Foxconn has reportedly sent the entire supply of its Co-Packaged Optics (CPO) switch racks to NVIDIA ahead of schedule, a move that highlights just how fast the AI infrastructure race is accelerating. Industry chatter suggests there may be little to nothing left for Foxconn to physically showcase at this year’s major Taipei computer exhibition, because even units typically reserved for display have allegedly been shipped out.
The bigger story is what this says about demand. NVIDIA is pushing hard to advance co-packaged optics well before its next major platform era, with CPO positioned as a key answer to mounting data center networking limits. As AI clusters scale up, traditional interconnect approaches increasingly run into bandwidth, power, and reach constraints. CPO aims to reduce those bottlenecks by bringing optical connectivity closer to the switching silicon, improving efficiency and enabling denser, faster scale-out networking.
According to reports from Taiwanese supply chain sources, Foxconn originally projected shipments of more than 10,000 CPO rack units for 2026. That forecast has now reportedly been raised to around 50,000 units across 2026 and 2027—about five times the initial expectation. If accurate, that kind of ramp suggests NVIDIA is preparing for a significant deployment wave, not just small pilots.
Foxconn’s role in this is especially important because it’s said to be NVIDIA’s primary, and possibly sole, supplier for these all-optical CPO switch cabinets and racks. The manufacturing is reportedly taking place at Foxconn’s Vietnam facility. The same reports claim that prototype shipments began in early 2026, with mass production and broader shipments expected to follow later—yet the prototypes and available cabinets have already been pulled forward and delivered to NVIDIA, creating an unusually tight supply situation.
Financially, the shift toward co-packaged optics could be meaningful for Foxconn. The company is expected to see better pricing and stronger margins from CPO-related products compared with more conventional hardware builds, with some estimates suggesting CPO could contribute a notable share of revenue as deployments expand.
Looking ahead, the timeline being discussed points to initial CPO adoption emerging in 2027, followed by much broader rollout in 2028 as next-generation AI platforms reach larger-scale data center use. With competitors also investing heavily in next-gen interconnect technologies, CPO is shaping up to be one of the critical battlegrounds for AI infrastructure performance, efficiency, and scalability over the next several years.
If the “everything shipped out” claims are true, Foxconn’s empty shelves aren’t a marketing problem—they’re a signal: NVIDIA’s appetite for next-generation optical networking gear may be bigger, and sooner, than the industry expected.
