Nvidia, Cisco and partners introduce new pluggable optics format for AI data centers

Nvidia, Cisco, TE Connectivity and Amphenol have introduced a new pluggable optics form factor built for AI data centers.

In data centers, the format of a component decides who can build it, who can buy it, and how quickly it can be swapped out.

What the four companies are building

Pluggable optics are the small modules that slot into network switches. They turn electrical signals into light so data can travel across fiber between servers.

The new form factor arrives as each partner pushes its own optical roadmap. Cisco has been expanding its pluggable lineup with 1.6T OSFP modules and 800G Linear Pluggable Optics, known as LPO.

Cisco’s 800G LPO solutions cut power consumption by approximately 50% for the optical modules and around 30% overall for the switch.

LPO strips out some of the signal-processing electronics that normally sit inside each module, meaning less heat and a smaller electricity bill.

Cisco also unveiled its Silicon One G300 switching silicon on February 10, 2026. The chip can switch data at 102.4 Tbps.

TE Connectivity and Amphenol supply high-density connectors and interconnect hardware. Both showed components for co-packaged and linear pluggable architectures at OFC 2026, the optical networking industry’s main trade event.

Nvidia’s parallel bet on co-packaged optics

Nvidia’s involvement in a pluggable standard is notable because the company has also been championing the opposite approach. On March 18, 2025, at its GTC conference, Nvidia announced Spectrum-X Photonics and Quantum-X Photonics switches using co-packaged optics, or CPO.

CPO moves the optical engine right next to the switch chip instead of putting it in a separate plug-in module.

Nvidia says the design reduces the number of lasers needed by four times. It also claims 3.5 times better power efficiency, 63 times better signal integrity, and ten times better resiliency than traditional pluggable transceivers.

Deployment is also 1.3 times faster, according to Nvidia.

CPO cuts electrical signal loss from roughly 22 dB to roughly 4 dB, meaning far less of the signal degrades before it becomes light.

Quantum-X switches can deliver up to 115 Tb/s in configurations of 144 ports at 800 Gb/s.

Spectrum-X configurations go further, handling up to 409.6 Tb/s or 512 ports at 800 Gb/s. Quantum-X Photonics switches are expected to become available between late 2025 and early 2026, with Spectrum-X Photonics expected in the second half of 2026.

Why pluggables are not going away

CPO is best suited to the highest-radix switches at the core of a network with the most ports. Pluggables, including LPO, offer flexibility: operators can mix vendors, replace a failed module without touching the switch, and upgrade in pieces.

With co-packaged optics, a fault in the optical engine is a much bigger repair job than pulling out a module and plugging in a new one.

What this means for the AI networking market

Shipments of 800G-and-faster transceivers are projected to exceed 24 million units in 2025, driven by AI cluster buildouts.

A 50% cut in module power, multiplied across millions of units, is the kind of saving that shows up in a data center’s power budget.

For Nvidia, playing on both the CPO and pluggable sides means it wins regardless of which architecture a customer picks for a given layer of the network.

For Cisco, a common pluggable format protects its switching and optics business when customers keep buying modular, interoperable gear rather than locking into fully integrated designs.

TE Connectivity and Amphenol are positioned as suppliers to the whole market, as both CPO and advanced pluggables require dense, high-performance connectors.