A National Neural Network: AI-Broadband Symbiosis as Health Infrastructure

The December 2025 Executive Order on Artificial Intelligence has generated familiar responses from familiar quarters. Legal scholars debate federal preemption while state attorneys general prepare litigation. Technology advocates celebrate regulatory restraint, while civil society groups warn about the harms of algorithmic bias. This well-rehearsed debate overlooks what may prove most consequential for American health care: a buried provision in Section 5 that concerns infrastructure rather than regulation.

That provision makes state eligibility for broadband funding contingent upon compliance with federal artificial intelligence (AI) policy. On the surface this could be viewed as standard political leverage dating back to the Interstate Highway Program. A closer reading reveals something more interesting and potentially more important. Section 5 acknowledges, perhaps unintentionally, that AI applications and broadband infrastructure cannot be sensibly separated in policy or practice. Modern clinical AI applications run on cloud servers rather than local machines, requiring substantial bandwidth for transmitting imaging studies and maintaining continuous data streams. Linking broadband funding to AI policy alignment creates conditions for unified national broadband deployment that fragmented state approaches cannot achieve.

The Broadband Equity, Access, and Deployment (BEAD) program allocates $42.5 billion to expand high-speed internet in unserved and underserved communities. Section 5 grants the US Department of Commerce the authority to condition non-deployment BEAD funds on a state’s AI regulatory posture. States maintaining what the administration characterizes as “onerous AI laws” lose eligibility for workforce training, digital literacy programs, and application adoption funding. The careful targeting of non-deployment funds rather than deployment funds reveals the logic of infrastructure policy rather than purely punitive intent. Physical construction funding remains available regardless of regulatory alignment; however, the activities that enable successful AI adoption are contingent upon participation in unified frameworks.

This matters urgently because the US health care system is facing a physician shortage that traditional policy responses cannot address on timeframes that matter. The Association of American Medical Colleges projects deficits of 37,800 to 124,000 physicians by 2034, with primary care shortages expected to hit hardest.¹ Approximately 65 million Americans live in Primary Care Health Professional Shortage Areas with fewer than one physician per 3,500 people. Meeting minimally adequate coverage would require an additional 14,500 primary care physicians willing to practice where they are most needed.

Community health centers occupy particularly important terrain in this landscape. Roughly 1,400 organizations operating over 16,000 service delivery sites provide care to 32 million patients annually. These demographics correlate strongly with clinical and social complexity, including multiple chronic conditions, language barriers, housing instability, food insecurity, and limited health literacy. Physician vacancy rates are substantial in some regions despite targeted federal support and loan repayment programs. Traditional responses operate on timeframes measured in decades rather than years. Medical education from initial enrollment through residency completion requires seven to eleven years of intensive training. Rural communities losing their only primary care physician or their last general surgeon cannot reasonably wait ten years for a replacement.

Clinical AI might address some dimensions of this crisis, at least in principle. Algorithmic decision support could extend physician reach by helping generalists manage cases that currently require specialist referral. A primary care physician managing 2,000 patients cannot personally review every laboratory result, vital sign measurement, and medication adherence pattern on a continuous basis. Algorithms scanning these data streams can identify patients exhibiting concerning trends, allowing physicians to shift their attention from routine monitoring toward targeted interventions where they matter most. Predictive systems could identify high-risk patients needing intensive intervention, allowing scarce clinical attention to be allocated more rationally. Remote monitoring augmented by machine learning could support chronic disease management without requiring frequent office visits that strain already inadequate clinical capacity. Agentic AI (i.e., autonomous systems that can independently execute tasks with minimal human intervention) could handle administrative tasks that would reclaim 40% of the time currently consumed by physicians for scheduling, prior authorizations, and documentation, allowing them to focus on patient care.

These promising applications share a critical prerequisite: infrastructure that much of rural America and many urban underserved areas simply do not possess. Cloud-based AI requires reliable, high-speed connectivity rather than intermittent service that is subject to weather events, physical damage, and environmental disruptions. Clinical decision support systems integrated into workflows require sufficient bandwidth and low latency for real-time operation, allowing clinicians to safely depend on them. Remote patient monitoring requires not just institutional broadband but adequate home internet access. The familiar gap between principle and practice appears at the infrastructure layer. The populations that might benefit most from AI-augmented care often cannot access these technologies because the necessary infrastructure simply does not exist.

Section 5 establishes a policy architecture that could alter this situation. If AI regulatory standards achieve national uniformity while BEAD deployment prioritizes underserved areas, as mandated by statute, then the conditions are in place for deploying standardized applications across the entire infrastructure footprint. The interstate highway system offers an instructive analogy. Federal standards governing construction, signage, and specifications restricted state autonomy but enabled a unified national network, allowing vehicles to travel coast to coast on consistent roads. Commercial trucking, passenger travel, and national supply chains all depend on standardization that initially seemed like an unwelcome federal intrusion.

A Proposed Federal-Rural AI Corridor Architecture

A Federal-Rural AI Corridor would operationalize this logic specifically for health care (Figure 1). The concept draws on infrastructure policy terminology, describing standardized pathways that enable the reliable flow of goods, services, or information across geographical areas. For example, the Northeast Corridor supports high-speed rail through infrastructure meeting consistent specifications across multiple state jurisdictions. A Federal-Rural AI Corridor would enable seamless clinical applications to flow from development centers to deployment sites through infrastructure that meets uniform standards for connectivity, security, and interoperability. The framework would address interconnected problems simultaneously rather than treating them separately. It would provide clinical AI developers with regulatory clarity and defined market access, reducing barriers to navigating inconsistent state requirements. It would offer rural facilities and community health centers access to validated AI applications, along with implementation support, thereby reducing the technical and organizational barriers that these resource-constrained institutions currently face. It also would enable policymakers to address workforce shortages through technology while maintaining appropriate oversight and protecting patients from algorithmic harms.

Figure 1. The Proposed Federal-Rural AI Corridor Architecture

The Corridor would operate through three implementation tiers. The infrastructure foundation would establish connectivity standards specifically designed for clinical AI in workforce-constrained settings. Current BEAD requirements specify minimum speeds reflecting general consumer access rather than clinical application requirements. The Corridor framework would require enhanced specifications for health facilities, including security and privacy requirements embedded at the infrastructure level with network segmentation, strong encryption, and robust authentication.

The application standards tier would establish criteria for AI systems eligible for Corridor deployment. The Food and Drug Administration (FDA)’s Software as a Medical Device framework and the National Institute of Standards and Technology’s AI Risk Management Framework provide baseline requirements for software as a medical device. Corridor standards should include specific requirements that reflect deployment conditions in settings with limited specialist backup and constrained diagnostic infrastructure. The governing principle should emphasize minimizing extrapolation risk while maximizing genuine workforce extension. Drug repurposing applications, which identify potential new therapeutic uses for existing approved medications, primarily rely on interpolation within existing real-world evidence rather than extrapolation to novel mechanisms or patient populations. These approaches carry inherently lower risk profiles than systems attempting predictions for populations entirely absent from training datasets. Validation must demonstrate performance across genuinely diverse populations representative of community health center patients, and interoperability must be mandated rather than treated as optional.

The care delivery models tier would address integration into actual clinical practice. Technology succeeds or fails based on workflow integration rather than technical sophistication alone. Hospital-at-Home programs, which enable acute care delivery through remote monitoring and AI-powered decision support, represent promising models for areas with a workforce shortage. Agentic AI handling administrative tasks represents a particularly low-risk and high-value deployment. For community health centers where physicians spend nearly half their time on administrative work, well-designed systems could reclaim substantial capacity for activities requiring human judgment.

The federated architecture of the proposed Corridor deserves particular emphasis. Unlike centralized approaches that extract data from rural facilities to train models ultimately owned by distant technology companies, a federated structure enables genuine local data sovereignty. Individual facilities retain control over patient data that never leaves secure local systems. Model training is conducted through privacy-preserving techniques, where only parameters, rather than actual records, are shared. This prevents what critics accurately characterize as “data colonialism” (i.e., the systematic exploitation of underserved populations as data sources for algorithms serving wealthier markets).

Implementation of the proposed architecture requires coordination spanning federal, state, professional, and institutional levels. This deliberate multi-level approach mirrors existing health care governance arrangements while enabling unified AI deployment addressing workforce challenges. Multiple federal agencies possess relevant jurisdiction, including the National Telecommunications and Information Administration (NTIA) for broadband infrastructure, the Food and Drug Administration for medical device regulation, the Centers for Medicare and Medicaid Services for health care reimbursement, the Health Resources and Services Administration for community health center programs, and the Office of the National Coordinator for Health Information Technology for health information technology interoperability. Effective coordination across these distinct agencies with different missions and institutional cultures represents a substantial administrative challenge but proves absolutely essential for coherent policy.

The Executive Order establishes unambiguous deadlines, creating genuine urgency. The Department of Commerce must complete its evaluation identifying “onerous” state AI laws by approximately March 11, 2026. The NTIA Policy Notice specifying BEAD eligibility conditions follows shortly thereafter. Policy design decisions made during the first quarter of 2026 will establish frameworks that persist throughout program execution, extending through 2028. Stakeholder input during this formative period carries significantly greater weight than feedback offered after basic frameworks are solidified.

The question facing medicine is whether to engage this formative process constructively or to leave infrastructure design entirely to telecommunications engineers, corporate lawyers, and technology companies whose expertise and incentives do not extend to managing diabetic patients in towns with one struggling clinic and no endocrinologist within 90 miles. The 90-day window remains open but is closing rapidly. The Department of Commerce evaluation and subsequent NTIA Policy Notice will establish durable frameworks that shape health care infrastructure investment for years to come. For those committed to ensuring that AI deployment addresses workforce shortages and expands access for genuinely underserved populations, rather than merely serving those already advantaged by geography and an abundant physician supply, meaningful engagement cannot wait.

References

About the Authors