ECOC 2025 Tech Focus: Datacom Modulators at a Crossroads

Introduction

At ECOC 2025, market research firm fibeReality, LLC shared its distinctive take on the development of data center modulators and optical interconnect technology. Unlike vendor technology showcases, this talk had a clearly opinionated tone, questioning the optical circuit switching (OCS) hype, the real value of CPO, and the bottlenecks in silicon photonics, and discussing the tug-of-war between the “fast and narrow” and “slow and wide” technology paths.

Content

1. The optical circuit switching (OCS) controversy

• Source of the hype: Google’s and CloudLight’s involvement sparked the OCS buzz, but the speaker questioned its practicality, arguing its operational problems are serious, and didn’t rule out Google abandoning it in the future.

• Critique: the OCS hype resembles the dot-com bubble era, with marketing and consulting services benefiting the most rather than real operational value.

• Industry comparison: if OCS really had huge value, Amazon would have followed long ago, but in fact it hasn’t.

2. The debate over CPO and network power

• NVIDIA’s role: the speaker argued that the push for CPO is driven more by NVIDIA’s desire to grow switch revenue than by a genuine power bottleneck.

• Power share: in his view, networking accounts for <5% of total data center power, and CPO’s savings are under 1%, so its value shouldn’t be overstated.

• Possible upside: the CPO hype at least keeps industry attention on silicon photonics, keeping the 200G-and-above SiPh path competitive.

3. Silicon photonics’ bottlenecks and alternatives

• The 200G limit: the speaker believes silicon photonics is already close to its limit at 200G/lane.

• Potential paths:

  • 8 × 200G (1.6T aggregate) configurations may still have a market.

  • InP (indium phosphide) and DFB lasers will keep their edge, especially in 1.6T ZR applications.

  • Chinese vendors may enter the market quickly through conventional InP/DFB processes.

4. The modulator dilemma

• Fast and narrow vs. slow and wide:

  • One camp backs high-speed lanes (200G/400G per lane), while the other believes lowering per-lane speed and adding more lanes is more practical.

  • Hyperscalers such as Google may stick with their existing solutions, and differing requirements across customers will add complexity for module makers and suppliers.

• Uncertainty ahead:

  • The 800G → 1.6T → 3.2T transition may require multiple combinations (a mix of 200G and lower-speed lanes).

  • Revolutionary new modulator technologies (e.g., thin-film materials, non-silicon platforms) are still far off and may take 5–8 years to gradually mature.

5. A warning about industry hype

• “Don’t believe in so-called paradigm shifts”: the speaker argued that most material vendors’ narratives contain a marketing element.

• Practical criteria: as long as module size and bandwidth meet requirements, the market will choose the cheapest, most reliable solution rather than chasing “revolutionary” new technology.

Conclusion

fibeReality’s talk wasn’t a product showcase but a sober look at the state of the industry:

1. OCS’s value is questionable: although backed by Google and CloudLight, real-world operational problems may outweigh its advantages.

2. The intent behind CPO: driven more by business motives (NVIDIA’s dominance strategy) than by genuine power needs.

3. Silicon photonics bottleneck: 200G is close to the limit, and hybrid technologies (InP, DFB, SiPh) may coexist going forward.

4. The modulator problem: raising frequency is hard, low-speed multi-lane designs may be a transitional solution, but revolutionary material technologies are still years away.

5. A pragmatic market: final decisions hinge on cost, reliability, and supply chain, not technology hype.

This talk reminds us that, in the race to 1.6T and 3.2T, the industry needs to distinguish real technical breakthroughs from marketing talk and stay calm and pragmatic.