Nvidia spends $10 billion on dark fiber to build a moat beyond chips

Nvidia's plan to acquire dark fiber across the US and build a carrier-grade network with 7.6 Petabits/second of capacity threatens to reshape the cloud computing market by cutting out the middlemen — Amazon's AWS, Microsoft's Azure and Google Cloud.

"This is Nvidia's ultimate insurance policy against the rise of custom ASICs," Wolfe Research analysts wrote in a report, framing the network build as a hedge against GPU market share erosion from chips designed by Broadcom and Marvell for cloud hyperscalers.

The three-year, $5 billion to $10 billion project targets dark fiber — already-laid but unlit cables — giving Nvidia full control over optical transmission equipment and network topology. At 7.6 Petabits/s, the aggregate bandwidth would rival major telecom carriers, far exceeding what any single enterprise customer requires. Needham and Wolfe Research disclosed the plans Tuesday.

Nvidia shares, trading at roughly 35 times forward earnings, face a structural risk as Amazon, Google and Microsoft develop in-house AI chips for inference workloads. The dark fiber network lets Nvidia deliver computing power directly to end customers, preserving a revenue stream even if those clients shift compute layers away from GPUs.

A 7.6 Petabit Network That Rivals Telecom Carriers

The bandwidth target — 7.6 Petabits/s — needs context. Current global internet backbone single-fiber transmission typically runs in the tens to hundreds of Terabits per second. Nvidia's planned aggregate capacity equals thousands of backbone-grade fibers combined. The network would connect multiple data centers into a single massive cluster, which Nvidia could then lease to Neoclouds — smaller cloud providers that lack the fiber assets of hyperscalers.

Dark fiber acquisition and activation typically takes 18 to 36 months from purchase to commercial operation. Nvidia aims to complete the build in three years, a timeline that compresses what traditional telecom operators achieve in five to seven years. The company has near-zero experience in optical transmission equipment, network operations or last-mile connections to data centers and enterprise sites — all areas where execution risk is high.

The Cloud Hyperscaler Dilemma

For AWS, Azure and Google Cloud, the plan creates an uncomfortable dynamic: their most important hardware supplier is becoming a direct competitor in specific segments. When enterprise clients rent GPU instances on cloud platforms today, the hyperscalers capture a significant platform premium. If Nvidia delivers computing power through its own network, that margin gets redistributed between Nvidia and the end customer.

The dark fiber strategy also extends Nvidia's broader shift from a chip company into a full-stack AI infrastructure provider. Over the past year, Nvidia has invested $2 billion in Groq's technology and team to establish a GPU-plus-LPU computing model, committed $30 billion to OpenAI to lock in the largest model training demand, deepened its HBM supply partnership with SK Hynix, and expanded into robotics and automotive through Japan-based collaborations. The dark fiber network fills the last gap in the chain — physical transport.

For traditional telecom operators, Nvidia's entry as a buyer of long-haul dark fiber could trigger a revaluation of assets held by Zayo and Crown Castle, the dominant players in the US dark fiber market. A $5 trillion buyer entering the space raises acquisition costs and accelerates the trend of technology giants penetrating the telecom infrastructure layer.

But commercial operation of a carrier-grade network is a different challenge from designing GPUs. Nvidia must deploy optical transmission equipment, hire network operations teams and establish last-mile connections — all areas where it has no track record. Even if the network is built on time and budget, converting capacity into recurring revenue at scale remains unproven.

Nvidia shares have gained more than 170% over the past 12 months, reflecting investor confidence in AI infrastructure spending. The dark fiber investment shows management's conviction that the spending cycle will persist — and that controlling the physical layer of compute delivery is worth $10 billion.

This article is for informational purposes only and does not constitute investment advice.