Jensen Huang Says AI Could Need 100GW of Distributed Compute in Telecom Networks, Pointing to Verizon/AT&T-Style Buildouts and Partner Moves

NVIDIA CEO Jensen Huang has argued that the next wave of artificial intelligence infrastructure will not be confined to hyperscale data centers. In remarks cited by market media, Huang outlined an opportunity he pegged at about 100,000 telecom network sites and roughly 100 gigawatts (100GW) of distributed data center capacity, describing telecom networks as a potential platform for where AI compute could be deployed.

The framing is significant because telecom companies already manage large, geographically distributed networks with power and cooling capabilities at many locations. Huang’s comments suggest a scenario where AI workload placement could increasingly reflect latency, location, and connectivity needs, pushing compute closer to end users rather than routing everything to centralized facilities.

Market coverage around the comments also noted that major players in wireless and telecom equipment are already taking steps that align with that idea. The cited write-up references T-Mobile US, and Nokia, as examples of companies “already making moves,” though it does not detail what those moves specifically are in the available excerpt.

For NVIDIA, telecom-adjacent positioning matters because it expands the addressable market for its data center platform and AI software ecosystem. Rather than viewing new demand only through the lens of large enterprise and hyperscale campuses, distributed deployments could broaden the set of environments where NVIDIA’s GPUs and related infrastructure are targeted.

AI’s power draw and cooling requirements have been recurring constraints across the industry. Huang’s estimate of “100 gigawatts” indicates he believes telecom networks can provide a meaningful portion of the physical footprint needed to scale compute, potentially reducing the mismatch between where AI demand originates and where compute is built.

NVIDIA has increasingly emphasized the idea of AI as an infrastructure layer, with performance and software workflows built to support deployments across a wide range of environments. While the telecom-specific estimate comes from commentary reported by market media, it reinforces the company’s broader strategy of linking AI scale-out to enterprise and industrial rollouts, not just factory-style data center expansions.

Still, the public details in the cited report are limited. The excerpt does not include the context of when Huang made the remarks, the specific business deals or technical plans tied to any named operator, or any quantified timeline for delivering the “100GW” figure. It also does not clarify whether the estimate refers to an industry-wide potential addressable demand or NVIDIA’s expected share of that demand.

What to watch next is whether NVIDIA, telecom operators, or telecom equipment suppliers provide more concrete disclosures about pilots, network-to-edge deployment designs, or capacity plans that connect telecom site buildouts to AI compute rollouts. Additional specificity would help determine whether “100,000 sites” and “100GW” are near-term engineering targets or longer-cycle estimates for how telecom networks could evolve into distributed computing platforms.