Nvidia is preparing a major push into United States telecommunications infrastructure by acquiring unused dark fiber to build a private, high-capacity network that could support its expanding artificial intelligence business, according to research reports from Needham and Wolfe Research.
The plan would mark a significant shift for Nvidia, which is best known as the dominant supplier of graphics processing units used in AI training and inference. Instead of relying only on telecom carriers and cloud partners to move data between computing sites, the company is seeking greater control over the physical layer that connects data centers, customers and AI workloads.
Needham and Wolfe Research estimate Nvidia could spend between $5 billion and $10 billion over the next three years on the network buildout. Once completed, the system is expected to reach total bandwidth capacity of about 7.6 petabits per second, a level that would place it among the most powerful private data transmission networks in the country.
The initiative comes as AI computing demand is forcing technology companies to rethink not only chips and servers, but also the underlying networks that carry massive volumes of data. For Nvidia, the move suggests a broader strategy: become less dependent on third-party infrastructure and more directly involved in delivering end-to-end AI computing services.
A move from chips to infrastructure
Nvidia’s dark fiber project would push the company beyond its traditional role as a semiconductor supplier and closer to the operating model of a full-stack AI infrastructure provider. Dark fiber refers to fiber-optic cable that has already been laid but is not currently active. Companies can either lease links from telecom providers or acquire the fiber directly and install their own equipment to “light” the network.
By purchasing dark fiber rather than simply leasing bandwidth, Nvidia would gain exclusive control over a large portion of the physical network. That control could allow the company to choose its own optical equipment, network design, routing architecture and expansion schedule.
This is a heavier, more capital-intensive model than the one typically used by software or chip companies. It is closer to the approach taken by large telecom firms and hyperscale cloud operators that own or control key infrastructure assets over long periods. The benefit is exclusivity and performance control. The trade-off is higher upfront cost, longer deployment timelines and greater operational responsibility.
For Nvidia, the timing is important. The company’s AI chips are central to the current boom in generative AI, large language models and high-performance computing. But the next stage of competition may depend on how efficiently computing resources can be distributed, connected and delivered to customers. As AI models grow larger, the bottleneck is no longer limited to processing power. Network capacity is becoming a core constraint.
Why dark fiber matters
Owning dark fiber gives Nvidia a level of control that leasing bandwidth cannot provide. A leased connection gives a company access to capacity, but the underlying network remains managed by the carrier. The carrier determines parts of the route, service structure, maintenance process and long-term upgrade path.
Direct ownership or long-term control of dark fiber changes that equation. Nvidia could build a private network optimized specifically for AI workloads, with low-latency links between data centers, dedicated paths for large training jobs and customized routing between key compute hubs.
That capability could be especially valuable as AI traffic becomes more demanding. Omdia has estimated that roughly 90% of machine learning data remains inside the main host facility. This reflects the fact that many AI training systems are still concentrated inside single data centers or tightly linked campuses. But if AI workloads increasingly need to move across regions, states or multiple facilities, existing telecom routes may not be sufficient.
Moving training jobs, model updates and inference data across wider geographic areas could require new ultra-high-strand fiber corridors between major computing hubs. In that environment, direct access to dark fiber becomes a strategic asset rather than a simple connectivity expense.
Cisco executive Rakesh Chopra has said that scaling advanced AI models can create traffic demands far beyond standard server connections, estimating that some new loads may generate hundreds of times more traffic than traditional links. That kind of growth would require deeper changes in routing, switching and fiber deployment.
Nvidia’s plan appears designed to address that problem early. If the company can build a private transmission network around its AI platform, it could reduce dependence on public internet routes and outside carrier capacity while improving service consistency for customers running large AI workloads.
The scale of the planned network
The reported target of 7.6 petabits per second is enormous. One petabit equals one million gigabits, meaning the proposed network would have capacity measured in millions of gigabits per second. Reports have also described the target in more precise terms of about 7.68 million gigabits per second.
Such capacity would be comparable to thousands of current backbone fiber routes. It could support high-speed connectivity between many large data centers, depending on how the system is designed, where the fiber is located and what optical equipment Nvidia deploys.
Engineers could use dense wavelength division multiplexing, or DWDM, to reach ultra-high capacity across fiber strands. DWDM works by splitting a single glass fiber into multiple wavelengths, or channels, allowing many streams of data to travel through the same strand at the same time. A network using 96 channels at 800 gigabits per second per channel could produce extremely high throughput across each route.
This technology is already used in major telecom and cloud networks, but Nvidia’s entry would be notable because the company is not a traditional telecom operator. Its main advantage is not a long history in fiber management, but its central position in AI hardware, accelerator systems, networking chips and data center architecture.
If successful, Nvidia could combine computing, storage, networking and physical transport into a more integrated offering. That would give the company more control over the performance of AI services from chip to customer.
Competition is changing
Nvidia’s move also reflects a changing competitive landscape in AI hardware. While the company remains the leading supplier of GPUs for AI, large cloud service providers have been developing or adopting custom chips known as ASICs, or application-specific integrated circuits.
Companies such as Broadcom and Marvell have worked on custom processors and connectivity systems for major cloud customers. These chips are designed for specific workloads and can reduce reliance on general-purpose GPUs in some cases. Large cloud platforms have also developed their own in-house AI accelerators, giving them more control over cost, supply and performance.
By building a private transmission network, Nvidia could strengthen its relationship with customers even if some of them reduce direct dependence on its GPUs. Instead of selling only chips, Nvidia could deliver access to computing power through its own managed infrastructure. That model would allow customers to use Nvidia-powered systems without necessarily owning the hardware or relying entirely on a third-party cloud platform.
This is a major strategic distinction. Nvidia has traditionally benefited from selling hardware to cloud providers such as Amazon Web Services, Microsoft and Google. Those companies then resell cloud computing services to businesses and developers. A direct-delivery model could change how revenue is shared across the AI value chain by reducing the role of intermediaries in some services.
Such a shift would not necessarily mean Nvidia replaces the major cloud platforms. The largest cloud companies remain essential partners and customers. But it could give Nvidia more leverage, more optionality and a stronger ability to serve smaller customers that lack direct access to massive AI infrastructure.
Potential impact on cloud and telecom markets
If Nvidia builds a large private fiber network, the effects could extend beyond AI computing. Smaller cloud providers and enterprise customers may seek alternatives to the largest hyperscale platforms. A Nvidia-controlled network could potentially offer connectivity between data centers, GPU clusters and AI service locations.
That possibility could be attractive to companies that want access to advanced AI infrastructure but do not want to depend entirely on a single hyperscale provider. It could also support new forms of distributed cloud computing, where workloads move across multiple facilities based on cost, latency, availability and data requirements.
For telecom infrastructure companies such as Zayo and Crown Castle, Nvidia’s entry into the dark fiber market could change asset valuations. Dormant or underused fiber routes may become more valuable if AI companies begin competing for control of physical network assets. Large-scale purchases by a major technology company could also encourage consolidation among fiber owners.
Telecom companies have long treated dark fiber as a strategic asset, but AI demand is changing the economics. If data movement becomes as important as data processing, control of fiber corridors could attract more attention from cloud operators, chip companies, data center owners and private infrastructure funds.
The first-quarter size of the broader server switching market has been estimated at $15.4 billion, underscoring how much capital is now flowing into the systems that move data inside and between computing environments. The networking layer is no longer a secondary part of the AI buildout. It is becoming a major spending category in its own right.
Execution risks remain high
The ambition of Nvidia’s plan is significant, but execution will not be simple. Typical dark fiber deployment and activation projects can take 18 to 36 months, even for experienced operators. Larger national network projects can require five to seven years, depending on route availability, permitting, equipment supply, labor, interconnection agreements and customer needs.
Reports suggest Nvidia wants to complete the project within three years. That would be an aggressive schedule for a company with limited direct experience operating large optical transport networks.
Building private fiber capacity requires more than buying cables. Nvidia would need to manage optical transport systems, network operations centers, route redundancy, repair crews, power access, security, regulatory compliance and service-level reliability. It would also need to integrate the transmission layer with its computing services so that customers see a seamless experience rather than a complicated set of technical systems.
Last-mile connectivity could be another challenge. A national or regional backbone may connect large data centers, but customers often need reliable access from their own facilities or local cloud zones. Unless Nvidia partners with carriers or develops additional access arrangements, the final connection to customers could remain dependent on existing telecom providers.
There is also a balance-sheet question. Nvidia has the financial capacity to pursue large infrastructure projects, but a $5 billion to $10 billion program would still represent a major commitment. Telecom-style assets usually take time to generate returns. They also require maintenance and upgrades over many years.
A broader full-stack strategy
The dark fiber effort fits into a larger pattern of Nvidia expanding beyond standalone chip sales. The company has moved deeper into AI systems, software platforms, networking products and partnerships across the global hardware supply chain.
Nvidia has invested in heterogeneous computing technologies, including areas associated with companies such as Groq. It has also built partnerships with OpenAI and worked closely with memory and hardware suppliers across Asia. Together, these steps point toward a more complete AI infrastructure strategy.
The company’s long-term goal appears to be broader control over the AI computing stack. That includes the chips used to process data, the systems that connect those chips inside data centers, the software that manages workloads, the memory and storage ecosystem that feeds models, and now potentially the physical transmission network connecting facilities.
This approach could make Nvidia more resilient as AI demand evolves. If customers want chips, Nvidia can sell chips. If they want cloud-style access to computing power, Nvidia can support that model. If network capacity becomes a bottleneck, Nvidia could use its own fiber to improve performance and reduce reliance on third parties.
Implications for digital asset networks
The expansion of private fiber networks also has implications for digital asset markets and public blockchain systems, though the link is indirect. Public ledger networks depend on broad, open internet connectivity among nodes, validators, exchanges, wallets and trading systems. If more high-speed capacity shifts into private corporate networks, some market participants may watch closely for any effect on public network performance during periods of heavy activity.
Digital asset traders have become more focused on infrastructure reliability, especially when congestion, latency or network outages affect execution. A private cable system controlled by a major technology company would not automatically threaten public blockchain systems, but it could sharpen the debate over centralized infrastructure and open access.
Peer-to-peer compute networks and decentralized physical infrastructure projects may use Nvidia’s move as evidence that physical connectivity is becoming strategically important. However, traders should distinguish between long-term infrastructure themes and short-term market claims. A private AI fiber network would primarily be designed for data center and enterprise traffic, not for public blockchain routing.
A new phase of AI infrastructure competition
Nvidia’s reported dark fiber plan shows how the AI race is expanding into physical infrastructure. The first phase centered on chips. The next phase is increasingly about data centers, power supplies, cooling systems, networking gear and fiber routes.
For Nvidia, owning or controlling dark fiber could help protect margins, improve service reliability and create new ways to deliver AI computing directly to customers. It could also reduce dependence on cloud intermediaries, even while Nvidia continues to work closely with those same platforms.
The project carries clear risks, including cost, execution complexity and the challenge of operating telecom-grade infrastructure. But the strategic logic is clear: as AI workloads become larger and more distributed, the companies that control both computing power and data movement may hold a stronger position.
If completed on the reported schedule, Nvidia’s private network would be one of the clearest signs yet that AI infrastructure is no longer confined to chips and servers. It is extending into the ground, through fiber corridors, and across the physical systems that determine how fast intelligence can move.
Want deeper insight into AI infrastructure trends? Explore Toobit’s Academy article on crypto infrastructure pillars transforming digital networks.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
