Verizon Converts Central Offices Into Edge AI Hubs While Building Billion-Dollar Fiber Networks

The Pivot From Pipes to Processing

Verizon has committed over $1 billion to provision dark fiber routes connecting data center facilities operated by Google, marking the telecom operator's most aggressive move yet into AI infrastructure services. The agreement, announced during a Q2 2026 earnings call, forms the cornerstone of what Verizon now brands "AI Connect," a business unit designed to capture revenue from the compute boom reshaping enterprise technology spending.

At DailyTechWire, we've tracked how incumbents across Asia and North America are racing to repurpose legacy telecom assets for AI workloads. Verizon's strategy stands out not only for the deal size but for the surgical transformation underway inside its own facilities: the company is physically removing copper wiring from central offices and converting those spaces into compact data centers optimized for inference tasks that demand ultra-low latency.

Dan Schulman, who leads Verizon Communications, framed the shift in stark terms. The question is no longer whether telcos can compete in cloud infrastructure, he suggested, but whether they can solve the connectivity puzzle that hyperscalers now face. "How do you connect data center to data center to assure that you can optimize to maximize the ever-exploding need for compute?" he said. Verizon's answer involves both long-haul fiber linking massive training clusters and edge nodes placed closer to end-users, where inference happens in milliseconds rather than round-trips measured in tens or hundreds of milliseconds.

Dark Fiber Economics and the Hyperscaler Bet

Dark fiber, unlit strands of optical cable that customers can activate with their own transmission equipment, has become a strategic asset as AI training runs and model-serving traffic overwhelm public internet backbones. Verizon controls thousands of route-miles accumulated over decades of network buildout, much of it underutilized or idle. By leasing dedicated fiber paths to Google, Verizon monetizes infrastructure that would otherwise sit dormant while locking in predictable revenue streams that span multiple years.

The deal's structure also signals a broader shift in how hyperscalers procure connectivity. Rather than relying on shared bandwidth or third-party transit, companies running planetary-scale AI workloads increasingly prefer exclusive fiber routes that guarantee bandwidth, minimize jitter, and isolate traffic from the open internet. Verizon's existing footprint, particularly in metro areas and between major colocation hubs, positions it to serve that demand without the capital expense of greenfield fiber construction.

Schulman indicated that additional agreements are in advanced stages, collectively worth "multiple billions of dollars in revenue over the next several years." He did not name other customers, but the pipeline suggests Verizon is negotiating with more than one hyperscaler, possibly including Microsoft, Amazon Web Services, or Oracle, all of which are expanding AI data center capacity in North America and globally.

Edge Inference and the Central Office Transformation

While the Google fiber deal addresses data center-to-data center connectivity, Verizon's parallel effort to convert central offices into edge AI facilities targets a different bottleneck: latency-sensitive applications that cannot tolerate the round-trip delay to a distant cloud region. Schulman cited robotics, remote surgery, and autonomous vehicle control as workloads that require inference to happen within single-digit milliseconds, a threshold achievable only when compute sits geographically close to sensors, cameras, and actuators.

Central offices, the neighborhood-scale facilities that traditionally house telephone switches and broadband equipment, are distributed across urban and suburban footprints. That geographic spread makes them natural candidates for edge deployment. Verizon's approach involves gutting legacy copper plant, freeing up rack space, power, and cooling capacity to install GPU servers and inference accelerators.

The economics hinge on repurposing sunk real estate and power infrastructure rather than building new edge sites from scratch. Verizon already owns the land, holds the utility interconnections, and operates the fiber backhaul. By swapping out aging voice hardware for inference compute, the company aims to extract new revenue from assets that were depreciating under the weight of declining landline subscriptions.

The Competitive Landscape and Regional Implications

Verizon's AI Connect initiative arrives as telecom operators worldwide grapple with stagnant or declining consumer revenue and search for growth in enterprise and infrastructure services. In Asia, operators such as Singtel, SK Telecom, and NTT have each announced edge AI pilots or partnerships with cloud providers, though few have disclosed contracts at the billion-dollar scale. Verizon's willingness to commit capital and transform physical plant at speed may force competitors to accelerate their own edge strategies or risk ceding enterprise AI revenue to a handful of early movers.

The dark fiber model also has implications for regional connectivity markets. In Southeast Asia, submarine cable consortia and neutral fiber providers have traditionally dominated inter-city and international routes. If incumbents in markets like Indonesia, Thailand, or the Philippines begin to monetize dormant fiber for AI workloads, the balance of power in wholesale connectivity could shift back toward legacy telcos, particularly those with dense metro fiber rings and data center adjacency.

Verizon's announcement notably omits detail on power provisioning, a constraint that has stalled data center projects across multiple geographies. Central offices were not designed for the power density of GPU clusters, and retrofitting electrical infrastructure can require utility upgrades, permitting, and months of construction. The feasibility of Verizon's edge buildout will depend on whether local substations can supply the kilowatts per rack that inference accelerators demand, a variable that differs sharply by market and vintage of facility.

Revenue Recognition and the Multiyear Horizon

Schulman's guidance of "multiple billions of dollars in revenue over the next several years" suggests contracts with staggered recognition, typical of infrastructure-as-a-service deals where customers commit to minimum spend over three to five years. For Verizon, the appeal lies in predictable cash flow that offsets the capital intensity of fiber deployment and central office conversion. For hyperscalers, the value is guaranteed capacity at a time when colocation space, power, and fiber are all supply-constrained.

The timing also aligns with the broader AI infrastructure investment cycle. Training clusters for frontier models continue to scale, with some facilities now exceeding 100,000 GPUs, while inference workloads are proliferating as enterprises embed AI into customer-facing applications, analytics pipelines, and operational systems. Verizon is betting that both trends, training connectivity and edge inference, will sustain elevated spending through the end of the decade.

Whether that bet pays off depends on execution risk, competitive response, and the pace at which AI workloads actually migrate to the edge. Remote surgery and autonomous driving remain niche or experimental in most markets; if mainstream inference stays centralized in hyperscaler regions, Verizon's edge buildout may struggle to fill capacity. Conversely, if latency-sensitive use cases proliferate faster than expected, the company's head start in converting real estate and securing fiber deals could deliver outsize returns.

What Comes Next

Verizon has committed to disclose additional AI-related agreements before the close of 2026. The scale and structure of those deals will clarify whether the company is pursuing a handful of large hyperscaler partnerships or a broader portfolio that includes enterprise customers, sovereign cloud operators, or AI-native startups. The answer will also reveal how much of the AI infrastructure stack telcos can realistically capture, and whether their role remains confined to connectivity or expands into hosting, orchestration, or even model serving.

For now, the Google deal and central office conversion represent the clearest signal yet that legacy telecom operators see AI infrastructure as a path back to growth. The question is whether Verizon can execute the physical transformation, manage the power and cooling challenges, and deliver the ultra-low latency that edge inference promises, all while competitors and hyperscalers weigh their own build-versus-partner decisions. The next twelve months will test whether AI Connect is a durable revenue engine or a high-cost experiment in a market that may yet consolidate around a few vertically integrated players.