Argentum AI, Kyndryl Sign $630M GPU Data Center Deal

Argentum, Kyndryl Sign $630M NVIDIA GPU Deal Argentum, Kyndryl Sign $630M NVIDIA GPU Deal

Argentum AI and Kyndryl have signed a 20-year, $630 million data center collaboration to deploy NVIDIA Blackwell Ultra B300 GPU systems at Kyndryl’s Boulder, Colorado, facility. The agreement covers an initial 10 megawatts of capacity in a Tier III data center environment, with deployment already underway. The partnership gives Argentum access to infrastructure designed for high-density GPU workloads and aims to expand its ability to supply enterprise AI inference capacity. The deal also highlights how specialized GPU cloud providers are partnering with established infrastructure operators to address the growing demand for computing resources needed to run large-scale AI applications.

Argentum and Kyndryl Expand GPU Infrastructure for AI Inference

The agreement gives Argentum access to a data center facility configured for the power demands of current-generation GPU systems. For Kyndryl, it creates an opportunity to use existing infrastructure and its enterprise technology services capabilities to support the expanding AI compute market.

The companies have not disclosed the number of NVIDIA B300 systems or GPUs planned for the initial deployment, the expected completion date, or how the $630 million contract value is divided between facility capacity, power, equipment and services. Those details will be important for assessing the project’s eventual computing capacity and commercial economics.

NVIDIA B300 Systems Target Demanding Inference Workloads

The deployment centers on NVIDIA’s B300 systems, part of its Blackwell Ultra GPU platform. Blackwell Ultra is designed for demanding AI workloads, including large-scale inference, reasoning models and applications that require substantial memory and computing throughput.

NVIDIA’s published specifications for its HGX B300 platform list eight Blackwell Ultra GPUs, with up to 2.1 terabytes of GPU memory and 144 petaflops of FP4 tensor performance in sparse configurations. These figures describe NVIDIA’s specific HGX B300 configuration and should not be interpreted as confirmed specifications for Argentum’s deployment, whose exact server configuration has not been disclosed.

Inference infrastructure is becoming increasingly important as enterprises move generative AI applications from experimentation into production. Training large models requires substantial computing resources, but inference capacity is needed continuously to generate responses, process requests and support AI-powered applications.

The workloads can include conversational AI, enterprise copilots, coding assistants, document processing and specialized AI services. As usage grows, providers need infrastructure that can deliver predictable performance, support high utilization and accommodate increasingly demanding models.

Argentum’s GPU cloud model is focused on providing dedicated bare-metal GPU capacity to AI model builders, inference providers and enterprise customers. The Kyndryl agreement adds a long-term infrastructure arrangement to support that business.

Why the Tier III Facility Matters

The Boulder deployment is described by the companies as operating in a fully Tier III environment. Under Uptime Institute’s classification, Tier III infrastructure is concurrently maintainable: planned maintenance or replacement of individual capacity components and distribution paths can take place without disrupting IT operations. It does not provide the same fault-tolerance characteristics as Tier IV infrastructure.

That distinction matters for AI services supporting enterprise applications. GPU clusters are expensive to operate, and interruptions can affect customer workloads, service availability and computing economics.

High-density GPU deployments also place greater demands on electrical distribution and cooling systems than conventional enterprise server environments. Infrastructure operators must accommodate substantial power loads while maintaining stable operating conditions for compute equipment.

Kyndryl’s existing facility gives Argentum access to infrastructure intended for this class of deployment. The companies have not provided detailed information about the site’s cooling architecture, power redundancy configuration or the specific certification status of the deployed environment.

A Long-Term Infrastructure Partnership

The 20-year term suggests that both companies are approaching the collaboration as a long-term infrastructure arrangement rather than a short-term capacity reservation.

For Argentum, securing data center capacity through an established infrastructure provider could reduce the time and capital required to develop an entirely new facility. It also gives the company an operating environment in which to deploy GPU systems and serve customers seeking dedicated AI compute.

For Kyndryl, the agreement extends its role in the AI infrastructure ecosystem. Its services business includes designing, building, managing and modernizing enterprise IT environments. Working with specialized GPU cloud providers creates another route to apply those capabilities as demand for AI infrastructure expands.

The deal also reflects a wider shift in the data center market: specialist compute providers are increasingly working with established operators that already have facilities, power infrastructure and operational experience.

AI Compute Growth Brings Energy and Capacity Challenges

The partnership arrives amid a rapid expansion in global data center investment. The International Energy Agency’s 2026 analysis estimates that electricity consumption from data centers could rise from approximately 485 terawatt-hours in 2025 to 950 terawatt-hours by 2030. AI-focused facilities are expected to account for a growing share of that demand.

This growth makes access to suitable power infrastructure a strategic consideration for GPU cloud providers. Securing GPUs alone is not enough; operators also need data center capacity, electrical equipment, cooling systems and reliable operations.

Argentum and Kyndryl have not disclosed the project’s power sourcing, energy-efficiency targets or water-management approach. Those factors will matter as the deployment moves from installation into sustained operations.

The companies also intend to broaden their collaboration as Argentum’s infrastructure requirements grow. Whether that leads to additional capacity at Boulder or deployments at other facilities remains unclear.

For enterprise AI buyers, the agreement signals another effort to expand the infrastructure available for high-performance inference. Its eventual significance will depend on deployment progress, customer access to the capacity and the reliability and economics of the resulting GPU cloud services.

Market Landscape

The AI infrastructure market is expanding beyond hyperscale cloud providers as specialized GPU cloud companies secure dedicated computing capacity through partnerships with data center operators.

Key market developments include:

  • GPU cloud expansion: Specialized providers are building dedicated infrastructure for AI model developers and inference workloads.
  • Blackwell Ultra adoption: NVIDIA’s B300 systems target increasingly demanding inference and reasoning applications.
  • Data center partnerships: GPU cloud companies are using existing facilities to accelerate deployment rather than developing every site independently.
  • Power and cooling constraints: High-density AI systems require specialized electrical and thermal infrastructure.
  • Long-term capacity agreements: Multi-year contracts provide infrastructure providers and GPU operators with a framework for planning investment and customer capacity.

Argentum and Kyndryl’s agreement combines a specialist GPU cloud business with an established enterprise infrastructure operator. The initial 10 MW deployment is a meaningful infrastructure commitment, but its commercial impact will depend on how quickly the systems become operational and how effectively Argentum converts capacity into customer workloads.

Top Insights

  • Argentum AI and Kyndryl signed a 20-year, $630 million agreement for an initial 10 MW of GPU infrastructure at Boulder.
  • The deployment uses NVIDIA B300 Blackwell Ultra systems designed for demanding AI inference and enterprise workloads.
  • The companies say deployment is underway, but have not disclosed GPU quantities, completion timelines or detailed system configurations.
  • The Tier III environment is designed for concurrent maintenance, supporting operational continuity during planned infrastructure maintenance.
  • The partnership reflects growing demand for dedicated GPU cloud capacity as enterprises move generative AI applications into production.

Power Tomorrow’s Intelligence — Build It with TechEdgeAI

Grow Your
Brand Visibility

Looking to publish a press release, guest article, interview or podcast? Connect with us.

GET FEATURED
Subscribe

Sign up today for exclusive insights and updates.

Newsletter Signup