The Infrastructure Decade: Why Healthcare's AI Future Is Being Built in Layers, Not Launched Overnight
Image: Depositphotos
A few months ago, I overheard a health-system CIO, who summed up the problem perfectly.
“Our issue isn’t ambition. It’s amperage. We’ve got ideas for a hundred AI pilots—but not the power, pipes, or data discipline to run them safely at scale.”
He’s right.
Every week, I meet healthcare leaders who are thinking boldly about AI but are quietly bumping into the same invisible wall: the limits of their infrastructure.
We’re not suffering from a shortage of algorithms. We’re suffering from an absence of architectural readiness.
The Decade of Infrastructure
Across industries, we are entering what Jensen Huang calls the industrial revolution of intelligence¹. In this week’s NVIDIA keynote, he described how the world is moving from retrieval-based computing to generative-based computing—from traditional data centers to what he calls AI factories¹.
These aren’t marketing metaphors. They represent a structural shift in how computation itself is organized. As Huang put it, “In the future, the computer’s going to generate the tokens for the software… the computer has become a generator of tokens, not a retrieval of files.”
He explained that global data-center build-outs are reaching an inflection point, projecting nearly a trillion dollars in new infrastructure investment over the next decade¹. Each “AI factory,” he noted, will consume roughly a gigawatt of power and cost $50–60 billion to build.
That scale might sound far removed from healthcare—but it mirrors our challenge precisely. Just as energy and power constrain AI factories, bandwidth and governance constrain health enterprises.
You can’t run intelligence on top of fragility.
Andrej Karpathy captured this well, in a recent interview, when he said we are entering the decade of agents, not the year². That’s especially true in healthcare. Our biggest breakthroughs will come not from larger models, but from systems that can reason, recall, and operate within clinical and regulatory constraints. Those systems demand infrastructure that can think in real time.
Architecture Over Algorithms
Healthcare’s inefficiencies are not moral failures; they’re architectural ones.
Nearly $400 billion in annual administrative waste³.
43 percent of patients waiting more than two weeks for claims settlement³.
55 percent growth in Medicare Advantage denials in a single year³.
These numbers don’t reflect bad intentions—they reflect systems designed for file exchange, not reasoning. We still push PDFs through fax machines and call it interoperability.
To move beyond that, we need infrastructures that support continuous reasoning and verification:
Parallel compute environments that analyze clinical, contractual, and utilization data concurrently.
Unified data fabrics that let patient context, eligibility, and evidence travel together.
Low-latency coordination between payers, providers, and regulators so every decision has provenance.
Huang calls this extreme co-design: when chips, systems, networking, and software evolve together as one stack. Healthcare needs the same mindset—co-designing compute, policy, and care delivery into an intelligent system of record.
Five Signals That the Infrastructure Decade Has Begun
1. Intelligence is moving to the bedside.
Hospitals are beginning to deploy distributed inference nodes inside their own networks, bringing decision support within milliseconds of data creation⁴. In emergency and radiology settings, that latency reduction saves time—and lives.
2. Hospitals are confronting network debt.
A 2025 World Journal of Advanced Research study found that two-thirds of public hospitals cite unreliable connectivity as their main AI barrier⁵. That’s why some executives now rank bandwidth upgrades alongside cybersecurity as top priorities for 2026 budgets.
3. The economics of compute are improving.
Accelerated computing architectures are delivering multi-fold efficiency gains per watt compared with general-purpose CPUs. The real story isn’t raw speed—it’s the cost of reasoning finally dropping to levels where continuous inference becomes financially sustainable.
4. Integration is the new bottleneck.
Studies show that 95 percent of enterprise AI pilots fail to reach production⁶. The cause isn’t model quality—it’s integration debt. Healthcare’s legacy interfaces, regulatory checks, and data silos make plug-and-play impossible. The organizations investing early in modular, event-driven architectures will break that cycle first.
5. Governance is maturing into engineering.
Boards are treating AI oversight the same way they treat clinical quality. Explainability, bias monitoring, and traceability are now part of design specifications, not afterthoughts. That cultural shift—embedding trust into the build—is what makes scaling safe.
Together, these trends prove that healthcare’s AI future won’t arrive as a single product launch. It will arrive layer by layer, as infrastructure aligns with intent.
Building the Cognitive Infrastructure
In my own work, I see this transformation forming three interdependent layers:
Physical Infrastructure – distributed compute capacity, energy efficiency, and resilient connectivity that can handle real-time reasoning across the enterprise.
Logical Infrastructure – the communication fabric that routes data and inference securely across payers, providers, and regulators.
Cognitive Infrastructure – the reasoning layer: systems that understand causality, quantify uncertainty, and learn from outcomes under governance.
These layers together form what I call the neural architecture of healthcare—an ecosystem where intelligence is decentralized, auditable, and continuously learning.
A Phased Path Forward
2025–2028 | Foundation Building: Hybrid compute environments, strong governance frameworks, pilot deployments of real-time prior authorization and edge diagnostics.
2028–2035 | Ecosystem Development: Multi-agent collaboration between payer and provider systems; autonomous document generation; cross-enterprise auditability.
2035–2042 | Full Transformation: End-to-end intelligent operations—claims, utilization management, population health—under transparent AI oversight, with administrative overhead below 5 percent.
This isn’t speculative—it’s consistent with what Huang described: years of planning to secure land, power, and capital before large-scale autonomy becomes viable. The same applies to healthcare. We must plan for data, trust, and compute before autonomy can be realized safely.
Why I’m Optimistic
In every major wave of digital progress, from electronic records to cloud migration, transformation followed the same pattern: vision → infrastructure → acceleration.
We are in the infrastructure phase now—the hardest, least glamorous, but most important part.
Huang predicts that over the next decade, tens of gigawatts of AI infrastructure will come online worldwide. Healthcare will be one of the beneficiaries, but only if we design our systems with the same rigor that semiconductor engineers apply to their fabs.
This is why I remain confident in the 2042 timeline. It’s not the year healthcare adopts AI—it’s the year our infrastructure catches up to our intelligence.
And when that happens, we won’t be talking about algorithms or pilots. We’ll be talking about healthcare that finally thinks—and learns—in real time.
References
Huang, J. (2025). NVIDIA GTC Keynote, Washington D.C. Verified transcript.
Karpathy, A. (2025). Interview with Dwarkesh Patel on Agentic AI and Infrastructure Requirements.
Ronanki, R. (2025). From Automation to Autonomy: Preparing for Health Plan 2042. Lyric White Paper.
Ronanki, R. (2025). NeuralGrid and NeuralFabric: Designing the Intelligent Infrastructure for Healthcare’s Autonomous Future. Substack.
World Journal of Advanced Research. (2025). Building AI-Ready Infrastructure for U.S. Healthcare.
Massachusetts Institute of Technology. (2025). Enterprise GenAI Pilot Outcomes Study.
This article was posted by Rajeev Ronanki on his substack on November 4, 2025: here.
Rajeev Ronanki
Rajeev Ronanki continues to reimagine the future of healthcare by harnessing the power of AI and data to provide consumers with predictive, proactive, and personalized insights at the intersection of healthcare supply and demand. His experience spans over 25 years of innovation-driven industry and social change across healthcare and technology, and he regularly speaks on topics related to navigating the future of healthcare, harnessing data-driven insights, and delivering personalized experiences. In November 2021, Rajeev released “You and AI: A Citizen’s Guide to AI, Blockchain, and Puzzling Together the Future of Healthcare,” which has become an Amazon Best Seller.
When he served as the President of Carelon Digital Platforms, Rajeev led efforts to transform Elevance Health into a digital platform for health and wellbeing. He and his leadership team collaborated with internal and external partners to expand virtual care, create a longitudinal patient record to improve care and reduce overall administrative burden, deploy AI to increase auto-adjudication of claims and expedite manual claims review processes, modernize the provider data lifecycle into a single source of truth, and pilot innovative solutions to transform the way consumers interact within the healthcare ecosystem.
Before Elevance Health, Rajeev was a partner at Deloitte Consulting, LLP, where he established and led Deloitte’s life sciences and healthcare advanced analytics, artificial intelligence, and innovation practices. Additionally, he was instrumental in shaping Deloitte’s blockchain and cryptocurrency solutions and authored pieces on various exponential technology topics such as artificial intelligence, blockchain, and precision medicine.
Rajeev obtained a bachelor’s degree in mechanical engineering from Osmania University in India and a master’s degree in computer science from the University of Pennsylvania.