ZR Is the Meat, Coherent Lite Is the Bone: The Long Game in Optical Transceivers for the Next Decade
The two terms most often misused in optical communications lately are ZR and Coherent Lite.
Many people treat them as different stops on the same road: both are coherent pluggables, just getting faster. That is badly wrong. They are not two stops on one road; they are two routes pushed to opposite ends of the same technology spectrum. One handles long-haul transport beyond 80 km, the other squeezes into the data center at under 20 km. They serve different customers, at different price points, on different timelines.
More importantly, the two routes are in completely different market positions today. ZR is already making real money; Coherent Lite is still waiting for its first customer to commit.
This article breaks down three things: why ZR is "really" ramping, why Coherent Lite is still in the narrative stage, and, if you care about Taiwanese suppliers or the investment angle, how to think about each of the two lines.
1. Two Things Often Lumped Together, Fighting on Different Battlefields
First, let's draw clean definitions.
The ZR / ZR+ family (400ZR, 800ZR, 1600ZR) targets data center interconnect beyond 80–120 km (DCI) and metro networks. OIF began defining 400ZR in 2019, volume production started in 2021, 800ZR/ZR+ is ramping at scale in 2026, and 1600ZR is expected to begin adoption in 2027. This route is characterized by long reach, high ASP, modest volume but healthy margins.
Coherent Lite targets 2–20 km intra-data-center and campus interconnect. Its core pitch is using cheaper, lower-power coherent technology to replace IMDD (intensity modulation direct detection), which is already hitting its bottleneck. OIF has defined a 1.6T version, now called 1600 Coherent Lite (1600CL). It is characterized by short reach, ASP that must be pushed close to IMDD, and very high volume.
The supply chain view makes this clearer. At ECOC 2025 Marvell offered a contrast: in 2024, IMDD optical modules shipped inside data centers exceeded 60 million units, while full coherent ports between data centers numbered only about 1 million ports. That 60x gap is exactly what Coherent Lite wants to eat.
In one sentence: ZR goes after the long-reach market IMDD can never reach; Coherent Lite goes after the short-reach market where IMDD can no longer keep up beyond 1.6T.
2. ZR Is in the "Real" Stage: All Three Conditions Are Aligned
To judge whether a technology trend is "real," look at three things: has demand committed, is supply ramping, and is the timeline concrete? For ZR, all three are yes.
Demand: the six major hyperscalers (Google, Microsoft, Meta, Amazon, Apple, TikTok) issued large 800G+ DCI tenders in the second half of 2025, with volume shipments in 2026. Cignal AI's 2026 Q1 report states it outright: pluggable coherent optics are now the dominant form of telecom optics outside the data center, leading in deployed bandwidth, market size and bandwidth growth. Embedded coherent (traditional chassis-based) will in the long run be left only with the toughest routes, such as submarine cables and ultra-long-haul.
Supply: Coherent Corp., Lumentum, Cisco (Acacia), 富通電子 and Marvell are all ramping 800ZRx. On the Taiwan side, Elaser and Luxnet are also expanding EML and packaging capacity. Overall monthly capacity in 2026 keeps up with demand but is tight; the biggest bottleneck remains the upstream 200G EML chip.
Timeline: the OIF specifications are finalized and IEEE 802.3 standardization continues. Market size also has third-party backing: Research and Markets estimates ZR+ coherent optics at US$1.84 billion in 2025, US$2.19 billion in 2026 (18.9% CAGR), and US$4.32 billion by 2030. Dell'Oro earlier put overall coherent transceiver shipments at a 17% CAGR over the next five years.
Verdict: for ZR, the question is not whether it takes off, but how fast 800ZRx penetration climbs and when 1600ZR takes over in volume. Only two indicators matter in the second half: 800ZR/ZR+ penetration in DCI applications, and progress on production-grade 1600ZR demos before OFC 2027.
3. Coherent Lite: The Technology Has Passed, but Only One Customer Is Waiting to Commit
Coherent Lite is a different story; the state of all three conditions is very different.
On technology, it has cleared the first hurdle, and cleared it well. At ISSCC 2026, Marvell presented the industry's first dual-channel 800 Gb/s Coherent Lite silicon photonics transceiver built on a 5nm FinFET process. The chip achieves three things that matter a great deal for the industry:
First, total round-trip latency below 300 ns, versus more than 3 μs for traditional 400ZR, a 10x reduction. Latency has been the core barrier keeping coherent out of the data center: server interconnect is so latency-sensitive that even microseconds are too slow. Marvell pushed it into the nanosecond range, effectively earning coherent its physical ticket into intra-DC.
Second, 2x better analog power efficiency than mainstream coherent solutions. How does Coherent Lite achieve that? It skips the complex chromatic dispersion compensation and polarization tracking that long-haul coherent requires; at only 2–20 km, these can be drastically simplified. Using O-band lasers also relaxes frequency-accuracy requirements, allowing the low-cost DFB lasers used in IMDD systems to be reused.
Third, a single chip integrating 8 lanes of 100G PAM-4 host interface plus 2 lanes of 400G Coherent Lite line interface, mapping directly to 1.6T OSFP-XD module requirements. In testing it covered 40 km of standard single-mode fiber without optical amplification, with ample link margin.
Technically, Coherent Lite is no longer a question of whether it can be built, but of who will pay for it.
That is the problem. Cignal AI's September 2025 report, "Coherent Lite: Technology, Applications and Forecast," puts it bluntly: "Coherent Lite currently has only one major, uncommitted potential customer (Amazon), which makes the market size difficult to estimate." Most suppliers view today's Coherent Lite as a stepping stone toward intra-data-center coherent optics, with real volume waiting for the 3.2T generation.
Suppliers are betting the same way. Nokia's full coherent pluggable lineup launched in March 2026 includes 1600ZR/ZR+, a 2.4T high-performance pluggable, and a Coherent Lite solution that jumps straight to 3.2T, with samples expected in mid-2027. Note the signal: Nokia did not build a 1.6T Coherent Lite first; it went straight to 3.2T. That means suppliers themselves believe 1.6T Coherent Lite volume is not big enough, and that the real takeoff requires betting on 3.2T.
So Coherent Lite's current status is: technology yes, customer commitment half a yes, volume production timeline not yet concrete. This is the narrative stage: you can follow the story, but don't overstay.
4. Why Some Say Coherent Lite Volume Could Be 10x DCI
This is the most important number in the entire Coherent Lite bet.
Cignal AI estimates that if Coherent Lite truly penetrates inside the data center, deployments could reach 10x DCI module deployments. The math is simple: DCI connects data center to data center, and there are not many such links worldwide; intra-DC connects rack to rack and hall to hall. A single GB200 NVL72 rack alone needs 72 800G plus 18 1.6T optical modules, and one hyperscaler builds hundreds of these racks a year.
By that math, if DCI ZR stays in the million-port class, the ceiling for Coherent Lite is the ten-million-port class.
But an ASP discount has to be factored in. To replace IMDD inside the data center, Coherent Lite must be priced at "slightly more expensive, but worth it for the performance". It cannot command ZR-like prices, because its competitor is IMDD optics that have been in volume production for years at extremely low cost. A rough estimate puts Coherent Lite ASP at roughly one-third to one-half of same-speed ZR.
So the realistic revenue ceiling: 10x the ports at 0.3–0.5x the ASP gives a revenue TAM of roughly 3–5x DCI ZR. If DCI ZR is US$4.3 billion in 2030, a fully ramped Coherent Lite would be a US$13–20 billion market.
That is why both Marvell and Nokia are betting on it. ZR is the meat, Coherent Lite is the bone: the meat tastes good, but the bone holds up longer, and whoever secures the position first gets a 10-year moat.
But this 10x estimate has a precondition: Coherent Lite must actually penetrate intra-DC. If it stops at the campus layer (2–20 km) and cannot get into the true data center interior, the ceiling shrinks to 3–5x rather than 10x. That depends on how the LPO and CPO competition plays out over the next 18 months.
5. Three Signals That Would Reverse This View
Every view must be testable. The Coherent Lite view should be reassessed based on three signals:
If Amazon publicly commits to a volume production timeline, Coherent Lite immediately upgrades from narrative to "real," pulling the entire industry timeline forward by six months to a year. This is the single most important signal and the most worth tracking.
If LPO or CPO wins hyperscaler design wins in the 1.6T generation, Coherent Lite's window gets squeezed. The worst case is that it is skipped entirely at 1.6T and only gets a chance at 3.2T, by which time CPO will be competing for the same sockets.
If OFC 2027 shows no production-grade 1600 Coherent Lite demo, the 2027–2028 deployment timeline slips to 2028–2029, and the whole narrative cools for at least a year.
The most concrete near-term event to watch: the OIF session at OFC in March 2026 (SF8: 800ZR/LR and 1600ZR/ZR+/CL Implementation Agreements), which lays out the full status of the 1600CL specification. After that session, the Coherent Lite timeline goes from "a vague two or three years" to "a specific quarter."
6. Conclusion: 2026 Is About Making Money on ZR; 2028 Is About Whether Coherent Lite Takes Off
Put the two lines side by side, and the industry rhythm is actually very clear:
2026–2027 is ZR's volume boom. 800ZR/ZR+ penetration rises fast, and 1600ZR starts taking over in volume in the second half of 2027. The money in this period goes to vendors with existing ZR product lines (Coherent Corp., Lumentum, Cisco, Marvell) and to Taiwanese suppliers of EML and packaging/test. This is cash flow, happening now.
2027–2028 is Coherent Lite's test period. If Amazon commits between the second half of 2026 and the first half of 2027, the first volume design wins will appear in late 2027 and ramp from 2028. If Amazon never commits and the other hyperscalers choose LPO, Coherent Lite will be pushed to the 3.2T generation (i.e., after 2029) before it takes off. This is a bet, on the future.
After 2028 comes the showdown between Coherent Lite and CPO. Whether the 3.2T-generation data center interior is dominated by Coherent Lite or CPO (co-packaged optics) will be settled during 2027–2028. Cignal AI itself has said: "CPO is still inevitable, but whether it is imminent, the next 12 months will tell." Those 12 months run from now to early 2027.
For Taiwanese suppliers, this rhythm means two things:
Near term (2026–2027): lock in the 800ZR/ZR+ EML chip, silicon photonics foundry and packaging/test links. The market is already clear; whoever can expand capacity and get into tier-1 customer supply chains will capture this wave.
Mid term (2027–2028): start tracking Coherent Lite's low-power DSP and O-band silicon photonics capabilities. Taiwanese suppliers already have a foundation in silicon photonics, EML chips and packaging/test, but low-power DSP and system integration are the home turf of design houses like Marvell, Broadcom and Nokia. Taiwan's opportunity is in upstream components, not the DSP itself.
One last line to close:
Twenty years of optical communications history tell us that every new technology takes 3–5 years to go from "physically feasible" to "volume ramp." Coherent Lite has already crossed the physical-feasibility threshold. What comes next is whether hyperscalers are willing to pay, when, and how much. This is no longer a technology question; it is a business question.
This article is for technology and industry trend analysis only and does not constitute investment advice.
Comments