Nvidia's "Underground" Campaign: Why $10 Billion in Dark Fiber Is a Deeper Moat Than Chips
- Core Thesis: Nvidia plans to invest between $5 billion and $10 billion to acquire dark fiber across the U.S. and build a carrier-grade network. This aims to transform the company from a chip supplier into a full-stack AI infrastructure operator, constructing an "ultimate insurance policy" against the rise of ASIC chips by controlling the physical transport layer, thereby extending competition from chip performance to direct computing power delivery.
- Key Elements:
- The project will cost between $5 billion and $10 billion. Upon completion within three years, the total bandwidth will reach 7.6 Petabits per second, surpassing enterprise-level demands and targeting control of infrastructure at the telecom operator level.
- Acquiring dark fiber grants Nvidia exclusive usage rights to the lines, allowing it to independently define network designs and equipment choices, creating a strategic path characterized by heavy assets, long cycles, and high exclusivity.
- This deployment is seen as a strategy against the erosion of the GPU market by ASIC chips customized for cloud giants by companies like Broadcom and Marvell. Even if customers choose ASICs at the computing layer, Nvidia can retain influence and economic benefits through the network delivery link.
- Building its own network shifts Nvidia's relationship with cloud giants like AWS from a supplier dynamic to a competitive one in specific scenarios, potentially redistributing the platform premium profits previously captured by cloud service providers.
- For traditional telecom operators like Zayo and Crown Castle, Nvidia's large-scale acquisitions may drive a revaluation of dark fiber assets, accelerating tech giants' penetration into the infrastructure layer.
- Operational risks are significant: Nvidia lacks experience in optical transmission equipment, network operations, and last-mile connectivity, presenting major execution challenges for efficient commercialization.
Original Author: Zhao Ying
Original Source: Wall Street CN
Nvidia is quietly advancing a project across the United States that is unrelated to chips yet potentially more strategically defensive than any single GPU: acquiring dark fiber nationwide that has been laid but never activated, to build its own carrier-grade network infrastructure.
On Tuesday, according to disclosures from Needham and Wolfe Research, the project is expected to cost between $5 billion and $10 billion over the next three years. Upon completion, its total bandwidth potential could reach 7.6 Petabits per second—a scale that surpasses enterprise-level demand and targets infrastructure control at the level of a telecommunications carrier.
Unlike directly purchasing bandwidth, acquiring dark fiber grants Nvidia full usage rights to the fiber optic lines, allowing it to independently define everything from optical transmission equipment selection to network topology design. This is a path that is more asset-heavy, longer-cycle, but also significantly more exclusive.
Wolfe Research characterized this move in its report as Nvidia's "ultimate insurance policy against the rise of ASICs." Against the backdrop of major clients' in-house chips eroding GPU market share, Nvidia is attempting to shift the competition from "whose chip is faster" to "who can deliver computing power directly to the customer"—by controlling the physical transmission layer, creating a second moat beyond hardware competition.
For investors, this news reflects a fundamental shift in Nvidia's strategic positioning: evolving from an upstream component supplier for cloud service providers into a computing infrastructure operator that directly reaches end customers. This implies that Nvidia's long-standing, delicate symbiotic relationship with AWS, Azure, and GCP may enter a new phase of competition and cooperation.
7.6 Petabits per Second: An Underestimated Number
The total bandwidth ceiling of Nvidia's dark fiber network—7.6 Petabits per second—needs to be understood within an appropriate frame of reference. The current single-fiber transmission capacity of the global internet backbone typically ranges from tens to hundreds of Tbps, while Nvidia's planned scale equates to the aggregated capacity of thousands of backbone-grade fibers.
A dark fiber network can connect multiple data centers, forming a massive cluster. Analysts believe Nvidia could lease this network to Neoclouds, thereby narrowing the gap with hyperscale data center operators, many of which possess extensive fiber optic networks.
The activation and commercialization cycle for dark fiber typically ranges from 18 to 36 months. From fiber acquisition to deploying optical transmission equipment, and establishing last-mile connections to various data centers and enterprise nodes, Nvidia needs to complete in three years a journey that would take traditional telecom operators five to seven years. The aggressive nature of this execution pace itself sends a signal to the market.
"Ultimate Insurance": How Dark Fiber Hedges ASIC Risk
Understanding the strategic urgency of dark fiber requires examining the changing competitive landscape Nvidia faces. Companies like Broadcom and Marvell are customizing AI-specific chips (ASICs) for cloud giants such as Google and Amazon, gradually eroding GPU market share in specific scenarios like inference. When major customers have the option of in-house chips, Nvidia's bargaining power in the computing power delivery chain becomes less secure.
The dark fiber deployment is precisely a response to this structural risk—by building its own network to directly reach end customers, Nvidia can transform computing power from being "resold via cloud service providers" to "end-to-end self-delivery." The deeper meaning of Wolfe Research using the term "ultimate insurance" lies in this: Even if future customers choose ASICs over Nvidia GPUs at the computing layer, as long as the computing power is delivered via Nvidia's network infrastructure, Nvidia retains a say and economic interest in the delivery layer.
This logic is highly consistent with a series of strategic moves by Nvidia in recent years: a $20 billion acquisition of Groq's technology and team establishing the GPU+LPU heterogeneous computing paradigm; a $30 billion investment in OpenAI securing the largest demand-side entity for model training. The dark fiber network fills the last—and most difficult to replicate—piece of the entire chain from computing power production to delivery: the physical transmission layer.
Dual Pressure from Cloud Giants and Telecom Operators
For AWS, Azure, and GCP, a direct sales model for GPU-as-a-Service presents a thorny prospect: their most important upstream supplier is becoming a direct competitor in specific scenarios. Previously, when enterprise customers rented GPU instances on cloud platforms, cloud service providers extracted a significant platform premium. In the future, if Nvidia delivers directly via its own network, this profit margin will be redistributed between supply and demand.
For traditional telecom operators, a new buyer valued at over $5 trillion aggressively acquiring long-haul dark fiber could trigger a revaluation of related assets. Dominant players in the dark fiber market, such as Zayo and Crown Castle, face not just cost increases from a competitive buyer, but an acceleration of tech giants' penetration into the telecommunications infrastructure layer—a trend that has already extended from data centers to transmission networks.
However, it must be noted that the commercial operation of a dark fiber network is no easy task. Optical transmission equipment, network operations and maintenance teams, and last-mile connections to various data center enterprises—Nvidia has nearly zero experience in these areas. Even after the network is built, operating it efficiently and achieving commercialization poses significant execution risks.
The dark fiber acquisition is not an isolated venture for Nvidia but a component of its transformation from a chip company into a full-stack AI infrastructure operator. In just the past year, Nvidia's deployments have spanned computing architecture (Groq), customer ecosystem (OpenAI), supply chain (SK hynix HBM cooperation), and geopolitical layout (Japan robotics and automotive industry cooperation)—and the dark fiber network represents the part of this puzzle focused on physical control.
