A sharp debate has ignited over the timeline for Co-Packaged Optics (CPO), a critical technology for next-generation AI data centers.
The conflict began when research firm SemiAnalysis distributed a note to institutional clients suggesting that the CPO rollout could be delayed to 2028–2029. This report, citing concerns over manufacturing yields, repairability, and total cost of ownership (TCO), sent shares of optical communications companies tumbling. The core of the concern is that packaging the optical components directly with the switch ASIC (chip) makes field repairs much more complex than with current pluggable modules.
However, NVIDIA quickly refuted this claim. On the same day, at the Computex conference, NVIDIA's Senior VP of Networking, Gilad Shainer, stated, "We are ready to start shipping, and we are ramping up CPO in the second half of 2026." This direct contradiction created two competing narratives in the market: one of delay and another of an imminent, albeit initial, launch.
So, why is CPO so important? Its main advantage is power efficiency. By placing optics closer to the chip, CPO can reduce the power consumption of a high-speed port by up to 70%—from 30 watts to just 9 watts for an 800G port, according to NVIDIA. For massive AI factories that consume megawatts of power, this translates into enormous savings on electricity and cooling, directly improving the TCO.
Recent evidence seems to favor NVIDIA's timeline for an initial launch. First, reports from early June, such as one by the Taipei Times, indicated that NVIDIA had already begun shipping its CPO switches, co-developed with TSMC, to select partners. Second, this was corroborated when cloud provider Lambda shared a blog post unboxing one of these first CPO switches. Third, NVIDIA has been consistently messaging a 2026 CPO commercialization target for over a year across blogs, webinars, and earnings calls.
Ultimately, both narratives may hold a piece of the truth. The evidence strongly suggests that a limited ramp-up for select partners, likely focusing on switch-side (scale-out) applications, will indeed begin in late 2026. However, the manufacturing and serviceability challenges highlighted by SemiAnalysis are real hurdles to overcome for widespread, mass-market adoption. Therefore, it's plausible that while early adopters deploy CPO in 2026, the broader market may rely on interim technologies like LPO/NPO (Low-Power/Near-Packaged Optics) until CPO technology matures and becomes more cost-effective and reliable, likely in 2027 or beyond.
- Co-Packaged Optics (CPO): A technology that integrates optical components directly onto the same package as a silicon chip (like a switch or processor) to improve power efficiency and performance.
- Pluggable Optics: The current standard where optical transceivers are modules that can be easily plugged into or removed from switches and other network devices, allowing for easy replacement and upgrades.
- Total Cost of Ownership (TCO): A financial estimate that includes the purchase price of an asset plus all direct and indirect costs of operating it over its lifespan, such as power, cooling, and maintenance.