Supply chain issues with optical transceivers threaten to stall the development of AI-era networks, according to a new McKinsey report.

The opportunities in networking optics report forecasts optical-transceiver shortages, not GPU developments, will throttle 800 Gbps and 1.6 Tbps data center networks through to the end of the decade.

Production of 800 Gbps transceivers is projected to run 40 to 60 percent below demand until 2027, and shortfalls of 30 to 40 percent are likely for 1.6 Tbps devices through 2029.

The laser squeeze comes from hyperscaler demand. According to McKinsey, hyperscalers will shift around 87 percent of their back-end optics to 800 Gbps and greater by 2029, with 1.6 Tbps transceivers accounting for more than 40 percent of this demand.

The front-end metro optics market is also undergoing rapid expansion fueled by the adoption of coherent zero dispersion-shifted range (ZR/ZR+) transceivers with high data rates. The analyst firm suggests that about half of the metro segment currently uses 400-Gbps ZR/ZR+ transceivers.

As AI training and inference require low-latency interconnects, hyperscalers and data center operators are increasingly deploying high-speed coherent transceivers and investing in high-speed fiber backbones to connect data center clusters.

As such, McKinsey forecasts the market could transition quickly, with 800 Gbps ZR/ZR+ transceivers and above capturing about 70 percent of market share by 2029.

As the front-end metro optics market continues to expand, new Original Equipment Manufacturer (OEM) players are entering the field, led by laser suppliers’ vertical integration.

The crunch is seeing OEMs buying or partnering with laser fabs, while China-based suppliers have increased their share of the back-end pluggable market rapidly, growing by about 20 percentage points between 2017 and 2023 to roughly 60 percent.

In reaction to tariff and geopolitical background noise, assembly is migrating to Southeast Asia and parts of Europe.

McKinsey’s report advises the networking optics industry to “work quickly and collaboratively” to address the laser supply shortfall - otherwise, it risks becoming a bottleneck in the AI-fueled expansion of today’s data centers.