HomeTech NewsData Centres & EnergyArgentum AI and Kyndryl Sign $630 Million, 20-Year Boulder Deal

Argentum AI and Kyndryl Sign $630 Million, 20-Year Boulder Deal

Argentum AI and Kyndryl have announced a 20-year, $630 million data-centre collaboration for an initial 10 megawatts of capacity at Kyndryl’s Boulder, Colorado, facility, to host NVIDIA B300 systems for AI inference. The Kyndryl newsroom and a matching Argentum release on PR Newswire, both dated 17 November 2026, use the same wording.

The $630 million is the total value of the 20-year agreement, not an annual payment. The companies have not named the customer that will sit in the hall.

What the two companies say they have agreed

Both releases describe Argentum as a premium GPU cloud provider and Kyndryl as an enterprise technology-services company. Argentum, they say, “is deploying NVIDIA B300 systems at the site in a fully Tier III environment,” and “Deployment is underway.” The deal is meant to give Argentum access to a facility built for the power density of current-generation GPU systems, and to give Argentum’s customers “a faster path to high-performance inference capacity.”

Andrew Sobko, Argentum’s founder and chief executive, called the partnership with Kyndryl “an important milestone for Argentum AI.” Jamie Rutledge, president of Kyndryl U.S., said the agreement “gives us a unique opportunity to put Kyndryl capacity to work in an innovative way.” Both companies “intend to broaden their collaboration to support Argentum’s growth and evolving mission-critical enterprise AI infrastructure needs.” Those are the companies’ statements.

The releases also say the facility “welcomes a fast-growing, enterprise-grade AI inference company, whose technology powers some of the world’s most widely used AI applications, as its flagship anchor tenant.” That tenant is not identified. Nothing in either official text names a model provider, a consumer app, or a contract at the tenant layer.

Argentum describes its business as originating, financing and operating GPU clusters, and as delivering dedicated bare-metal GPU capacity to model builders, inference providers and enterprises. Kyndryl is presented as a provider of mission-critical infrastructure services. The releases do not say how many B300 systems will be installed, what the 10 megawatts will draw as a peak load, or when the first tenant workload will go live, beyond the statement that deployment is already under way.

Where this sits among GPU-capacity deals

This is a long lease of hall space and power for a specific GPU generation, not a loan against already deployed chips. That distinguishes it from USD.AI’s $106 million loan for NexGen Cloud’s GPUs in Finland and from IREN’s $363 million Wingspire financing against deployed AI GPUs. Those posts are about debt secured on hardware. This one is about a 20-year collaboration at an existing Kyndryl site.

The closest product comparison is other B300 cluster announcements, including Scaleway’s plan for an NVIDIA B300 cluster in Copenhagen. Scaleway was describing its own cloud regions. Argentum is describing capacity it will operate in a Kyndryl hall, with an unnamed anchor tenant.

A 20-year, $630 million headline is not a named customer, and 10 MW is an opening figure

The firm facts from the two official releases are that Argentum and Kyndryl have announced a 20-year collaboration valued at $630 million, that the capacity is an initial 10 megawatts at Kyndryl’s Boulder facility, that Argentum is deploying NVIDIA B300 systems there in what it calls a fully Tier III environment, and that the companies say deployment is underway. The $630 million should be read as a total over two decades. Dividing it into an implied annual rent, or treating it as a one-off capital outlay, would go beyond what the releases say.

The unnamed tenant is a genuine gap, not a detail to be filled in from elsewhere. The companies chose a broad description — a “fast-growing, enterprise-grade AI inference company” whose technology, they say, powers widely used applications — and stopped there. That claim is theirs. It is not a substitute for a name, a workload, or a utilisation figure. Until a tenant is identified, or a later official update supplies one, the deal should be read as a hall-and-hardware agreement with a still-undisclosed customer.

Ten megawatts is also described as an “initial” allocation. Both companies say they intend to broaden the collaboration. That is a statement of intent, not a second contracted phase with a published size or date. The releases likewise give no schedule for when the B300 systems will be serving production inference, only that work on the deployment has started.

The useful way to place the news is beside other GPU-capacity financings and cluster announcements, not beside a funding round. This is a commercial collaboration for a defined hall, a defined term and a defined opening wattage. The next official facts to watch are ordinary ones: whether the tenant is named, whether the companies publish a go-live date, and whether the “initial” 10 megawatts is followed by a further contracted block. Until then, the news is a 20-year, $630 million Boulder deal for B300 inference capacity, with the customer still unstated.

Sources

Related stories

Follow @tsnmediahub on X for breaking AI news as it happens.

Follow on X

Share this story

Latest stories

  1. Johns Hopkins Warns of US Transformer Shortages from 2027
    Data Centres & Energy

More in this category

Latest stories

  1. Johns Hopkins Warns of US Transformer Shortages from 2027
    Data Centres & Energy

Explore the TSN Atlas: Every topic we cover, mapped. Click a bubble to dive in.