The CPO Era Officially Begins: Market to Top $1.5B by 2030 | Decoding Cignal AI's First Quantitative Forecast and the Key Role of ELSFP

Introduction

For years, co-packaged optics (CPO) has been the optical industry's phantom technology - "always five years away." But with market research firm Cignal AI recently publishing the industry's first quantitative forecast for CPO and ELSFP, that phantom has turned into real capital spending. Driven hard by Nvidia and Broadcom, CPO has crossed the commercial threshold. This is not just a 1.6T/3.2T spec upgrade; it is a survival battle as AI compute architecture moves from copper to optics - and the high-power laser packaging process hidden behind it will decide who captures this windfall.

Key Insights

Standardization: 200G/lane Is the Singularity for the CPO Boom

According to Cignal AI, CPO port shipments will begin to emerge in 2027 and accelerate sharply in 2029-2030, with global deployments expected to exceed 30 million ports in 2030. The underlying logic of this boom is the technology node where per-lane speed reaches 200G PAM4. As SerDes speeds double, traditional pluggable modules face major power and signal integrity (SI) challenges in the 1.6T and even 3.2T era. By shortening the electrical trace from ASIC to optical engine, CPO eliminates the power-hungry DSP and fundamentally breaks through the power wall.

Differentiation: ELSFP Is the Real Hero Behind CPO Adoption

Why is CPO for real this time? The key is the maturity of the External Laser Small Form-factor Pluggable (ELSFP). CPO's biggest pain point has always been that "lasers hate heat" - packaging a laser next to a switch ASIC drawing up to a kilowatt is a slow death for the laser. ELSFP's design logic is to pull the heat source away from the ASIC and onto the front panel, feeding light into the optical engine via polarization-maintaining fiber (PMF). This not only solves thermal management but, more importantly, delivers serviceability - when a laser ages and fails, you simply hot-swap the ELSFP instead of scrapping an expensive switch. Cignal AI forecasts the ELSFP module market will grow to $1.5 billion per year by 2030.

The Hidden Barrier for High-Power Lasers: The TO-CAN to CoS Process Shakeout

Behind the ELSFP ramp, the industry tends to fixate on InP (indium phosphide) epitaxy capacity, but the real volume bottleneck lies in the sub-assembly packaging process. To overcome CPO's high optical loss, ELSFP requires very high-power continuous-wave (CW) lasers. At this power level, the TO-CAN sub-assembly packaging common in datacom is completely out of the game - its thermal structure and high-frequency characteristics simply cannot handle the heat density of the CPO era. Laser bare die must move entirely to Chip on Submount (CoS). Only by precisely bonding the bare die directly onto high-thermal-conductivity substrates such as aluminum nitride (AlN) can you secure an optimal thermal path and high-frequency signal integrity. CoS imposes stringent requirements on die-attach accuracy, thermal stress control and subsequent high-precision coupling to the FAU (fiber array unit). In other words, vendors that cannot master CoS packaging and high-yield volume production will be eliminated from the CPO supply chain.

Industry Chain Reaction

This shift is not about wiping out traditional pluggables entirely, but about a new division of labor in network architecture:

• Scale-out (cross-rack networking): Pluggables will still dominate in the near term. We see Marvell and Coherent continuing to optimize DSP and LPO solutions to extend the pluggable life cycle.

• Scale-up (GPU-to-GPU interconnect): This is CPO's real battlefield. As copper's effective reach at the 200G node shrinks to under 1 meter, CPO is seamlessly taking over from copper cables.

• Big players' dual-track strategies: Nvidia is taking on the risk of the first wave of system-level deployments with Quantum-X and Spectrum-X, while Broadcom is preparing to harvest at scale with its Davisson platform once the technology is validated. Across the supply chain, beyond the major InP capacity holders (such as Lumentum), OSATs with high-precision CoS packaging capability and advanced silicon photonics wafer-level packaging (such as TSMC COUPE) are set for a re-rating.

Conclusion

STT's conclusion is clear: in AI scale-up networks, CPO is definitely the "endgame," and 1.6T pluggables are only a transition.

Technology indicators to watch over the next 6-12 months:

1. CoS packaging yield: Watch the actual volume yield of Chip on Submount for high-power CW lasers in the supply chain - it will directly determine how smoothly ELSFP supply flows.

2. OIF standardization progress: Results from major vendors' Plugfest testing of ELSFP blind-mate and optical interface interoperability.

3. Nvidia deployment data: The real-world MTBF (mean time between failures) of optical paths and ELSFPs once Blackwell goes live in large data centers.