The listing application that Yuanjie Semiconductor filed with the Hong Kong Stock Exchange on March 25, 2026 cites a CIC forecast stating that the data center optical interconnect market will reach $144.4 billion by 2030. That represents a 48.1% CAGR from $13.7 billion in 2024—or simply put, a 10.54x increase. While the number might easily evoke images of AI data center wiring demand, this is not a market forecast for CPO (Co-Packaged Optics) alone. It is a forecast that divides the same market along three separate axes: application, transmission speed, and optical technology.
The application explicitly states that it is a draft, incomplete and subject to change. The CIC research was commissioned and paid for by Yuanjie, which verified public information and interviews with industry sources using its own internal model. Therefore, the $144.4 billion figure should not be read as confirmed revenue, but as a projected value that assumes continued demand growth, technological progress, and policy support, with no extreme force majeure events or unforeseen industry regulations that would have a material or fundamental impact on the market.
Even so, the contents of the tables provide material for considering where supply constraints are likely to emerge, by application and by component. The reason 2030 growth cannot be attributed to CPO adoption alone is reflected in these three breakdowns.
The Three Tables That Make Up the $144.4 Billion: Application, Speed, and Technology
By application, scale-out is projected to be the largest category in 2030 at $64.5 billion. Non-AI data centers are forecast at $40.2 billion, scale-up at $32.1 billion, and scale-across at $7.6 billion. Scale-up has the highest CAGR at 561.5%, but this is because its 2024 base was so small that it displays as 0.0 in the table. The application with the highest growth rate and the application with the largest 2030 revenue are not the same thing.
| Application | 2024 Revenue | 2030 Revenue | 2024–2030 CAGR |
|---|---|---|---|
| Non-AI | $5.9 billion | $40.2 billion | 37.5% |
| Scale-out | $7.6 billion | $64.5 billion | 42.7% |
| Scale-across | $0.1 billion | $7.6 billion | 108.5% |
| Scale-up | $0.0 billion (displayed as such due to small base) | $32.1 billion | 561.5% |
By speed, 1.6T is forecast to reach $65.6 billion and 3.2T $44.5 billion. Combined, these two total $110.1 billion, accounting for 76.2% of the $144.4 billion. 200G-and-below is projected to decline from $3.3 billion to $1.2 billion, while 400G remains roughly flat, moving from $5.9 billion to $6.3 billion. Meanwhile, 800G is projected to grow from $4.5 billion to $26.7 billion. This is consistent with industry descriptions stating that as of 2025, 400G and 800G have reached full-scale commercial deployment and account for roughly half of unit sales volume.
| Speed | 2024 Revenue | 2030 Revenue | 2024–2030 CAGR |
|---|---|---|---|
| 200G and below | $3.3 billion | $1.2 billion | -15.4% |
| 400G | $5.9 billion | $6.3 billion | 1.0% |
| 800G | $4.5 billion | $26.7 billion | 34.8% |
| 1.6T | $0.0 billion (displayed as such due to small base) | $65.6 billion | 867.3% |
| 3.2T | $0.0 billion (displayed as such due to small base) | $44.5 billion | No 2024 base |
The speed table should be read alongside the 2025-era explanation that 1.6T has begun mass production and shipment while 3.2T remains in R&D. The 2030 revenue distribution is not a snapshot of today's availability status. Note also that the three tables—by application, by speed, and by technology—divide the same market along different axes; the figures in each table cannot be added horizontally across tables to construct total market size.
The 63.7% SiPh Share Is Not CPO's Market Share
In the technology breakdown, SiPh (Silicon Photonics) revenue is projected to grow from $4.0 billion in 2024 to $91.9 billion in 2030. Its share of the market is projected to rise from 29.4% to 50.2% by 2027 and 63.7% by 2030. This 63.7% figure represents SiPh's technology revenue share—it is neither CPO's revenue share nor CPO's adoption rate.
SiPh is a technology category that integrates modulators and passive components onto a PIC (Photonic Integrated Circuit). Non-SiPh transceivers use optical and electrical chips plus passive optical components and substrate/mechanical parts. The application explains that in SiPh, the bill of materials concentrates on the SiPh chip and the laser. Because silicon is not an efficient material for emitting laser light, the SiPh approach uses CW lasers as an external or heterogeneously integrated light source. As an illustrative example (which varies by product design), a 400G SiPh transceiver might use two 50mW or 70mW CW lasers, while a 1.6T might use four 70mW or 100mW lasers.
The growth of SiPh does not directly imply reduced demand for III-V-based lasers, precisely because of this division of labor around the light source. In March 2026, OpenLight announced that its PH18DA InP-on-silicon platform, developed with Tower Semiconductor, had received its first mass-production order for NewPhotonics' 800G and 1.6T laser-integrated PICs. The targeted applications include DSP/LPO pluggables and NPO. The expansion of SiPh revenue cannot be explained by CPO adoption alone.
Pluggable, NPO, and CPO differ in where the optical engine is placed relative to the switch ASIC, which shifts the balance among power, density, and maintainability. According to the comparison in the application, electrical link length is 150–250mm for pluggable, 30–100mm for NPO, and under 15mm for CPO. While CPO shortens the electrical path, giving it advantages in density and power, it is harder to maintain than pluggable solutions. Optical transceivers, the current mainstream, can be front-loaded and removed, maintained on a module-by-module basis, and offer mature compatibility and long-distance adaptability. This is not a story where CPO will immediately replace all use cases.
Between the Start of Mass Production and Market Adoption
Primary sources from 2026 feature a lineup of announcements indicating progress in the implementation of CPO and high-speed optical components. However, each points to a different stage of adoption. On May 31, NVIDIA announced that its Spectrum-X Ethernet Photonics is in production, naming CoreWeave, Lambda, and Oracle Cloud Infrastructure as early ecosystem partners and adopters. Meanwhile, the availability timeline listed on the product page is the second half of 2026.
NVIDIA's SN6800 features four Spectrum-6 ASICs and claims 409.6Tbps, with 512 ports at 800Gbps or 2048 ports at 200Gbps. It also presents comparison figures of 5x power efficiency and 5x AI uptime, but these are values based on comparison conditions defined by the company itself and cannot be generalized as performance figures for CPO in general. The term "in production" must be considered separately from questions of which customers will have access, and when.
In October 2025, Broadcom announced the TH6-Davisson, delivering 102.4Tbps at 200Gbps/lane. It is a CPO switch equipped with sixteen 6.4Tbps optical engines and replaceable external laser modules. While the announcement text uses the word "shipping," the availability section states "sampling to early-access customers and partners." The word "shipping" should not be assumed to mean broad commercial availability.
TSMC also announced in April 2026 that its CPO solution using COUPE on substrate would begin production in 2026, claiming 2x power efficiency and one-tenth the latency compared to a pluggable version—though this too is a TSMC-specific comparison. OpenLight's mass-production order is a corporate announcement with limited customers and applications, not a figure representing market-wide mass-production scale. Counting production, early-access sampling, planned availability, and mass-production orders together as a single "adoption" metric overestimates the current state too quickly.
The Conditions for Turning Forecasts into Real Revenue
Behind the CIC forecast lies the premise of a shift toward higher-priced, high-speed products. But demand outlooks for 1.6T and 3.2T do not automatically translate into shipment revenue. Beyond satisfying yield requirements for mass-producing high-speed products, supply of optical engines and CW lasers, and advanced packaging capacity, the process must also clear customer qualification and operational procedures. The fact that pluggable solutions retain an advantage in maintainability is itself a condition that will determine where deployment occurs.
The calculation formula, interview sources, and margin of error behind this forecast have not been disclosed. Given that this is a commissioned study, it would be better to treat the $144.4 billion figure not as a single definitive answer, but as a benchmark for confirming three directional trends: that scale-out will be largest by application, 1.6T largest by speed, and SiPh largest by technology. The availability status of Spectrum-X in the second half of 2026, the commercial rollout beyond early-access sampling, and the mass production and shipment of 3.2T will be the next pieces of evidence bringing this forecast closer to actual demand.