The Final Phase of the Intelligence Revolution
The AI investment cycle has moved through chips and data centers. The next wave — Physical AI — brings autonomous intelligence into factories, hospitals, and the machines themselves. This is what makes 365 million annual MRIs possible.
The AI Investment Cycle
Each phase of the AI build-out enables the next. The productivity gains from Physical AI are the final destination.
The Picks & Shovels
GPU and accelerator chip manufacturers. Every AI application requires massive compute. Nvidia's data center revenue grew 400%+ as every company raced to buy H100s.
The Infrastructure
Hyperscaler data centers. Trillion-dollar build-outs of the physical infrastructure to run AI models at scale. Data centers are the off-site brains for the entire AI economy.
The Enablers
Foundation model companies building the intelligence layer. Large language models, vision models, and agentic frameworks that power applications on top of Phase 1 and 2.
Physical AI
Productivity beneficiaries. Data centers act as off-site brains for agentic process control, advanced robotics, and predictive maintenance in factories. This is where MRI manufacturing automation lives.
Manufacturing the Machines at Scale
Today, the global MRI machine market produces approximately 6,000–6,500 units per year. The entire industry — Siemens Healthineers, GE HealthCare, Philips, Canon Medical — builds fewer machines annually than Tesla builds cars in a week.
The global MRI market is valued at approximately $9 billion today, projected to grow to $17 billion by 2035. But to hit 2040 goals for universal annual scanning, production must grow dramatically — requiring a 25%+ CAGR in manufacturing output.
This is only achievable through Physical AI: robotic assembly lines guided by AI process control, predictive quality inspection using computer vision, and supply chain optimization via agentic AI. The same technology that will read MRI scans must also manufacture the machines that take them.
Three Paths to 365 Million Annual Scans
The number of MRI machines required by 2040 depends entirely on how many scans each machine can perform per year. Physical AI changes that denominator dramatically.
| Scenario | Scan Throughput | Machines Needed | Production Challenge | Key Requirement | Likelihood |
|---|---|---|---|---|---|
A: Traditional Current technology, no AI optimization | ~4,000 scans/machine/year | 91,250 machines | 14x current global fleet Requires ~$140B+ investment | Business-as-usual manufacturing scale-up | Impractical |
B: AI-Optimized AI reading + agentic scheduling | ~10,000 scans/machine/year | 36,500 machines | ~5.5x current global fleet Requires ~$55B investment | AI reads + AI scheduling + extended hours | Achievable |
C: 24/7 Hyper-Automated Full Physical AI + 10-min AI scans | ~25,000 scans/machine/year | 14,600 machines | ≈ Current U.S. fleet No fleet growth required! | AI scans (10 min) + 24/7 ops + predictive maintenance | Target |
The key insight for Scenario C: The U.S. already has ~13,000 MRI machines. If AI-driven scan protocols reduce scan time from 45–90 minutes to 10 minutes, and predictive maintenance enables 24/7 operation, we don't need to build a single new machine — we need to use the existing fleet far more intelligently.
The Diagnostic Bottleneck
Even if we had the machines, the radiologist shortage would kill the vision. AI agents must solve the reading problem completely.
The AI Reading Pipeline
AI receives scan, performs quality check, identifies patient from metadata, accesses prior scan history.
Multi-model ensemble analyzes all anatomical regions simultaneously. Compares to longitudinal baseline. Generates structured findings report.
95%+ of scans show no significant findings. AI generates signed report, patient is notified automatically. Zero human involvement.
5% of scans with significant findings are routed to a radiologist with AI pre-annotations, risk scores, and relevant literature. Human reviews in minutes, not hours.
Result: 45,000 radiologists can handle 365 million annual scans — because they're only reading the 18 million flagged cases (5%), not all 365 million.
The Road to 2040
The milestones required to achieve universal annual MRI scanning.
AI Reading Reaches Clinical Grade
- ▸FDA clearances for autonomous AI MRI reading in specific anatomical regions
- ▸DTC companies (Prenuvo, Ezra) achieve sub-$500 full-body scan prices
- ▸First hospital systems deploy AI reading for routine screening
Agentic Operations Go Mainstream
- ▸AI handles end-to-end scheduling, billing, and insurance processing
- ▸Physical AI robots begin MRI machine assembly in manufacturing facilities
- ▸Scan times reduced to 20–30 minutes via AI protocol optimization
- ▸Price drops to $500–$1,000 for out-of-pocket scans
The Infrastructure Build-Out
- ▸Autonomous 24/7 MRI facilities operating in major metro areas
- ▸10-minute AI-optimized scan protocols achieve diagnostic parity
- ▸Insurance and Medicare begin covering annual scans for high-risk groups
- ▸Price drops below $300 for standard full-body scan
Universal Coverage
- ▸Annual MRI covered as preventative care for all Americans
- ▸AI reads 95%+ of scans autonomously
- ▸365 million annual scans achieved with current or near-current fleet
- ▸Healthcare costs begin declining as early detection prevents expensive late-stage interventions