Yole Says AI Interconnect Demand Could Push Transceivers to $112B by 2031

Yole Group argues that optical transceivers are no longer just supporting hardware in AI data centers: they are becoming a system-level requirement for scaling AI. Its new forecast puts the optical-transceiver market at $112 billion in 2031, as larger clusters make the connection between accelerators as consequential as the accelerators themselves.

The argument is a change in where the hard problem sits. In a large training cluster, a slow or power-hungry connection can leave compute waiting: latency can stall synchronization, bandwidth can limit utilization, and interconnect power competes with compute power. Yole’s framing is that the bottleneck is moving from compute to communication.

The gap between model growth and compute

Yole estimates that AI models are growing roughly 100 times every two years, while compute improves about 3.3 times over the same period. It says that mismatch would require about 70 times more interconnects every two years—a pace it says the optics-module industry is not currently equipped to sustain.

Those are Yole estimates rather than observed industry outcomes, but they explain the scale of its market call. The firm forecasts optical-transceiver revenue will rise from about $10 billion in 2021 to $112 billion in 2031, or roughly 35% annual compound growth. It also projects hyperscaler capital expenditures of $670 billion in 2026 and $5.3 trillion across 2026 through 2031.

Not one networking market, and not one replacement cycle

Yole divides the interconnect problem into four domains: Scale-In, Scale-Up, Scale-Out and Scale-Across. The distinction matters because each domain has a different technology entry point, incumbent technology and commercialization timeline. A single prediction that “optics replaces copper” would flatten those differences.

Copper, in particular, is still being extended through megawatt-class rack engineering. Meanwhile, co-packaged optics is entering commercial deployment only narrowly, in proprietary switching platforms. Yole says this is currently suited to a limited set of vertically integrated players able to take on manufacturing risk.

The largest CPO opportunity is inside the cluster

The most aggressive number in Yole’s analysis is for co-packaged optics, or CPO. It forecasts CPO optical-engine revenue of $112.1 billion in 2031, up from $0.6 billion in 2026. Scale-Up applications are expected to account for about 94% of that 2031 market.

That creates the report’s central tension. The projected revenue opportunity is enormous, yet the technology associated with it is not described as a broad, mature replacement for today’s links. The near-term question is whether networking and photonic components can become available at the required scale, timing and cost as AI infrastructure grows.