The promise and the peril of data center optics. First, the promise as shown by the image below. Demand is exploding. Module shipments rose from 5.1 million in 2023 to 23 million in 2024 says Andrew Schmitt of Cignal AI. Speeds are rising too with 400G peaking in 2025, 800G peaking in 2026, and 3.2T shipping in 'meaningful' numbers in 2029 (Initial shipments around 2027.)
Now the peril. Namely, power, with 1.6T modules (coming to market now) hitting 30W. The alternatives for reducing power are all in flux, Schmitt adds. LPO (linear packaged optics) is interesting, but was late to 100G and the market is going to 200G. TRO (transmit retimed optics) shows promise, but doesn't save as much power. NPO (near packaged optics) waits for a champion. CPO (co-packaged optics), meanwhile is "inevitable, not imminent...and insanely hard," says Schmitt.
So what's the solution? Call it good engineering housekeeping. Shifting to 3nm, reducing host channel losses and other fine-tuning can reduce the power consumption of a 1.6T TRO module to under 16 watts says Xi Wang at Marvell Technology. More advanced silicon photonics and possibly technologies like co-packaged copper can continue the efficiency gains. When we get to 3.2T, pluggable modules will still be the mainstream solution.
This is from a webinar taking place now. Link for rewatching soon.
