Introduction

President Donald Trump recently signed National Security Presidential Memorandum 11 (NSPM-11), a directive instructing US military and intelligence agencies to significantly accelerate their adoption of advanced artificial intelligence. This strategic move aims to bolster the nation's technological capabilities in a sector critical for national security.

The NSPM-11 directive replaces the previous NSM-25 from the Biden administration, which had been in effect since 2024, introducing new priorities and a stronger focus on critical aspects. Among the most relevant provisions is the requirement to protect "frontier AI models"—the most advanced and strategic ones—from potential theft by foreign adversaries, a clear signal of intensifying geopolitical competition in the AI field.

Directive Details and Deployment Implications

A key element of NSPM-11 is the clause preventing any commercial vendor from "pulling the plug" or discontinuing service for AI systems critical to national security. This provision has profound implications for technology deployment strategies. For government agencies, it signifies a clear preference for solutions that guarantee total control over infrastructure and data, reducing reliance on third parties.

This requirement pushes towards the adoption of self-hosted or on-premise deployments, where hardware and software are managed directly by the organization. This approach, although it may involve a higher initial capital expenditure (CapEx) compared to cloud-based models (OpEx), offers advantages in terms of data sovereignty, security, and operational resilience. The ability to operate in air-gapped environments, completely isolated from external networks, becomes fundamental for protecting the most sensitive models.

Data Sovereignty and Technological Control

Trump's directive highlights a growing trend among government entities and large enterprises handling sensitive data: the need to maintain full control over their AI assets. Protecting "frontier models" from external theft is not just about cybersecurity; it's also about ensuring that intellectual property and strategic capabilities remain under national control.

This context makes on-premise deployments particularly attractive. They allow for precise definition of access policies, implementation of rigorous physical and logical security controls, and adherence to stringent compliance regulations. Internal management of infrastructure, including servers with high-performance GPUs (such as those with high VRAM for Large Language Models inference), dedicated storage, and secure networks, is essential for building robust and protected AI pipelines.

Future Outlook and Trade-offs

The acceleration of AI adoption in the military and intelligence sectors, combined with the vendor control clause, sets the stage for a significant evolution in the approach to AI deployment. Organizations will need to balance the need for rapid implementation with the necessity of maintaining unassailable technological sovereignty.

This implies a careful evaluation of the Total Cost of Ownership (TCO) for self-hosted solutions, considering not only the purchase of hardware and licenses but also the costs of management, maintenance, and upgrades. Although complexity may increase, the benefits in terms of security, control, and strategic autonomy are considered paramount. For those evaluating on-premise deployments for LLM workloads, AI-RADAR offers analytical frameworks on /llm-onpremise to assess these trade-offs, providing objective guidance on the various available options.