NVIDIA's announcement of the GB200 NVL72 system at Computex 2024 set off a new wave of demand forecasting in the optical transceiver industry, with implications that will shape the market through 2026 and beyond.
The GB200 NVL72 is a rack-scale AI computing system featuring:
• 72 Blackwell GPUs interconnected via NVLink, requiring 800G optical connectivity for each NVLink connection.
• An estimated 5,000+ optical transceivers per NVL72 rack, depending on the specific configuration.
• Projected deployments of thousands of NVL72 racks across hyperscale customers in 2025–2026.
The optical module implications are staggering. Industry analysts estimated that NVIDIA's AI infrastructure alone could consume 10–15 million 800G optical transceivers over the 2024–2026 period—a volume that dwarfs the entire historical production of 400G modules.
This unprecedented demand is reshaping the competitive landscape:
• Incumbent OEM module suppliers are allocating the majority of their capacity to hyperscale contracts.
• Mid-market and enterprise operators face longer lead times and higher prices, creating a significant opportunity for compatible module manufacturers.
• New entrants are attempting to scale 800G production, but quality and consistency remain challenges.
For SZVAN, the GB200-driven demand represents our largest addressable opportunity: serving the vast "middle market" of operators who need 800G connectivity but cannot access OEM supply at reasonable terms or timelines.
