Argentum AI, Kyndryl announce $630M Boulder data center deal
Argentum AI and Kyndryl announced a 20-year, $630 million collaboration on November 17, 2026, deploying NVIDIA B300 systems into an initial 10 megawatts at Kyndryl’s Boulder data center.

Why it matters
- Argentum AI and Kyndryl announced the collaboration on November 17, 2026.
- The 20-year agreement is valued at $630 million and covers an initial 10 megawatts in Boulder.
- Argentum is deploying NVIDIA B300 systems for AI inference at the Tier III facility.
- Boulder’s NVIDIA B300 deployment is already underway.
- Argentum reports 77,000 contracted GPUs, 344 megawatts of contracted power and 13 exclusive data center sites.
By Marcus Bennett
Argentum AI and Kyndryl announced a 20-year, $630 million data center collaboration on November 17, 2026, covering an initial 10 megawatts at Kyndryl’s Boulder, Colorado facility.
Argentum is deploying NVIDIA B300 systems there for AI inference, the execution of trained models to produce outputs. The companies said deployment is already underway in a fully Tier III environment, which they describe as enterprise-grade infrastructure for demanding AI applications at scale.
Under the agreement, Argentum becomes the facility’s flagship anchor tenant, the primary customer under the arrangement, and gains access to Kyndryl’s existing high-density capacity. The 20-year commitment pairs defined infrastructure with Argentum’s financing and GPU cluster-operation role, while giving Kyndryl a long-duration customer for the site.
The arrangement matters because Argentum’s power-first model starts with a site and its electricity before it procures GPUs. Kyndryl contributes an operating facility, and Argentum says the approach can deliver inference capacity faster than a buildout dependent on new construction.
What does the $630 million agreement cover?
The announced scope begins with 10 megawatts at Boulder, where Argentum will install NVIDIA B300 systems for inference workloads. The companies say the existing facility was built for the power density of current-generation GPU systems, aligning the site with Argentum’s hardware selection.
On its official website, Argentum reports 77,000 GPUs contracted for deployment, 344 megawatts of contracted power, and 13 global data center sites under exclusivity. Those figures describe a much wider infrastructure pipeline; they do not measure the Kyndryl site or its initial 10 megawatts.
The agreement gives Argentum access to Kyndryl’s existing high-density capacity, and the companies say that placement gives customers a faster path to high-performance inference capacity. The arrangement combines a ready operating site with Argentum’s responsibility for the customer-facing cluster service.
Argentum’s model extends beyond hardware supply. The company originates, finances and operates GPU clusters, then delivers dedicated bare-metal GPU capacity, meaning physical systems assigned to one customer, to AI model builders, inference providers and enterprises.
Jamie Rutledge, president of Kyndryl U.S., said: “Our agreement gives us a unique opportunity to put Kyndryl capacity to work in an innovative way.” He said Kyndryl’s mission-critical capabilities and decades of operational and services experience would support Argentum as it creates capacity for customers.
Rutledge tied Kyndryl’s operational history to Argentum’s plans to create customer capacity. Under the division of duties, Argentum runs the GPU clusters and customer service, while Kyndryl contributes the facility, power and operational support named in the agreement.
How does the power-first deployment work?
Argentum secures site control and power before purchasing GPUs. After those two inputs are in place, the company coordinates financing, relationships with original equipment manufacturers, cluster design and operational partners.
The company says the sequence is designed around customer demand, with capacity matched to workload requirements. Argentum’s process therefore begins with controlled infrastructure, then adds hardware and operating partners around the committed deployment.
Under the Kyndryl agreement, Kyndryl supplies the facility and power-capacity side, while Argentum coordinates GPU deployment and customer-facing service. The 20-year term gives the arrangement a defined infrastructure commitment, which the companies say will broaden as Argentum’s enterprise AI infrastructure needs evolve.
The agreement illustrates why existing high-density sites matter to Argentum’s approach. Kyndryl’s operating facility already has the power density required by current-generation GPU systems, allowing the company to use an existing site instead of relying only on construction-dependent buildouts.
Argentum and Kyndryl intend to broaden their collaboration as Argentum’s enterprise AI infrastructure needs evolve. They describe Boulder as the immediate test: an active GPU deployment that could anchor a broader infrastructure relationship if expansion plans become further commitments.
How does financing fit into the project?
Argentum says its financing structures are designed for contracted AI deployment. The stated aim is to scale capacity without waiting for capital to be assembled deal by deal, allowing the company to organize GPU investment around contracted workloads.
The 20-year Kyndryl commitment fits that model by pairing a defined infrastructure requirement with Argentum’s broader financing and cluster-operation role. Kyndryl still supplies the facility and power capacity, while Argentum retains responsibility for the GPUs and customer-facing service.
What services can customers choose?
Argentum says customers can choose how much of the infrastructure stack it manages. Its three service levels divide responsibility for hardware, software and applications.
- Hardware-as-a-Service provides physical infrastructure, power and connectivity. The customer remains responsible for the operating system, software and operations.
- Bare-Metal-as-a-Service adds operating-system delivery, break-fix repair work and maintenance to the infrastructure covered by the lower tier.
- Platform-as-a-Service extends Argentum’s responsibility through AI application management and reporting.
The tiers give Argentum a modular role in contracted deployments. It can supply physical capacity alone, add operating-system delivery and repair work, or assume application management and reporting, while the customer chooses which operating layer it retains.
How large is Argentum’s infrastructure pipeline?
Against that companywide pipeline, Boulder starts with 10 megawatts and an NVIDIA B300 deployment. Kyndryl’s agreement is one defined infrastructure commitment; Argentum’s reported 77,000 GPUs, 344 megawatts and 13 exclusive sites extend well beyond the Colorado facility.
The site model also explains why existing high-density capacity carries value for Argentum. The company promotes power-first sourcing as a faster alternative to construction-dependent buildouts, and Boulder already supplies a Tier III environment aligned with current-generation GPU systems.
Argentum and Kyndryl intend to broaden their collaboration as enterprise AI infrastructure needs evolve. Under their stated plan, Boulder is the immediate test of an active GPU deployment that could anchor a broader infrastructure relationship if expansion plans become further commitments.
What happens after the initial 10 megawatts?
For Kyndryl, the 20-year agreement supplies a flagship anchor tenant and a long-duration commitment tied to the Boulder site’s capacity. For Argentum, the arrangement tests whether power-first site control, staged GPU procurement and contracted deployment financing can operate as one customer-capacity model.
Further commitments would show whether the $630 million, 20-year Boulder structure can scale beyond its initial 10 megawatts. Without them, the disclosed project remains a single-site deployment; with them, it would become a template for the broader relationship the companies have described.
Source: unite.ai
More from Marcus Bennett
Show full bio
Senior reporter covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.
718 articles
Related articles
- NVIDIA to invest up to $100 billion in OpenAI for 10 gigawatts of AI infrastructure
- Edgevana Commits Over $700 Million to NVIDIA Blackwell Deployment
- Naver Lands $1.5 Billion AI Computing Deal With US Inference Firm
- Firmus Seeks $3 Billion Privately, With Nvidia and Meta Prepayments
- TensorWave Signs Six-Year OpenAI Deal for 457 MW of AMD AI Capacity