AI infrastructure pivots to power, cooling, and speed as Vestberg joins Digipower X
The former Verizon and Ericsson chief Hans Vestberg has been recruited as a senior adviser to Digipower X at a moment when AI growth is gated less by chips and more by electricity, cooling, and build speed.
From telco networks to AI compute infrastructure
Vestberg brings rare end-to-end experience spanning national network rollouts, spectrum strategy, fiber densification, and mobile edge compute from his years running Ericsson and later Verizon, where he presided over the first commercial 5G launch in 2018 and pushed a network-first operating model.
That playbook is relevant to AI data centers shifting toward distributed architectures, where latency, availability, and resiliency resemble telco-grade design choices rather than traditional enterprise IT.
Power, cooling, and time-to-capacity now constrain AI growth
Digipower X positions itself as a vertically integrated AI infrastructure operator combining Tier III-certified modular data centers with owned and controlled energy assets to compress deployment cycles.
The company cites more than 200 MW currently online across a combined-cycle plant and three additional operational sites, development pathways for up to 1.5 GW over the next three years, and a letter of intent tied to a 1.3 GW power plant in West Virginia that is being evaluated as a long-term AI campus anchor, with additional scale targeted in North Carolina.
Its AI-Ready Modular Solution (ARMS) aims to deliver Tier III modular capacity in roughly 180 days, emphasizing redundancy, energy optimization, and liquid-cooling readiness for high-density AI clusters.
What AI’s power shift means for telecom, cloud, and enterprises
Vestberg’s involvement underscores a convergence of telecom-hardened engineering with cloud-scale AI, pushing infrastructure closer to end users and industrial sites.
Distributed AI inference will pull compute to the edge
As training consolidates in hyperscale regions, inference is fragmenting toward regional and metro locations to hit tight latency targets, keep data local, and control costs.
For operators and neutral hosts, this is an opportunity to repurpose central offices, aggregation hubs, and fiber-fed sites into AI-ready edge facilities with liquid cooling, higher rack densities, and Tier III-level uptime, integrated with 5G standalone, MEC, and private networks.
Enterprises in sectors like automotive, healthcare, and logistics will prioritize proximity to plants and campuses, favoring modular builds that can land power at 50–100 kW per rack and support direct-to-chip or immersion cooling without lengthy substation lead times.
Cloud on-ramps and interconnect will define AI edge competitiveness
Hyperscalers including AWS, Microsoft Azure, and Google Cloud increasingly require fast-turn capacity expansions, dedicated interconnect, and predictable power envelopes for GPU clusters.
Sites that pair behind-the-meter generation, short interconnection paths, and dense fiber routes will win placement for cloud edge zones, managed inference services, and telco-integrated MECs.
Standards and metrics matter: Uptime Institute Tier III for redundancy, robust power usage effectiveness (PUE) targets, water usage visibility, and credible renewable energy strategies, including PPAs, RECs, and grid services participation.
Digipower X’s energy-first, modular AI data center model
The company’s strategy prioritizes controlled power supply and fast deployment over traditional real estate footprints.
Energy-first site selection for firm, fast power
By operating a combined-cycle plant and securing rights to additional large-scale capacity, Digipower X is attempting to de-risk interconnection queues and transformer bottlenecks that are delaying competitors by years.
An energy-first approach enables flexible load management, potential demand response revenue, and the option to blend grid, on-site generation, and storage for cost and resiliency.
Modular builds with Tier III resilience and liquid cooling
ARMS packages standardized power and cooling blocks, N+1 topologies, and liquid-cooling support into repeatable modules that can be snapped to demand, slicing months from greenfield timelines.
For buyers, the attraction is predictable delivery, audited uptime characteristics, and the ability to scale GPU clusters without re-architecting the facility each increment.
Risks and open questions for gigawatt-scale AI
Execution at gigawatt scale is capital intensive and faces policy and supply-chain friction.
Permitting, grid, and supply chain constraints for AI capacity
Utility interconnections, environmental approvals, and water rights remain gating items even with on-site generation, while long-lead equipment—HV transformers, switchgear, and liquid-cooling components—can stretch to 18–36 months.
Evolving EPA rules, local siting opposition, and transmission congestion in PJM and neighboring regions could affect timelines and capacity factors.
Capital structure and demand durability for AI infrastructure
Gigawatt programs require multi-year financing, contracted offtake, and disciplined energy risk management; exposure to spot power prices can erode margins during heat waves or grid stress.
Buyers must scrutinize customer mix, term lengths, and the operator’s track record, especially for companies diversifying from prior digital asset activities into AI compute.
What leaders should do now to scale AI infrastructure
Telecoms, enterprises, and investors should treat AI power and proximity as first-order design choices, not afterthoughts.
For telecom operators: retrofit edge sites for AI
Inventory central offices and metro aggregation sites for AI-ready retrofits; pre-negotiate liquid-cooling upgrades and higher-density power trains; and align with modular providers for 6–12 month edge capacity adds.
Bundle dark fiber, slicing-ready 5G standalone, and cross-connects to create turnkey AI edge zones for hyperscalers and enterprises.
For enterprises and AI builders: co-design power and proximity
Co-design siting with energy strategy: secure PPAs or behind-the-meter options, model total cost of inference (power, cooling, network), and insist on Tier III or better resilience for mission-critical workloads.
Pilot modular deployments near plants or campuses to validate latency and data sovereignty while preserving a path to burst to hyperscale regions.
For investors and policymakers: back firm power and modular playbooks
Prioritize projects with line of sight to firm power, credible interconnects, and repeatable modular playbooks; support grid upgrades, transformer manufacturing, and permitting reform that unlock AI capacity without compromising reliability.
Vestberg’s appointment signals that AI infrastructure is becoming a telecom-grade, energy-centric discipline, and the winners will be those that execute like network builders, not just data center landlords.
