim Anderson reported that wafer starts on 6-inch lines have reached 80% of target capacity — ahead of schedule — with production ramping in parallel at Sherman, Texas, and Järfälla, Sweden. He noted that 6-inch yields continue to exceed those of their mature 3-inch lines, which is a genuine technical accomplishment.
But here’s the contradiction that matters for investors: despite this progress, Coherent is simultaneously increasing their purchases of EML laser chips from Lumentum.
If your 6-inch wafers are yielding better than your 3-inch wafers, why would you simultaneously be buying more of the same components from your primary competitor?
Anderson provided the answer: the typical lead time from wafer start to transceiver shipment is approximately six months. Six-inch production began in the September 2025 quarter, meaning the cost benefits are only now beginning to flow through. By year-end, Coherent expects roughly half of internal InP capacity to run on 6-inch wafers.
So the 6-inch program is real — but the market is pricing in the narrative before the economics arrive. The cost advantages haven’t yet materialized in the margin profile (non-GAAP gross margin improved just 77 basis points year-over-year). From my experience scaling a hardware business, when a company’s public claims about manufacturing capability don’t yet match their purchasing behavior, you watch the purchasing behavior. If the 6-inch program delivers as promised, EML purchases from Lumentum should plateau or decline by mid-2027. That’s the metric to track.
Optical Circuit Switches: Transparency vs. Ambiguity
Optical Circuit Switches (OCS) represent one of the most important architectural shifts in AI datacenters. Instead of converting photons to electrons and back at every network switch — wasting power and adding latency — OCS routes optical signals directly. The OCS market forecast has been revised upward from approximately $500 million to nearly $1.5 billion.
Lumentum disclosed a backlog exceeding $400 million, with shipments to three customers and the majority shipping in H2 calendar 2026. They cleared their earlier target of a $10 million quarter three months ahead of schedule. CEO Michael Hurlston was specific: demand is “broad-based,” pricing is improving, and they received a multi-hundred-million-dollar CPO purchase order deliverable in H1 calendar 2027.
Coherent cited “10 customer engagements” in OCS — but disclosed no backlog figures, no revenue timelines, and no dollar commitments. They did disclose a large CPO purchase order from a market-leading AI datacenter customer manufactured on their 6-inch InP line, with initial revenue expected later this year — a meaningful data point. But when an analyst directly asked CEO Jim Anderson to quantify the OCS backlog, he declined.
The technology matters here too. Lumentum’s OCS uses MEMS — tiny physical mirrors that redirect light beams with minimal signal loss. Coherent’s approach relies on liquid crystal technology, which has inherently worse optical insertion loss. In a datacenter running thousands of these switches, that difference compounds at every hop
The Margin Story Is the Real Story
Revenue growth gets headlines, but margin expansion reveals whether a company is capturing value or just processing volume.
Lumentum’s non-GAAP gross margins expanded to 42.5%, up over 1,000 basis points year-over-year. Operating margins hit 25.2%, up from 7.9% a year ago. They’re raising prices on incremental capacity through long-term agreements and under-shipping customer demand by approximately 30%. Every incremental unit gets absorbed at favorable pricing. That’s the kind of leverage you see when a company is selling into supply-constrained demand with differentiated products.
Coherent’s non-GAAP gross margin of 39.0% and operating margin of 19.9% are respectable. Their long-term gross margin target of over 42% suggests management sees a path to Lumentum-level profitability as 6-inch economics scale. But today, one company has pricing power rooted in physics. The other is investing heavily in capacity that hasn’t yet translated to margin differentiation.
Jefferies’ optical model projects Lumentum’s operating margin climbing to 33.2% by September 2026 and 36.4% by mid-2027. Their CY27 revenue estimate of $5.6 billion sits 46% above Street consensus. Coherent’s projected margin trajectory is shallower — reaching 23.7% over the same timeframe — and their CY27 revenue estimate of $9.8 billion sits 13% above consensus.
The Picks-and-Shovels Confirmation: Keysight
If you want to know whether an infrastructure buildout is real, watch the companies selling the test equipment. On Monday, Keysight Technologies reported record Q1 FY2026 results — $1.6 billion in revenue (+23% YoY), $1.645 billion in orders (+30%), and non-GAAP EPS of $2.17 that beat consensus by nearly 9%. The stock surged 23% to an all-time high.
The detail that matters for this thesis: wireline revenue surpassed wireless for the first time in the company’s history. This is the segment that sells Digital Communication Analyzers and Lightwave Component Analyzers to optical transceiver and module suppliers — the instruments that validate 800G modules before they ship and qualify emerging 1.6T designs for standards compliance. Wireline orders have now grown for nine consecutive quarters. CEO Satish Dhanasekaran identified four AI datacenter demand drivers on the call: the shift to higher-speed Ethernet networking (800G/1.6T, with 3.2T R&D already underway), the increasing importance of optical interconnects including CPO and OCS architectures, and the need for system-level AI workload emulation. AI-related orders grew “significantly above” the company’s 30% average, and Keysight has doubled the number of customers driving that demand.
SVP of Global Sales Steve Yoon put it plainly: “In my 36 years with the company, this is one of the strongest funnels I’ve ever seen.” Management raised full-year guidance to revenue and earnings growth “just above 20%” — well above the company’s long-term target of 5–7%.
This is the demand signal you can’t fake. Test equipment orders are a leading indicator — they show up before the production ramp, not after. When Keysight’s customers are buying 1.6T-capable analyzers at record rates, it tells you that Lumentum’s sold-out laser capacity and Coherent’s 6-inch wafer investments aren’t ahead of demand. They’re behind it.
The Speculative Angle: AAOI and the Laser Shortage Thesis
Last night, Applied Optoelectronics reported Q4 2025 earnings — record quarterly revenue of $134.3 million, capping their strongest year at $455.7 million. The stock jumped roughly 19% after hours. But the number that caught my attention wasn’t the revenue beat. It was management’s claim that they expect approximately $378 million in monthly transceiver revenue by mid-2027 — implying an annualized run rate north of $4.5 billion.
That’s a 10x revenue increase in two years. The execution risks are severe. But the reason AAOI matters here isn’t the single-name equity call — it’s what their earnings call reveals about the structural dynamics at the laser layer.
AAOI is one of a handful of companies globally that fabricates its own indium phosphide laser chips — from epitaxial growth through to final transceiver assembly. In December 2025, they announced a 400mW narrow-linewidth DFB pump laser built on their buried heterostructure platform, delivering over 400mW at 50°C. Those specs position it as an alternative external laser source for silicon photonics and CPO architectures — directly in the supply chain lane occupied by Lumentum’s CW laser products. They’ve since signed an equipment deal with Oxford Instruments for InP etch and deposition systems in their Sugar Land, Texas fab, and broke ground on a new 210,000 square foot manufacturing facility with investment potentially reaching $300 million.
On the earnings call, management confirmed that the industry faces a laser shortage, with some suppliers quoting lead times of over a year. They plan to more than triple laser manufacturing capacity in Texas by mid-2027, with over 95% of production dedicated to AI-specific lasers by year-end 2026. Their “Made in USA” positioning — less than 10% China component exposure with a path to near zero — is a genuine tariff hedge while Chinese transceiver suppliers like InnoLight scramble to shift production to Southeast Asia.
Here’s the tension through the lens of my optical framework: AAOI’s transceiver ambitions are fighting a structural headwind. I wrote in December that “the shift to LPO and eventually CPO restructures this market — component suppliers gain at the expense of module assemblers.” The pluggable transceiver market is brutally competitive, with Chinese suppliers commanding roughly 60% of NVIDIA’s 800G orders. AAOI’s 800G ramp is still delayed by firmware interoperability issues — management says mid-March, but every month of delay is a month where competitors ship volume and build switching costs.
The laser business is a different story. If AAOI’s 400mW pump laser is genuinely competitive with Lumentum’s ELS products, they become a potential supplier into the CPO ecosystem — selling laser sources into Broadcom’s or NVIDIA’s switch architectures. That’s higher-margin and more defensible than assembling OSFP modules. The question — visible in the spec table above — is whether their InP technology can deliver the linewidth, RIN, and thermal stability that CPO environments demand. Lumentum has published those specs. AAOI has published a power number. The gap between a press release and a qualification report is measured in years and billions of test hours.
I don’t hold a position in AAOI and I’m not recommending one. But when a ~$4 billion market cap company says customer demand for transceivers and laser sources exceeds what they can physically manufacture, and they’re tripling fab capacity to try to keep up, that validates the structural thesis. The laser shortage is the new CoWoS bottleneck — and the companies that own their own indium phosphide fabs will set the pace of the 1.6T transition.
What I’m Watching Next
OFC 2026 (Los Angeles, March 16–19) and GTC 2026 (San Jose, March 15–19) run concurrently and are the next major catalysts. I cant be at both! Both Lumentum and Coherent will showcase new products at OFC, and the technical community will evaluate claims side-by-side. I’ll be attending GTC on-site for exclusive coverage of NVIDIA’s AI infrastructure announcements — architecture updates, networking roadmaps, and co-packaged optics developments. Register for GTC here: nvda.ws/4c4Etll.
For Lumentum: Can they execute on $805M midpoint guidance (+85% YoY)? Their InP capacity expansion — exceeding the planned 40% increase, with further expansion in H2 2026 and early 2027 — needs to deliver. The transceiver business remains a margin headwind; they’re considering new fabs and potential acquisitions to meet demand they describe as “challenging to cap at a billion dollars.”
For Coherent: The 6-inch InP yield claims need to show up in the financials. We should see meaningful gross margin expansion by H2 2026 and a plateau in external EML purchases from Lumentum by mid-2027. The OCS “engagements” need to convert to disclosed backlog with dollar figures.
For AAOI: The firmware qualification by mid-March is the near-term binary event. Beyond that, watch whether the 400mW laser moves from samples to volume production — that’s the tell for whether AAOI becomes a CPO supply chain participant or remains a transceiver assembler.
For the sector: The CPO transition looms as the next architectural wave. Jefferies projects CPO taking just 5% share of the DSP market by year-end 2028, with scale-up optical likely a 2029/2030 story. That extends the pluggable transceiver runway while the CPO opportunity builds — good news for all three companies. The external laser source market is materializing exactly as the physics predicted.
The Bottom Line
Light doesn’t lie. You can’t talk your way around insertion loss, linewidth specifications, or reliability qualification data.
Lumentum is performing to spec: record revenue, expanding margins, $400M+ in visible OCS backlog, sold-out high-power laser capacity, and guidance that keeps climbing. Coherent has strong datacenter exposure and a 6-inch InP program that represents a real long-term competitive asset — but the gap between narrative and financials creates an uncertainty premium the market is pricing in. Keysight’s record quarter confirms the demand is real and accelerating — you don’t see nine straight quarters of wireline order growth and a wireline-over-wireless crossover without a structural shift underneath. And AAOI, for all its execution risk, is sending a supply-side signal that matters: indium phosphide capacity is the constraint, and any company with a working InP fab is being pulled into the AI infrastructure supply chain.
The question is no longer whether optical wins — it’s which optical companies capture the margin. Laser go bing, and enjoy the weekend!
BEP Research Is Going Paid — and GTC Is Coming
Since launching, this newsletter has grown to over 1,000 subscribers — including readers from Point72, D.E. Shaw, NVIDIA, Intel, and Micron. In the coming weeks, BEP Research will be transitioning to a paid subscription. Free subscribers will receive an early-bird offer before the paywall goes up — if you’re not subscribed yet, now is the time.
GTC 2026 starts March 16. I’ll be there in person. On-site analysis of NVIDIA’s architecture announcements, networking roadmaps, and co-packaged optics developments will be a cornerstone of paid subscriber coverage.
Resources
About the Author
Ben Pouladian is a Los Angeles-based tech investor and entrepreneur focused on AI infrastructure, semiconductors, and the power systems enabling the next generation of compute. He was co-founder of Deco Lighting (2005–2019), where he helped build one of the leading commercial LED lighting manufacturers in North America. Ben holds an electrical engineering degree from UC San Diego, where he worked in Professor Fainman’s ultrafast nanoscale optics lab on silicon photonics and micro-ring resonators, and interned at Cymer, the company that manufactures the EUV light sources for ASML’s lithography systems. He currently serves as Chairman of the Leadership Board at Terasaki Institute for Biomedical Innovation and is a YPO member. His investment research focuses on AI datacenter infrastructure, GPU computing, and the semiconductor supply chain. Long-term NVIDIA investor since 2016.
Follow on Twitter/X: @benitoz | More at benpouladian.com
Disclosure: I hold positions in LITE and NVDA. I do not hold positions in COHR, KEYS, or AAOI. This is not investment advice — do your own research.
Great article, Ben!
Don’t go paid! Great article. Seems like this spaces’ multiples are so saturated - wondering how to value them and their potential to run? Especially $KEYS