Tech Firms Introduced New Optical Standards for AI Centers
New pluggable and co-packaged optical designs aim to address the power and speed requirements of massive AI networks.
Updated on Oct. 12, 2026 in Data Centers
Live Poll
Is choosing more flexible, modular equipment better for long-term infrastructure stability?
Nvidia, Cisco, TE Connectivity, and Amphenol have announced new pluggable optics form factors and co-packaged designs for AI data centers. These technologies represent a strategic shift to manage the extreme power and signal transmission demands of large-scale AI workloads.
Why it matters
AI networks necessitate high-speed data movement at lower costs, driving a move toward more efficient hardware. These new optics options provide operators the flexibility to mix vendors while reducing the thermal and power footprint of intensive computing clusters.
Cisco's 800G Linear Pluggable Optics reduce module power by 50% and overall switch consumption by 30% compared to traditional designs. Additionally, co-packaged optics configurations reduce electrical signal loss from approximately 22 dB to 4 dB.
The players
Nvidia
A designer of graphics processing units and networking hardware that currently leads the market for AI computing infrastructure.
Cisco
A networking giant specializing in switching silicon, routers, and enterprise data center hardware.
TE Connectivity
A manufacturer of high-precision connectors and sensors essential for data center electrical and optical infrastructure.
Amphenol
A provider of high-speed interconnect products and cabling systems used in massive data center deployments.
The details
Linear Pluggable Optics (LPO) function by removing signal-processing electronics from inside the module, which directly reduces heat and electricity usage. Alternatively, co-packaged optics (CPO) place the optical engine—the component that translates data—immediately next to the switch chip instead of using separate plug-in modules. These methods allow for more efficient data handling by minimizing the distance electrical signals must travel.
Timeline
March 18, 2025: Nvidia announced Spectrum-X and Quantum-X Photonics switches.
2025: Projected 800G-and-faster transceiver shipments to exceed 24 million units.
Late 2025 – Early 2026: Expected availability of Quantum-X Photonics switches.
February 10, 2026: Cisco unveiled Silicon One G300 switching silicon.
Second half of 2026: Expected availability of Spectrum-X Photonics switches.
The Tech Race
The introduction of these optics components directly supports the roadmap for high-bandwidth chips like Cisco’s 102.4 Tbps Silicon One G300. By optimizing signal integrity and power efficiency, these designs compete for dominance in the infrastructure required to scale next-generation AI clusters.
Data center operators can expect these new modules to lower operating electricity costs starting in late 2025. The shift allows for more flexible hardware procurement, though adoption requires matching these optics with compatible switches like the forthcoming Spectrum-X and Quantum-X series.
The takeaway
These optical modules are critical for scaling AI capacity without exceeding the power limits of current facilities. Watch for the mid-2026 commercial release of Spectrum-X Photonics switches as a primary indicator of whether these efficiencies materialize in large-scale deployments.
Further reading
For broader trends in infrastructure, read more in our Data Centers section.
Live Poll
Is choosing more flexible, modular equipment better for long-term infrastructure stability?
