Loading...
Loading...
Updated May 2026
Albuquerque is the command center for US national security research and development. Sandia National Laboratories, Kirtland Air Force Base, and a sprawling ecosystem of defense contractors, research institutions, and classified computing facilities all operate from the Albuquerque area. The city's AI implementation challenge is uniquely constrained by national security classification, compartmentalization rules, and the requirement that AI systems integrate with networks that cannot touch the public internet. A Sandia researcher might want to use an LLM to help analyze massive scientific datasets, classify research findings, or accelerate literature reviews on materials science or nuclear security—but the systems hosting that data live on classified networks (Secret, Top Secret, or compartmented), the LLM itself must be vetted for security compliance before it can access classified information, and the entire implementation must operate under Department of Energy (DoE) or Department of Defense (DoD) security protocols that are far more stringent than commercial security standards. Albuquerque implementation partners need Secret/Top Secret clearances, DoE and DoD compliance expertise, and the ability to design AI systems that operate in air-gapped or compartmented networks. LocalAISource connects Albuquerque research and contracting leaders with implementation partners who understand both the scientific opportunity of AI and the security constraints that govern national security research.
Most AI implementation projects at Sandia or Los Alamos start with a fundamental constraint: the LLM cannot ever see unclassified information mixed with classified data, and the classified network cannot touch the unclassified internet. A researcher might want to use Claude or GPT-4 to help analyze terabytes of scientific simulation output or research findings, but that analysis must happen on an isolated, classified network where the LLM itself has been vetted for compartment compliance. The implementation pattern is conservative: deploy a private LLM (Llama 2 or Mistral) on a classified network running inside Sandia's computing infrastructure, feed it only classified or carefully de-identified research data, and use the model only within the classified compartment. That infrastructure requires: Secret/Top Secret security clearance for the implementation team, DoE/DoD compliance certification of the LLM and inference infrastructure, full audit trails of every model interaction for security audits, and potentially classified reviews by the lab's security office before the system can go operational. Most classified AI implementations run 18-24 weeks and cost $300,000 to $600,000 depending on the classification level and the breadth of security audits required.
Not all Albuquerque research is classified. Universities (University of New Mexico, UNM College of Engineering), unclassified research contractors, and the unclassified portions of Sandia and Los Alamos work with scientific data that can be shared, published, and analyzed using commercial LLMs. For unclassified research, the implementation challenge shifts: how to integrate LLMs into scientific workflows (literature review automation, hypothesis generation from data, paper classification) without compromising research integrity or creating data governance nightmares. An unclassified research group might feed research papers, simulation data, and experimental results to Claude or GPT-4 to generate summaries, identify novel patterns, or suggest future research directions. The integration challenge is data quality and reproducibility: science requires traceability (which data did the model see? what was the exact prompt?), and the AI system needs to be designed so that research outputs can be published and the methodology can be audited. These implementations typically run 10-14 weeks and cost $80,000 to $200,000. Partners need both AI and research methodology expertise; partners who only understand AI will miss critical reproducibility and methodology requirements.
Defense contractors in Albuquerque (companies working on DoD or DoE contracts) operate under Federal Acquisition Regulation (FAR) compliance, ITAR (International Traffic in Arms Regulations) if they handle defense technology, and contract-specific security requirements. Any AI implementation touching defense contract work must: comply with DFARS clauses (DoD FAR Supplements) on cybersecurity and supply chain risk management, potentially undergo CMMC (Cybersecurity Maturity Model Certification) audits, and provide full traceability of the AI system's provenance and training data. That means: no commercial LLM training on your data (because you lose control of proprietary defense information), no cloud-hosted inference unless the cloud provider has appropriate DoD security certifications, and no integration with the defense contractor's systems unless the AI implementation has been security-tested and approved. These compliance requirements add 6-12 weeks to the project timeline and can increase costs by 30-50%. Partners need DFARS and CMMC expertise; generic AI implementation partners will miss critical compliance requirements.
Yes, but only under strict conditions. The LLM itself must be deployed on classified computing infrastructure (inside the lab's network, not on the internet), the model must be vetted for compartment compliance by the lab's security office, and every interaction must be logged for audit. The model cannot have been trained on any data outside the lab (so proprietary commercial models like ChatGPT are not acceptable), and the lab must have confidence that the model cannot be attacked to extract or leak classified information. Most national labs use private-hosted open-source models (Llama 2, Mistral) deployed on isolated infrastructure. Budget 18-24 weeks for the security vetting alone; that's separate from the implementation timeline.
Plan for 6-12 additional weeks for vetting if you're classified, 4-8 weeks if you're unclassified but DFARS-compliant. The vetting process involves the lab's security office, potentially the DoD or DoE, and requires documentation of the model's architecture, training data, inference infrastructure, and security controls. Partners need Secret clearance minimum for classified work; Top Secret clearance is often required. If your partner doesn't have appropriate clearance, they cannot work on classified systems. This is a hard constraint, not something you can work around.
Unclassified research or contractor AI: $80,000 to $200,000, 10-14 weeks. Classified network AI (Secret level): $300,000 to $600,000, 18-24 weeks. Top Secret or compartmented AI: $500,000+, 24+ weeks. The dramatic cost increase for classified systems is driven by security vetting, clearance requirements, and the need for in-house expertise. Budget for security vetting as a separate phase before implementation begins; you cannot compress this timeline.
No. Commercial models like GPT-4 or Claude are trained on internet data and live on internet-connected servers; sharing classified information with them is a security violation. The only acceptable model for classified work is a private-hosted, lab-approved model (typically Llama 2 or Mistral) deployed on the lab's own infrastructure. Some labs are developing approval processes for commercial models running on fully air-gapped infrastructure, but those are rare. Assume private hosting is required for any classified work.
Ask five things. First, do they have Secret or Top Secret security clearance? That's non-negotiable for classified work. Second, have they shipped AI systems at Sandia, Los Alamos, or other national labs? Ask for references that you can verify through your lab contacts. Third, do they understand DFARS, ITAR, and CMMC requirements if your work is defense-contractor related? Fourth, do they have relationships with lab security offices and can they guide you through the vetting process? Fifth, are they willing to work on fully air-gapped networks without regular internet access? Many contractors aren't, and that's a dealbreaker for classified work.
Get found by businesses in Albuquerque, NM.