OCP Global Summit 2026 | Genuine Optics | Haijiang Yu | Silicon Photonics as the Full-Stack Optical Interconnect Platform

1. Why Listen to Genuine Optics on Silicon Photonics

Genuine Optics is not a newcomer. Its slides describe it as a top-8 global optical module supplier in 2024, headquartered in San Jose, with R&D in Wuhan and a manufacturing plant in Chonburi, Thailand, delivering "millions" of transceivers.

What sets the company apart is that it treats silicon photonics as the foundation of its entire product line, not as one product. Haijiang Yu's talk boils down to one sentence: from 800G pluggables to CPO, everything is built on the same SiPh platform.

That is the opposite of most module makers, which lead with EML and treat SiPh as secondary. The question: is fighting three battles on one platform a moat, or a dilution of resources?

The value of silicon photonics is not how fast one chip runs, but whether one platform can grow from pluggables all the way to CPO.

2. Breaking Down the Market: Compute Grows 100x Every Two Years, Interconnect Only 1.4x

The opening chart is familiar: AI model compute grows about 100x every two years, GPU performance 3.3x, while memory bandwidth and interconnect bandwidth grow only 1.4x. Genuine Optics calls this gap the main bottleneck of AI clusters.

Citing LightCounting forecasts, the slides make two points:

• Pluggables (retimed TRX) remain the mainstream in the near term, but LPO, NPO and CPO start growing exponentially in 2027–2028.

• NPO leads in scale-out, CPO mainly in scale-up. Per the LightCounting charts, NPO volume ramps ahead of CPO, whose shipments depend on GPU and accelerator scale-up fabrics.

This matches what we argued in NPO Is Not a Stepping Stone to CPO: NPO is not a transitional product but a standalone market that will coexist with CPO for the long run.

3. Breaking Down the Architecture: Why Bet on NPO

The slides lay out the three architectures in one table:

Dimension	Pluggable	NPO	CPO
Optics position	Front panel	Host board, under 10 mm from the ASIC	Directly on the ASIC package
Power	Baseline, high electrical loss	30–50% savings	50%+ savings
Thermal	Easy, independent airflow	Manageable, separate cooling	Complex, high thermal density
Field repair	100% hot-swap	High, modular sockets	Extremely difficult, board scrap
Ecosystem	Mature, multi-vendor	Emerging OIF standard	Early R&D, proprietary

▲ Genuine Optics compares the trade-offs of pluggables, NPO and CPO in power, thermals, serviceability and ecosystem, and concludes that NPO is the most pragmatic (Image source: Genuine Optics | 2026 OCP Global Summit presentation (06:09))

Genuine Optics' conclusion is blunt: NPO balances performance, engineering feasibility and ecosystem readiness. In other words, it captures most of CPO's power savings without the risk that one failed optical engine scraps an entire switch board.

Huawei ran a similar scorecard at ECOC, with NPO beating CPO 4:2 (see ECOC 2026: Huawei Scores CPO vs NPO, and NPO Leads 4:2). Two suppliers reaching the same conclusion means NPO has moved from fallback to a direction vendors actively bet on.

4. Breaking Down the Roadmap: Three Parallel Tracks to 2028

The roadmap pushes lane rates from 100G and 200G to 400G, and aggregate bandwidth from 800G to 12.8T:

• Pluggable: 400G DR4 (2023) → 800G DR8 → 1.6T DR8 → 12.8T XPO / 1.6T DR4 (2026) → 3.2T DR8 (2027) → 25.6T XPO (2028+, with a question mark on the slide)

• NPO: 3.2T TRx MZM PIC (2024) → 3.2T NPO optical engine / ELSFP (2025) → 6.4T NPO (2026) → 12.8T NPO (2027–2028+)

• CPO / OIO: 3.2T MRM PIC (2024) → 3.2T MRM CPO optical engine / DWDM ELSFP (2025–2026) → 400G / 800G OCI optical engine (2027) → 1.6T+ OCI (2028+)

▲ Genuine Optics' three-track roadmap for pluggables, NPO and CPO/OIO, with 12.8T XPO and 6.4T NPO as the 2026 focus (Image source: Genuine Optics | 2026 OCP Global Summit presentation (08:13))

What is in mass production today are 800G and 1.6T modules on 100G and 200G lanes; 400G lanes and 3.2T NPO/CPO optical engines are still at the live-demo stage. Hard numbers from the demos: 12.8T XPO under 130 W with 500 m single-mode reach; 3.2T NPO with loopback BER better than 1E-10 on all 32 channels; and a 400G-lane 1.6T DR4 using in-house SiPh/TFLN plus a 3 nm 400G DSP at about 30 W.

Get the timing right: in 2026 you can buy 1.6T and 12.8T XPO pluggables; NPO is a 2026–2027 product; CPO waits for the OCI-MSA standard.

5. Breaking Down SiPh Economics: Less DSP, and Six Fewer Lasers

This is the most useful part of the talk. For its 1.6T OSFP (200G lanes), Genuine Optics gave maximum power for three architectures:

Architecture	Power	Loopback BER
Full DSP	Under 24 W	~1E-13
LRO (half DSP)	Under 15 W	~1E-13
LPO (no DSP)	Under 10 W	Under 1E-8

LRO saves about 40% and LPO about 60%. More important, LRO's BER is in the same class as full DSP; the slides say LRO still reaches 1E-13 at 24 dB die-to-die loss. That makes half-DSP the power-saving option most likely to ramp in the pluggable world.

The second ledger is lasers. SiPh uses an external CW laser with MZM modulation, so one laser can feed multiple lanes: 1x2 or 1x4 sharing in the DR4 transmitter, 1x4 or 1x8 in the DR8 transmitter; the 400G-lane 1.6T DR4 uses a single 100 mW CW laser split four ways. Versus EML designs with one EML per lane, an 8-lane DR8 drops from 8 lasers to 2.

▲ Power comparison of Full DSP, LRO and LPO 1.6T modules, and how SiPh laser sharing lets an 8-channel transmitter use only 2 CW lasers (Image source: Simple Tech Trend original, data from Genuine Optics' 2026 OCP Global Summit presentation)

With EML capacity tight, this matters beyond cost: silicon photonics cuts the use of the hardest-to-scale component, the laser, to a quarter, swapping it for high-power CW lasers and mature CMOS foundries.

Reliability has long been the main doubt about SiPh. Genuine Optics now offers two years of field data: over 1 million SiPh modules, more than 8 billion device-hours, and a PIC FIT below 0.2, with FIT falling quarter by quarter as shipments accumulate.

6. Breaking Down Lasers: External Light Sources Are the Real Challenge for NPO and CPO

Both NPO and CPO move the laser out of the optical engine into a pluggable external light source (ELSFP). Genuine Optics' specs:

• ELSFP for NPO: 1311 nm, ≥20 dBm per lane (100 mW), under 8.5 W, 12-fiber MT interface with 8 PMF, OIF compliant.

• DWDM ELSFP for CPO, with two approaches in parallel: a quantum-dot comb laser at about 10 dBm per comb with 200/400 GHz spacing, and a DFB array at 200 mW per channel at 50°C with 400 GHz spacing.

▲ KPI radar chart comparing a quantum-dot comb laser with a DFB-array ELSFP; the DFB array leads in per-channel power and conversion efficiency (Image source: Genuine Optics | 2026 OCP Global Summit presentation (16:41))

The radar chart is candid: the QD comb wins on coupling ease, fewer fibers and uniform channel spacing, but the DFB array clearly leads in per-channel power, power conversion efficiency and technology maturity.

We covered the split over laser power specs in CPO Lasers Don't Have to Chase 400 mW. Genuine Optics is building both and letting customers choose, which shows the OCI-MSA light source spec is not settled yet.

The CPO optical engine itself stands out for a 12-layer TGV glass substrate with 2.5D/3D packaging, unidirectional bandwidth density above 3 Tbps/mm, and live demos at OFC 2026 and CIOE 2026.

7. Risks and Counterarguments

• Resource strain of three tracks: pushing pluggable, NPO and CPO at once splits R&D and capacity three ways. For a top-8 rather than top-3 module maker, that is a real challenge.

• 400G lanes are still early: the 1.6T DR4 demo showed BER around 1E-6, a pre-FEC number with a way to go before production spec.

• LPO's BER gap: LPO saves 60% but only reaches 1E-8 BER, relying on host SerDes, and system vendors' acceptance remains in question.

• Scope of the FIT number: FIT below 0.2 counts only random PIC failures, excluding lasers, DSPs and assembly, so it is not the same as module reliability.

• Geopolitical risk: R&D in Wuhan and manufacturing in Thailand, with the slides stressing "3+ Tier-1 foundries outside China", show that supply chain origin review is now a sales gate.

8. What It Means for Taiwan's Supply Chain

• SiPh foundry: capacity outside China is a selling point. Genuine Optics states that its SiPh/TFLN wafers come from 3+ Tier-1 foundries "outside China", with more than 30 MPW and full-reticle tape-outs. The slides do not name the foundries, so whether Taiwanese foundries are among them cannot be confirmed; but demand for non-China SiPh capacity is itself an opening for Taiwan's SiPh foundries.

• CW lasers: demand shifts from EML to high-power CW. Pluggables use a 100 mW CW laser split four ways, NPO ELSFPs need 20 dBm per lane, and CPO DFB arrays 200 mW. These are clear spec sheets for Taiwan's InP laser chip makers, with the challenge being power and reliability at high temperature.

• Packaging and substrates: TGV glass enters CPO. The 12-layer TGV glass substrate is one of the cores of Genuine Optics' CPO engine. Taiwan's glass substrate and advanced packaging supply chain is worth tracking, though the slides name no suppliers.

• Manufacturing: Thailand is the common answer for module makers. Genuine Optics produces in Thailand, choosing Southeast Asian capacity outside China like many module makers, so Taiwanese module makers in Thailand face more direct competition.

9. Conclusion

The message is clear: silicon photonics is no longer an alternative to EML but the common chassis of the whole optical interconnect line. 40–60% power savings from less DSP, three-quarters fewer lasers from sharing, and FIT data from a million shipped modules together explain why SiPh wins in the 1.6T generation.

But separate the timelines. In 2026, 200G-lane 1.6T and 12.8T XPO ship in volume; 6.4T NPO is 2026–2027; CPO waits for OCI-MSA and external light source specs. Over the next 12 months, watch three things: when 6.4T NPO moves from demo to shipment, whether OCI-MSA picks QD comb or DFB array, and when 400G-lane BER reaches production spec.

By cutting laser use to a quarter, silicon photonics rewrites the optical module supply chain, not just the power table.

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