Co-packaged optics technology could become critical for data center network scalability as bandwidth demands surge.

When Cisco took to the stage at the Optical Fiber Conference in 2023 to demonstrate co-packaged optics on a 3.2TB switch, the technology did not receive a rapturous reception. "I would say the industry response at that time ranged from indifference or a yawn to fairly negative," says Bill Gartner, SVP & GM of optical systems and optics at Cisco.

The skepticism was not rooted in doubt about CPO's potential benefits. Rather, concerns centered on vendor lock-in—the fear that CPO would force customers into an all-Cisco or all-Broadcom world. Gartner explained that traditional pluggable optics give customers negotiating leverage. "There's a switch, which makes up the bulk of the router and its costs, and then there are ports on the face plate into which you plug these pluggable optics. But a customer has a choice of which pluggable optic they want to use, and there are many providers of pluggable optics. That gives them negotiating leverage and the ability to make sure that they've got supply chain diversity, which gives them a lot of benefit."

The indifference that characterized early reactions has since vanished. The industry now appears ready to embrace CPO, despite the vendor lock-in issue remaining unresolved.

Co-packaged optics emerged as a concept around 2020, when data centers faced mounting power and bandwidth bottlenecks. Roy Chua, founder and industry analyst at AvidThink, explains the progression: "In the old days, your electrical connections were copper. Your data center networks were connected by Ethernet, and everything was good. Then, as speeds increased, copper became unable to handle the volume of traffic, so the industry turned to fiber optics." Fiber optics required lasers to convert electrical signals into photons for transport over glass fiber, then back to electrical signals on arrival—a process demanding significant energy.

Today's data centers rely on pluggable optical transceivers, known as digital signal processors, which connect to switches. However, DSPs are expensive and power-hungry. Co-packaged optics addresses this by integrating optical and electrical components—optical engines and switching ASICs—on a single packaged substrate. "What co-packaged optics does is move the conversion onto the actual silicon itself. It moves those lasers onto the silicon, eliminating the need for any external units," Chua explains. "By putting those lasers inside the silicon, you are able to save power. By co-packaging these elements, it allows for more consolidation while lowering costs and energy use."

Manufacturing remains challenging. As Chua notes, "You have to manufacture everything – your wave guides and your modulators – all together." Early concerns about cost and reliability arose, with questions about replacing failed lasers in a monolithic structure. The industry developed hot-swappable laser sources known as External Laser Small Form-Factor Pluggables to address this, though Gartner acknowledges the solution represents only a partial fix.

According to AvidThink's Data Center Networking in 2026 report, CPO is driving a structural shift in optical interconnect architecture by removing the need for DSPs from optical transceiver modules. Competing approaches include linear pluggable optics (currently the most popular alternative) and near-packaged optics, alongside Arista's eXtra-dense pluggable optics, which Chua describes as more of a packaging form factor than a pure technology.

CPO transitioned from perpetual promise to production deployment in 2026, driven largely by announcements from Nvidia CEO Jensen Huang at GTC. Huang indicated that Nvidia's Spectrum-X CPO switches would enter production in the second half of 2026, and announced that the company's 2028 Feynman architecture would introduce NVLink 8 CPO, bringing silicon photonics directly to GPU-to-GPU scale-up interconnects for the first time. This represents a significant milestone—CPO has traditionally been applied to scale-out switch-to-switch connections, not latency-critical scale-up domains.

"Until Jensen said CPO, no one cared about CPO. No one had any idea what CPO was, and now I get asked that question all the time," Chua says. He credits Huang with accelerating the industry timeline. "Without Jensen sort of stoking the fire, I think it would have been the 2027-2028 conversation, and Jensen basically made it the 2025-2026 conversation."

Broadcom has proven particularly aggressive in response to Huang's announcement. Beyond Nvidia's initiatives, the past 18 months have seen significant CPO developments: AMD's January 2025 acquisition of Enosemi; Lightmatter's unveiling of Passage L200, described as the world's first 3D co-packaged optics product; Broadcom's announcement of its Tomahawk 6 CPO variant, Davisson; Marvell's acquisition of Celestial AI; Coherent's demonstration of multi-technology CPO at OFC 2026; and a joint reference design announcement from Ayar Labs and Wiwynn for optically connected, rack-scale AI systems.

In March 2026, Ciena, Coherent, Marvell, Molex, Samtec, and TeraHop announced the formation of the Open CPX MSA to develop specifications for optical engines "required to enable a broad ecosystem of interoperable co-packaged and near-package interconnect solutions."

While these announcements suggest substantial demand, Gartner emphasizes that much early CPO work was vendor-driven rather than customer-driven. After Cisco and Broadcom's poor reception in 2023, Nvidia's CPO announcement catalyzed genuine industry momentum. "Nvidia announced that they were going to deploy CPO and, quite honestly, stimulated a lot of excitement in the industry," Gartner says.