Cisco sets 1.6T optics rollout for late 2024 to curb GPU downtime

The immediate move to 1.6T connectivity signifies an architectural shift where networking reliability is now a primary bottleneck for AI throughput rather than just a support service. For the streaming and cloud ecosystem, this highlights a growing economic inversion: the cost and power profile of the interconnects are becoming more volatile than the compute silicon they serve. As massive training clusters expand beyond single-facility power limits, the ability to maintain 1.6T speeds across distances will dictate which infrastructure providers can support the next wave of generative AI models. Watch for initial volume shipment reports in Q4 2024 to gauge hyperscale adoption speed.

The push toward 1.6T networking follows a period of significant growth in the preceding 800G market. Per LightCounting in Q2 2025, 800G transceiver sales saw a 10% quarter-over-quarter rebound, indicating that hyperscalers are aggressively filling capacity even as 1.6T qualifications begin. Market analyst 650 Group reported in March 2026 that AI-related networking investments are poised to dominate data center spending, with the broader Ethernet switch market projected to approach $200 billion annually by the mid-2030s. This surge is largely due to the high ratio of optical components needed to synchronize parameters across hundreds of thousands of GPUs.

While pluggable optics remain the industry standard, architectural alternatives are gaining traction to solve thermal issues. Broadcom announced in early 2026 that its Tomahawk 6 silicon, supporting 64 ports of 1.6Tbps, would officially integrate co-packaged optics (CPO). Per Semianalysis in January 2026, CPO is becoming essential specifically for 'scale-up' networking inside the rack, where copper links like NVLink are reaching physical distance and signal integrity limits. Meanwhile, Meta has reported significant trials in CPO reliability, achieving 50 million hours without a link flap, a crucial metric given that even a single failure can derail a large-scale training job.

Supply chain constraints remain a headwind for the 1.6T ramp. Per industrial reporting in late 2025, the production of electro-absorption modulated lasers (EMLs) and continuous-wave laser chips has emerged as a major bottleneck. To mitigate these risks, major hyperscalers including Google and Amazon have moved toward long-term purchase agreements to secure module capacity through 2026. This supply-side pressure coincides with the finalization of the IEEE 802.3dj standard for 200G-per-lane signaling, which is expected to underpin the standardized 1.6T ecosystem by mid-2026.