Co-packaged optics has long promised to transform datacenter connectivity, but deployment-ready products only arrived in 2025. Now a third path is emerging. LightMatter, the photonics startup backing a silicon-photonics interconnect platform, has unveiled a middle ground between conventional pluggables and fully integrated designs. The Passage L20, unveiled ahead of the Optical Fiber Conference this week, sits somewhere between the two extremes and may find a ready market among large cloud operators.
The tension at the heart of datacenter optics is straightforward: fitting optical engines directly into processor packages cuts power consumption and latency but introduces a catastrophic risk. If the photonics fail after being permanently bonded to a compute die, the entire chip becomes scrap. Hyperscalers have long expressed caution about this prospect, even as they acknowledge the potential benefits of co-packaged optics.
LightMatter's near-package approach keeps the photonics off the main die but close enough to avoid the electrical inefficiencies of a conventional pluggable module. The Passage L20 connects via copper to the ASIC, then breaks the signal out into up to 32 fibres, each capable of bidirectional signalling at 200 Gbps. The technical advantage lies in that bidirectionality. Each fibre carries traffic in both directions simultaneously, not one direction per fibre as conventional modules require. In real terms, this translates to cutting the fibre count in half compared to pluggable-based designs.
The numbers illustrate why this matters at scale. Sixteen L20 units can replace 512 200 Gbps pluggables in a 102.4 Tbps switch, whilst also reducing power consumption. LightMatter says each L20 draws 30 watts; conventional pluggable modules at higher bandwidths often exceed 10 watts each. At a Meta or Google scale, those watt-hour savings multiply across thousands of switches and millions of fibre-kilometres.
CEO Nick Harris acknowledges that near-package optics is likely a temporary waypoint on the road to full co-packaging. "I don't think that it's going to be a super-long roadmap for near package optics, because, of course, CPO is coming, and we think that's like a 2028 high-volume ramp," he said. Yet the market window may be substantial. Pluggable modules are expected to dominate through 2027 even as major vendors roll out CPO switches. That leaves a critical 18-24 month window in which hyperscalers are unwilling to commit fully to co-packaging but need a step up from conventional transceivers.
The economics of fibre supply reinforce this timing. Meta recently signed a deal worth up to $6 billion with Corning for optical fibre in its Louisiana data centre alone. Cutting fibre requirements in half is not a trivial cost saving at that scale. For operators calculating return on infrastructure investment over a decade, the L20 offers a safer incremental improvement than gambling on co-packaging maturity.
That said, the bet assumes rapid adoption in a narrow window. Once co-packaged optics mature and hyperscalers gain confidence in their reliability, customers will likely migrate to fully integrated designs. LightMatter has also announced higher-capacity CPO products of its own, positioning itself to serve the future market. The L20 is essentially a way to build revenue and field experience whilst hedging against the possibility that co-packaging's 2028 timeline slips.
The broader context is one of genuine urgency. Moving photons consumes between five and 20 times lower power than moving electrons, and as power becomes a bigger issue, fibre gets closer to compute. The question is not whether optical interconnects replace copper; it is how fast, at what cost, and with how much risk.