GE HealthCare Hits 600 FDA AI Authorizations, Outpacing Siemens 3:1 in Medical Imaging

Concentration Risk in Medical Imaging AI

GE HealthCare reached 600 FDA-authorized AI algorithms by mid-2025, establishing a 3:1 lead over Siemens Healthineers in cleared clinical AI applications. For enterprise buyers, this concentration creates a decision point: standardize on the dominant platform and accept vendor dependency, or maintain multi-vendor AI infrastructure at higher integration cost.

The gap matters because clinical AI in radiology and cardiology increasingly runs as embedded software within imaging equipment, not standalone point applications. A health system buying GE scanners gets native access to GE's AI library. Switching costs rise with each additional algorithm deployed. One large academic medical center reported spending $400,000 in integration work to run non-native AI models on existing imaging infrastructure.

What the Authorization Count Actually Measures

FDA authorization volume measures regulatory clearance speed, not clinical utility. GE's 600 clearances include incremental variations—different body parts, patient populations, or imaging protocols for the same underlying model. Siemens' 200 authorizations may represent more distinct clinical applications.

The relevant metric for buyers is workflow coverage: which algorithms address your institution's specific diagnostic bottlenecks. A community hospital focused on emergency radiology needs stroke detection and pneumothorax flagging. A cancer center needs tumor characterization and treatment response monitoring. Authorization count becomes meaningful only after filtering for applicable use cases.

GE's volume advantage does indicate sustained R&D investment and regulatory process efficiency. The company filed an average of 50 AI authorizations per quarter through 2025, suggesting an internal machine for moving models from development to deployment. That pace matters for buyers planning multi-year AI roadmaps—vendor momentum predicts feature availability.

The Multi-Vendor Integration Tax

Health systems running mixed imaging fleets face a choice. Deploy only vendor-native AI and accept capability gaps, or build integration infrastructure for third-party models. The integration path requires PACS modifications, separate AI inference servers, and ongoing maintenance as both imaging systems and AI models update.

One 800-bed health system disclosed $1.2 million in annual costs supporting 15 third-party AI models across a mixed Siemens-GE-Philips environment. That figure includes infrastructure, IT staff time, and vendor support contracts. The same institution estimated native-only deployment would eliminate those costs but reduce AI coverage by approximately 30% across their clinical priorities.

Radiology Partners' $80 million acquisition of an AI workflow company in late 2025 signals where the market is heading: toward platforms that orchestrate multiple AI models regardless of imaging vendor. For enterprise buyers, the viability of that integration layer determines whether vendor lock-in is temporary or structural.

Pharma's Parallel Bet on Clinical AI

AstraZeneca's acquisition of Modella AI in January 2026 represents pharmaceutical companies buying clinical AI capability for drug development, not patient care. Modella's technology analyzes medical imaging to measure treatment response in oncology trials. The $50 million transaction—AstraZeneca disclosed the price in regulatory filings—values AI that reduces clinical trial duration.

This matters for health system buyers because pharma-funded AI development focuses on narrow, high-value problems with clear ROI. The resulting models often have limited applicability outside research settings. By contrast, imaging vendors build AI for high-volume, lower-margin diagnostic workflows. Buyers should expect pharma-developed AI to remain largely inaccessible for routine clinical use.

What to Watch

Track FDA authorization rates quarterly, but weight by applicable use cases for your institution. Request vendor roadmaps showing AI development aligned to your clinical priorities, not raw algorithm counts. Model the total cost of multi-vendor AI infrastructure before committing to fleet standardization. And monitor whether third-party integration platforms gain enough adoption to credibly reduce vendor lock-in—that shift would fundamentally change the standardization calculus.