NSF Awards $5.1 Million to Northeastern Team to Develop Adaptive Wireless Infrastructure for AI Data Centers

VINES award unites Northeastern, Florida International U, and Teradio with Dell, NVIDIA, Lockheed, Keysight to demo sub-THz wireless links in AI data centers.

BURLINGTON, MA, UNITED STATES, November 3, 2026 /EINPresswire.com/ — The National Science Foundation has awarded Northeastern University $5.1 million to address one of artificial intelligence’s most consequential and least-discussed infrastructure challenges: the millions of cables that are choking the world’s largest AI data centers. The three-year Verticals-enabling Intelligent Network Systems (VINES) Track 2 award will fund the development and real-world demonstration of sub-terahertz (sub-THz) wireless links capable of carrying hundreds of gigabits per second — inside operating AI data centers where traditional cabling is rapidly reaching its physical limits.

AI clusters now scale to hundreds of thousands, and in some cases more than a million, GPUs housed in dense racks. A single AI data center can require millions of individual cable runs, driving up capital and operating costs, obstructing airflow, and complicating the power and cooling systems that sustain high-performance infrastructure. As GPU cluster density increases with each hardware generation, the cable plant does not scale gracefully. It becomes the bottleneck, and increasingly, the failure surface.

The Northeastern-led research program will design, build, and validate a sub-THz wireless network that augments and, in targeted applications, replaces today’s structured cabling for both out-of-band management networks and primary data fabric traffic. The project targets integration with standard data center network interfaces, enabling seamless deployment into existing AI data centers infrastructure

Why Sub-Terahertz?
Sub-terahertz wireless offers a rare combination of properties uniquely suited to the AI data center. Narrow, electronically steerable pencil beams provide the spatial isolation and physical-layer security of a wired connection, without the wire. The propagation distances inside a data center, where link distances measure in meters rather than kilometers are ideally suited for Sub-THz. Silicon technology has now matured to the point where sub-THz transceivers can be fabricated in standard semiconductor processes at commercially viable power levels and per-unit cost, a critical enabler that makes this research translatable into a deployable product.

A Multi-Institution Team Built for Translation
The project is designed from the outset for translation from research results into operational deployments. Principal Investigator Dr. Josep Miquel Jornet, COE Distinguished Professor, Associate Dean of Research, and Associate Director of the Institute for Intelligent Networked Systems (INSI) at Northeastern University, leads the program’s sub-THz system architecture, channel modeling, and beamforming research. Northeastern co-investigators Tommaso Melodia and Salvatore D’Oro contribute AI-driven spectrum orchestration, intelligent network control, and network slicing — the software layer that makes sub-THz links dynamically adaptive to the demands of large-scale GPU training workloads.

At Florida International University, co-investigators Dr. Konstantinos Zekios and Dr. Arjuna Madanayake lead the hardware frontier: advanced sub-THz antenna arrays, RF front-end design, and chiplet development using heterogeneous integration techniques that can bring sub-THz capability into the production-scale form factors that hyperscale data center operators require.

Teradio, Inc., a startup commercializing sub-THz radio technology developed within Northeastern’s INSI, provides the project’s path to commercial deployment. Teradio is developing sub-THz wireless systems for the AI data center. The company’s Adaptive Wireless Infrastructure platform delivers three capabilities simultaneously: resilience through a physically isolated out-of-band management plane, efficiency through traffic offload from primary training interconnects, and architectural freedom through software-defined on-demand wireless links provisioning that a fixed cable plant cannot match.

Industry Partners Ground the Research in Operational Reality

VINES Track 2 requires at-scale demonstration of the integrated system in real-world operational environments — making the project’s industry partnerships central to its structure rather than ancillary to it.

Dell Technologies provides operational AI factory environments for pilot deployment, live traffic data to validate real-world system performance, and direct engineering collaboration on integration with production rack infrastructure. NVIDIA contributes technical guidance on GPU cluster networking requirements and the specific management and data traffic patterns generated by large-scale AI training workloads, ensuring the sub-THz system is designed to the actual demands of current and next-generation GPU architectures. Lockheed Martin supports validation and testing in defense-relevant environments where secure, high-bandwidth wireless connectivity is operationally critical, extending the program’s demonstration into a second high-value vertical. Keysight Technologies provides advanced measurement, characterization, and protocol testing equipment for the sub-THz frequency range, ensuring that all performance results are independently validated with rigorous, calibrated instrumentation.

For Project Information please see: https://aifactories.sites.northeastern.edu/
Please visit Teradio at https://www.teradio.co

GREGORY WHELAN
Teradio, Inc.
gwhelan@teradio.co

Media gallery