ECOC 2026 | 4.7 pJ/bit and 2031: Coherent Moves CPO from Slides to the Lab Bench — and Pins the Crossover Six Years Out

2 days ago

9 min read

At ECOC 2026 Market Focus, Coherent didn't make the case for CPO one more time — it brought a working machine. On display was a 6.4T near-packaged optics (NPO) engine driven by Coherent's own external laser source module, the ELSFP, at 4.7 pJ/bit (typ) — and that figure already includes the ELSFP. Even more important, the demo was deliberately built as a mixed “NPO-to-pluggable transceiver” link, because the first real systems will inevitably be mixed. On the market side, Coherent cited LightCounting's July 2026 forecast: CPO/NPO port shipments will overtake pluggable transceivers around 2031, even as pluggables keep growing the whole way. The speaker's three-stage framework is Promise → Prototype Evidence → Volume, and his verdict is that the industry now stands firmly in stage two.

1. Why Now: AI Finally Gave CPO a Problem Worth Solving

The speaker opened from an unusual angle. Instead of starting with technology, he started with the idea that “a solution needs a good problem.”

His argument: the traditional data center never gave CPO that problem. Hyperscalers have always cared about power, density and core counts, but that pressure was never enough to justify an architectural leap this big.

AI changed that, because the entire focus of AI became packing as many accelerators as possible into a single cluster. Under that goal, optics and interconnect face three requirements: higher lane rates, lower power (with the savings going to the accelerators, not the interconnect), and — the toughest one — longer reach.

The slide laid out system pressure in four boxes: more accelerators per cluster, higher lane rates, tighter power budgets, and copper reach pressure. The optical response is a single line: Pluggable → LRO → NPO/CPO → Photonic I/O — and it points in only one direction: moving optics closer to compute.

Why does moving closer help? The speaker was clear: the long electrical channel on a switch, running from the ASIC to the front panel, was fine at low speeds and barely needed retimers. But as rates rise, that channel becomes more and more punishing — you need stronger, more expensive, more complex DSPs. Remove the channel and you remove the need for the DSP; once the DSP goes, power drops, integration goes up, and bandwidth density goes up.

System pressure in four boxes (more accelerators / higher lane rates / tighter power budgets / copper reach pressure) and the optical response path Pluggable → LRO → NPO/CPO → Photonic I/O. Source: Justin Abbott, Coherent Corp — ECOC 2026 Market Focus

2. The Most Valuable Slide of the Talk: A Machine That Actually Runs

At this point, most CPO talks drift into vision statements. Coherent didn't — it put the demo's measurement results straight on screen.

Architecture: two NPOs (vertically integrated inside Coherent), driven by its own ELSFP external laser source module — even the laser is in-house.

Power: 4.7 pJ/bit (typ). The speaker stressed that this figure includes the ELSFP, not just the optical engine. That matters enormously when comparing vendors' CPO power numbers, because many published figures leave the external light source out.

Design intent: the demo was deliberately built as an NPO-to-pluggable transceiver link. The reason: the first real-world systems won't be all-NPO; they will be mixed — even within the same switch (some optical engines, some transceivers). Interoperability is not a nice-to-have; it is the price of entry.

Measurements: top right is the Tx eye diagram of NPO #1, shown both “on-target” and “through a reference receiver” — clean eyes. Bottom right is the bit error rate of the NPO into a 1.6T FRO (reference receiver). The speaker candidly said this is still being optimized, but the point isn't the absolute number — the point is consistency across all 32 channels. In something this highly integrated, you want every channel equally good, not one hero lane.

The value of this slide is that it makes the “prototype evidence” stage concrete. A demo is not volume production, but without a demo you don't even get to talk about volume.

Key data from Coherent's live NPO demo at ECOC 2026: 6.4T NPO + ELSFP, 4.7 pJ/bit (ELSFP included), BER consistency across 32 channels, NPO-to-transceiver interop. Source: Simple Tech Trend | Data: Justin Abbott, Coherent Corp — ECOC 2026 Market Focus

3. The Market: The Crossover Is in 2031, but It Isn't Zero-Sum

Coherent cited LightCounting's July 2026 Cloud Data Center Optics report, scoped to cloud data center port shipments at 1.6T and 3.2T with reach of 2 km or less (in millions of units).

Three takeaways sat on the right side of the slide:

First, from concept to hard forecast. CPO/NPO is now a quantified forecast line, not a concept.

Second, it is not an either-or market. Transceivers keep growing across the entire forecast window — the two architectures scale up together.

Third, the crossover lands around 2031. What drives it is power per bit, bandwidth density and shrinking electrical reach.

The speaker added a line on stage that I think deserves to be quoted most: “The question is no longer whether CPO/NPO will ramp, but how fast.”

Here Taiwan's supply chain should keep a cool head: 2031 is six years away. The gray pluggable curve keeps growing over the whole period and remains larger than CPO/NPO until 2031. If your revenue still rests on pluggables, this chart is not an alarm; it is a reasonably generous transition window — but nothing more than that.

LightCounting's July 2026 forecast of the CPO/NPO vs transceiver shipment crossover (around 2031), and the very different adoption triggers for scale-out and scale-up. Source: Simple Tech Trend | Data: LightCounting Cloud Data Center Optics, July 2026, as cited by Coherent at ECOC 2026

4. The Socket Is the Point: OpenCPX Brings the Pluggable Playbook to CPO

This is the most institutional part of the talk — and the part technologists most easily underestimate.

The speaker's argument: the pluggable market succeeded because of standardized form factors. They spared optics suppliers and system vendors from each going down their own proprietary path (sometimes several inside the same company), let resources concentrate, and that is what allowed competition and scale to grow.

The OpenCPX MSA wants to bring the same thinking to CPO and NPO: define a form factor while not fixing what isn't broken — signaling, power/control and management all reuse existing standards (ITU, OIF, etc.) to reduce new risk. The goal is not to replace those standards but to give the market a common core document so investment converges on one path, and to enable multi-sourcing.

The four ecosystem benefits on the slide are straightforward: define a common form factor (moving away from proprietary solutions), reuse what has already been proven, enable multi-sourcing, and a roadmap with a future — 200G today, with a path to 400G connectivity.

And the closing line most worth remembering sat at the bottom:

The key value of the socket is not that it connects the optical engine to the system, but that it creates an open ecosystem.

The first OpenCPX specification was released a week before ECOC, and phase two is already under discussion. We broke this down earlier — Open CPX 1.0 Teardown: CPO Finally Has a Standard Socket, with Dimensions, Forces and Tolerances Laid Out.

5. Laser Strategy: External and Integrated Are Not Sequential — They Coexist

Coherent's first-generation solution uses an external light source, but the speaker explicitly rejected the narrative that “external is a stopgap and integrated is the destination.”

Advantages of external laser sources (ELS): serviceable, replaceable light sources; decoupling the light-source learning curve from the optical-engine adoption curve; letting the laser and optical-engine roadmaps evolve independently (he cited OCI as an example — if you're moving to tightly spaced DWDM lasers and the optical engine is microring-based, keeping the two separate actually makes progress easier); and thermal isolation from the ASIC package.

Advantages of integrated lasers: better density and easier fiber routing inside the switch — but at the cost of requiring package-level thermal and reliability qualification.

The architecture evolution chart has four stops: 6.4T external laser → 6.4T integrated laser → 6.4T hybrid integration → 12.8T OCI. But the line underneath is the real point: multiple architectures will coexist. The speaker even said the last two stops (hybrid integration and OCI) are likely to compete with each other in the market — and “competition” doesn't mean there will be only one winner.

The final line on this slide distills the whole talk: Requirements determine the architecture.

Coherent's four-stop architecture evolution (6.4T external laser → integrated laser → hybrid integration → 12.8T OCI) and the trade-offs between external light sources and integrated lasers. Source: Simple Tech Trend | Data: Justin Abbott, Coherent Corp — ECOC 2026 Market Focus

6. Two Markets, Two Completely Different Reasons to Pull the Trigger

This is another framework worth copying down. The speaker asked: is the adoption trigger for CPO/NPO the same in every application? The answer is no.

Scale-out: the system constraint is power and front-panel density; the value of CPO/NPO is more bandwidth in the same switch volume; the trigger is “when pluggable power or faceplate density becomes the limit.”

Scale-up: the system constraint is copper reach and topology limits; the value of CPO/NPO is extending accelerator connectivity and cluster size; the trigger is “when copper reach or retimer power becomes the limit.”

The reasons are entirely different, but the slide's conclusion is: both markets are approaching their respective adoption triggers.

For the supply chain, the implication is concrete: you can't use one product story to win both markets. The pitch into scale-out is “save the DSP power and free up front-panel space”; the pitch into scale-up is “copper can't get there, I can — and at lower power.” These are different customers, different procurement timelines and different qualification items.

7. What the Talk Left Unanswered

First, cost appeared as just a few words on one chart. “Lower cost per bit over time” — how much time? The speaker didn't elaborate, and that is exactly the number system vendors need most when they make procurement decisions in 2027.

Second, the preconditions for volume were deferred to “a future talk.” The speaker himself said confidence must come before volume, and that confidence comes from margin across electrical, optical and thermal dimensions — but he made clear he would not go deep on that today. It was the most honest line of the talk, and the one that most invites follow-up questions — because that is what really determines the slope from 2028 to 2031.

Third, the 2031 crossover is LightCounting's number, not Coherent's. The speaker himself said “any forecast can be wrong.” Use it as an indicator of industry consensus, but discount it as an investment timeline.

8. Conclusion

Placed in the broader context of ECOC 2026, Coherent's talk fills in the “evidence” box. Huawei provided a scorecard, Corning provided a reliability budget, and Coherent provided a running machine with a set of measurements.

For Taiwan's supply chain, three takeaways:

First, read the parentheses around 4.7 pJ/bit carefully. It includes the external laser source module. When you compare CPO/NPO power figures across vendors, the first question to ask is “does it include the ELS?” — the difference between with and without is enough to make two very different-looking numbers equally good, or the reverse. For Taiwanese companies building ELSFPs and high-power CW lasers, those parentheses are your proof of value.

Second, “mixed” is the default shape of first-generation systems, not a transitional compromise. Coherent deliberately built the demo as NPO-to-transceiver and even noted that a single switch will be mixed. That means pluggable volume won't drop just because CPO goes live, and that interoperability testing will become a distinct, required capability. For test, validation and interop service providers, this is a new demand taking shape.

Third, the openness of the socket determines whether Taiwanese vendors get a seat. By standardizing the form factor, OpenCPX shifts CPO from a market “only a few can play in” to a multi-sourced market structure — and multi-sourcing is exactly the battlefield Taiwan's supply chain has excelled at for 30 years. The path from 200G to 400G is the spec race of the next three years. For related context, see 2026 OCP APAC Summit | CPO / NPO / XPO Panel: Not a Route War, but the 409.6T Bottleneck 18 Months Out.

Verdict: The CPO debate is over; what remains is a fight over the slope. The 2031 crossover gives the pluggable camp a six-year buffer, but the message to the CPO supply chain is more urgent — because what determines that slope isn't technical feasibility, it's whether you can build a lot of them reliably. And preparing for that has to start now.

This article is for technology and industry trend analysis only and does not constitute investment advice.