Comcast launched a field trial that will see the cable giant deploy Nvidia GPUs at its regional facilities to support latency and speed sensitive edge services through what it calls an “AI grid.”

Comcast highlighted a trio of initial trial use cases. These include a personalized advertising agent using an ad-delivery engine from Decart real-time AI video models; a small business concierge agent that uses Personal AI’s small language model (SLM) and a memory platform deployed on Hewlett Packard Enterprise (HPE) ProLiant servers; and low-latency online gaming that builds on Comcast’s ongoing work, including work using Nvidia’s GeForce Now platform.

Comcast noted that initial lab testing “demonstrated strong performance,” with the field trials being used to “validate latency improvements, power and cost efficiencies, resiliency, scalability across Comcast’s footprint, and user experience benefits in a live environment.”

“Distributed AI Grid is the next big opportunity for the telecommunications industry, and Comcast’s nationwide, deeply distributed network is a perfect match for building it,” Ronnie Vasishta, SVP for AI and telecom at Nvidia, noted in a statement. “By bringing intelligent AI inference to the network edge, Comcast can unlock inherent cost efficiencies, while delivering deterministic, low‑latency experiences for customers at massively concurrent scale. This collaboration is powering the next era of hyper personalized experiences that run just milliseconds from users.”

The venture is also eyeing services for “AI-enhanced advertising,” small-business offerings, “premium low-latency gaming tiers,” and other third-party edge compute services.

Comcast’s cable network uses data over cable service interface specification (DOCSIS) full-duplex (FD) 4.0 technology, “smart amplifiers, and intelligent gateways” to serve approximately 65 million locations. Comcast claims this architecture can support “real-time AI inference at scale – something traditional centralized, fiber-only, or wireless networks cannot match.”

Despite that fiber dig, Comcast is deploying fiber in some locations, including a recent deal with Ciena to connect underserved locations.

The latest work comes on the heels of Nvidia this week unveiling its long-awaited language processing unit (LPU) efforts with Groq. The LPU is targeted at ultra-low-latency AI inference workloads, and positioned as an option alongside its GPU line.

“The GPU with its 288-gigabytes for memory, compared to only 500-megabytes of SRAM (static random access memory). The LPU is only one-five-hundredth of the amount of capacity per chip, but the bandwidth is exceptional: 22 terabytes to 150 terabytes per second,” Ian Buck, data center head at Nvidia, explained. “Nvidia is working to combine these two processors to do the decoding operations of a GPU and do it as one.”