Argentum AI signs $4.1B contract for 27,000 Nvidia GB300 GPUs as cloud infrastructure race intensifies

Argentum AI has signed a $4.1 billion contract to supply approximately 27,000 Nvidia GB300 GPUs to an unnamed leading AI company, with capacity delivery beginning in phases later this year. The deal underscores the relentless demand for dedicated AI cloud infrastructure.

The contract covers long-term access to the GPUs through Argentum's global cloud platform, which spans more than 15 countries either live or contracted. Argentum founder and CEO Andrew Sobko said AI demand continues to outpace infrastructure availability, validating the company's ability to deliver next-generation GPU infrastructure at scale.

AI capacity crunch spurs massive GPU procurement

Argentum's agreement comes as cloud providers and AI companies scramble for compute. The $4.1 billion deal is one of the largest publicly disclosed GPU contracts this year. It follows Argentum's $2.5 billion data center partnership with Boosteroid and DL Invest Group in Poland, where the company will deploy institutional-scale GPU infrastructure within a new 300MW facility.

Other recent developments in the AI infrastructure ecosystem include:

• Dragos launched EmberAI, an AI-powered tool for operational technology cybersecurity, built on its extensive OT threat dataset to help defenders detect and respond to attacks faster.
• AWS introduced cross-account memory access for Amazon Bedrock AgentCore, allowing agents to share memory resources across multiple AWS accounts for more flexible multi-account architectures.
• AWS HealthOmics added ephemeral scratch storage (mounted at /tmp) for private workflows, giving bioinformatics tasks dedicated local space that improves run consistency and reduces costs.

Cloud and security vendors adapt to specialized AI workloads

The new capabilities from AWS and Dragos reflect a broader trend: AI is driving not only massive GPU demand but also a need for more specialized infrastructure services. Cross-account memory in Bedrock lets organizations build complex AI agent systems that span teams, while HealthOmics' scratch storage directly addresses the data-intensive nature of genomic analysis. On the security side, OT environments increasingly require AI-based defenses as industrial systems become more connected.

Argentum's contract and the AWS updates point to an infrastructure market that is bifurcating: hyperscale players invest in general-purpose GPU clouds, while specialists like Argentum and domain-specific services like HealthOmics target particular use cases. The next year will likely see more large GPU deals and tighter integration between AI platforms and the underlying network, storage, and security stack.