Sources: Nscale is in talks for a second deal with Paradox worth ~$45B to deliver 460MW of computing power at its West Virginia campus using Vera Rubin chips (Bloomberg)

Nscale is proposing to provide Paradox with a very large supply of computing power. The planned capacity is 460 megawatts at Nscale’s campus in West Virginia. The possible deal could be worth about $45 billion. That matters because AI companies need major computing infrastructure to develop and run advanced models.

The key mechanism is a commercial arrangement between an AI cloud provider and an AI model company. Nscale would operate the infrastructure, while Paradox would use the computing capacity. The campus would use Vera Rubin chips, which are designed for demanding AI workloads. The article describes this as a second deal, meaning the companies are already involved in an earlier transaction.

The report describes talks, not a completed agreement. Its wording indicates that the proposed arrangement could become a major expansion of Paradox’s computing resources. It also shows how AI development increasingly depends on large, specialized data-center projects rather than only on individual computers or small server clusters.

An AI cloud computing company provides remote computing infrastructure for AI work. Customers access servers, accelerators, storage, and networking through a data center and pay for capacity or usage. This model matters because AI systems require specialized hardware and large facilities that can be expensive and difficult for one company to build alone.

In this case, Nscale would supply Paradox with computing power from its West Virginia campus. The hardware would include Vera Rubin chips, and Paradox could use the system to train models or operate them for users. The cloud provider manages the physical site, equipment, power systems, and connections, while the customer runs software on the rented infrastructure.

The proposed arrangement shows cloud computing becoming a central business for AI. Nscale would turn a large physical installation into a service that Paradox can access. The talks are not a completed deal, but the reported scale suggests customers are seeking long-term access to enormous dedicated capacity.

A megawatt measures power, or the rate at which electricity is being used. So 460 megawatts describes a huge operating load, not a quantity of stored electricity. For an AI campus, that load would support computing equipment, cooling systems, networking, lighting, backup systems, and other facility operations. It signals infrastructure far beyond a small server room.

The planned West Virginia campus would need large electrical substations, high-capacity connections, cooling equipment, and buildings filled with servers. The chips perform calculations, but the site must continuously supply and remove heat from thousands of machines. The article identifies 460 megawatts as computing capacity, though technically the figure expresses electrical power associated with the installation.

The exact number of buildings, servers, and construction phases is not provided. Still, 460 megawatts implies a major data-center development with regional infrastructure consequences. Its actual size would depend on how much power reaches computing equipment and how much is used by cooling and supporting systems.

Vera Rubin refers to a generation of specialized chips designed for accelerated computing, especially AI workloads. Unlike general-purpose processors, AI accelerators are built to perform many mathematical operations in parallel. The article identifies Vera Rubin chips as the hardware planned for Nscale’s West Virginia campus. It does not provide technical specifications or confirm the final configuration.

AI models rely heavily on repeated calculations involving large arrays of numbers. Specialized chips can perform those operations more efficiently than ordinary processors. They are installed in servers, connected through high-speed networking, and supplied with electricity and cooling. Paradox could use that infrastructure to train models and run them for users.

The reported use of Vera Rubin chips points to the importance of hardware choices in AI infrastructure. Newer accelerators can increase available performance, but the project still requires data centers, power, software, and networking. Because the companies are only in talks, the final chip deployment and schedule remain unsettled.

AI models learn by processing data and repeatedly adjusting vast networks of numerical parameters. That training process requires powerful processors, fast connections, large memory, and substantial electricity. After training, the model still needs computing power to answer users’ requests. The more users and complex tasks it handles, the more capacity Paradox may need.

A project like Nscale’s proposed campus would combine many specialized chips in one coordinated system. Software divides calculations across the machines, while networking moves data between them. Cooling and power equipment keep the system operating. Paradox could use such capacity both to develop models and to provide services based on them.

The article does not specify which Paradox models, workloads, or user volumes would use the proposed capacity. It does show the scale of the infrastructure being discussed: 460 megawatts and a potential value of about $45 billion. The talks suggest that access to computing is a major strategic need for AI companies.

A 460-megawatt campus would be a major new electricity customer. The grid might need stronger transmission lines, substations, generation capacity, and reliability planning to serve it. The facility could also increase demand for water, roads, fiber connections, and construction services. These effects depend on the project’s design, timing, and agreements with local utilities.

The campus could create temporary construction work and permanent roles in data-center operations, electrical systems, cooling, security, and maintenance. Local suppliers and nearby businesses might gain spending from the project. The key mechanism is investment: building and operating a large facility puts money into land, equipment, labor, services, and public infrastructure.

The article provides no estimates for jobs, taxes, water use, or grid upgrades. It only reports talks for a 460-megawatt campus in West Virginia. Therefore, possible benefits and pressures are implications, not confirmed outcomes. The final impact would depend on whether the deal closes and the campus is built as proposed.

Computing begins with electricity flowing into servers. Specialized chips convert that electrical energy into rapid mathematical operations. Memory holds data and model parameters, while networking connects many chips so they can work together. A data center supplies the buildings, racks, cooling, power distribution, security, and connectivity needed to keep the system running.

Software gives the hardware instructions. It loads AI models, divides calculations across chips, moves data between machines, and returns results to users. During training, software repeatedly processes data and updates model parameters. During operation, it accepts a request, calculates a response, and sends that response back through the network.

In the proposed arrangement, Nscale would provide the physical campus and Vera Rubin-based computing capacity, while Paradox would use the service for its AI work. The article does not detail the software stack or facility design. It does establish the overall scale: 460 megawatts at a West Virginia campus, in talks worth about $45 billion.