NVIDIA and Cisco have announced a joint initiative to introduce a new pluggable optics form factor designed in particular to scale the next generation of artificial intelligence data centers and ‘AI factories.’
As enterprise adoption of AI is scaling from experimental pilots to massive multi-trillion-parameter foundation models and autonomous agent networks, the world’s networking infrastructure is hitting a physical limit. Standard data center interconnects can’t handle the bandwidth, power, and thermal density needs of connecting huge GPU clusters. In a common interconnect form factor, NVIDIA and Cisco are helping to solve the physical limits of data center networking: standardizing a new high-density, low-latency, pluggable optical interface to enable AI factories to scale compute capacity at a reasonable power, heat, and infrastructure cost.
“Building the world’s largest AI factories starts with rethinking the smallest of connections,” said Gilad Shainer, senior vice president NVIDIA. “With USFP, NVIDIA and Cisco are building an ecosystem to reinvent pluggable optics — connecting more computing with greater efficiency to power the next generation of AI factories.”
High-Density Optics and AI Factory Interconnects.
Until now, scaling high-performance computing (HPC) and AI clusters has meant using costly, bespoke optical transceivers or loading up on power-hungry copper cables. The size of multi-node GPU clusters has scaled from thousands to hundreds of thousands of accelerators, though, and with it, conventional electrical networking interfaces with serious signal degradation over distance, requiring data center architects to add costly signal boosters and large cooling systems.
This new pluggable optics architecture, co-developed by NVIDIA and Cisco, overcomes these physical limitations with an optimized optical networking design:
Ultra-High Bandwidth Density Combines the latest silicon photonics with small, pluggable form factors to bring ultra-high bandwidth density to every rack. Provide the best possible throughput per rack unit to support dense GPU-to-GPU data pipelines.
Extreme Power Efficiency Gains: Minimizes total optical interconnect power, compared to standard optical transceivers to dramatically minimize energy bill of large AI clusters.
Standardized Interoperability: offers a standard, open ecosystem across the Cisco networking hardware and NVIDIA accelerated computing platforms, removing the vendor lock-in and reducing the complexity of deployment for CSPs and enterprise data center operators.
Also Read: TD SYNNEX Introduces AI Agents and MCP Connector to Streamline PartnerFirst Operations
Strategic Impact on the Data Center Infrastructure and AI Tech Industry
Deploying high-efficiency pluggable optics directly into AI compute infrastructure introduces fundamental structural realignments across the Data Center Infrastructure and Cloud Computing landscape:
1. Overcoming the “Power and Thermal Wall” in Next-Gen Data Centers.
Power has overtaken compute capacity as the primary bottleneck to enterprise AI scaling. Data center operators consistently run up against limits in the municipal power grid and cooling challenges with ultra-dense GPU racks. Standardized, low-power pluggable optics give data center designers an easy way to free up power capacity taking the power away from networking overhead and reinvesting it into compute.
2. Accelerating the Deployment of Distributed “AI Factories”
Seamless, ultra-low-latency fabric connectivity for AI factories Today’s AI factories need to connect thousands of the most powerful, distributed GPUs together at ultra-low-latency so they can all work together as a single, cohesive supercomputer. High-density pluggable optics enable infrastructure teams to break the physical distance barrier of copper cables without latency or thermal compromise. This helps to accelerate the build-out of the ultra-scale AI training and inference facilities globally.
3. Standardizing Optical Networking Across Open Ecosystems
Before the rise of open standards, proprietary optical interconnects fragmented the hardware ecosystem and increased CapEx for enterprise cloud builders. By doing so, the joint NVIDIA and Cisco initiative establish a compelling industry milestone for open, pluggable standards. Data center hardware OEMs and component suppliers will rally behind the new form factor, bringing supply chain scale, reduced component costs and AI-ready infrastructure faster.
General Impact on Business in the Data Center & amp; AI Infrastructure Market.
This hardware breakthrough between NVIDIA and Cisco sets performance and financial benchmarks for enterprise IT, hyperscalers, and infrastructure providers.
CapEx Tuning for Cloud Providers: reduced power consumption and eased cabling contribute to a decreased total cost of ownership (TCO) for hyperscalers and Tier 2 cloud providers, providing a boost to AI Cloud provider operating margins.
Pressure on suppliers of Legacy Networking Equipment Components: suppliers of components and legacy providers of networking infrastructure for existing, high-power optical formats will be quickly bypassed. Procurement through enterprise channels will focus on systems integrated to energy-efficient, AI-optimized optical formats.
Make Enterprise On-Premises AI Supercomputing More Accessible: As pluggable optical modules become a commodity, enterprise IT teams will be able to self-serve on-premises private AI clusters without requiring specialized and complex cooling and power upgrades making high-density compute accessible.
Conclusion
NVIDIA and Cisco’s new pluggable optics form factor to accelerate AI is an important milestone in today’s modern data center. Together, by removing the power, thermal, and bandwidth constraints of the cluster scale at GPU scale, the companies have built the physical infrastructure for the next decade of AI growth. And for the entire technology industry, this explosion validates that the most critical component of artificial intelligence is how quickly data traverses optical links, just as much as how quickly the chips run.