Healthcare AI Spending Hit $3.7B, But Procurement Cycles Stretched to 10 Months
Clinical AI budgets grew 2.5× year-over-year, reaching $3.7 billion in 2026. Health systems are demanding staged deployments and measurable outcomes before enterprise commitments.
Enterprise budgets are rising while buying decisions slow
Healthcare AI spending reached $3.7 billion in 2026, up 2.5× year-over-year, according to Menlo Ventures. Clinical AI copilots grew fastest at 5.5× to $550 million, while patient-engagement and front-office AI climbed 4.5× to $450 million. Adoption broadened from 22% of healthcare organizations in 2025 to 42% in 2026.
The money is flowing, but buyers are taking longer to commit. The average health-system procurement cycle stretched from seven months to 10 months. For outpatient providers, it increased from five months to 11 months.
The gap between budget growth and decision speed signals a market shift. Buyers are no longer asking whether to adopt clinical AI. They are demanding integration plans, security reviews, measurable clinical and financial outcomes, and staged deployments rather than enterprise-wide rollouts. Governance, workflow integration, and evidence now function as budget gates.
Mass General Brigham spun out longitudinal-patient-data startup with $22M
Aevius launched on November 17 with $22 million in seed funding led by Insight Partners. The Mass General Brigham spinout is building models that combine clinical notes, imaging, biomarkers, and other patient data into a continuously updated, computable representation of each patient.
The target use cases include risk stratification, care coordination, clinical trial matching, and personalized decision support. The pitch competes against existing data warehouses, predictive models, clinical registries, and disease-specific AI firms.
The seed round validates market interest, but it does not validate clinical performance or production adoption. A longitudinal patient representation raises unresolved questions about data provenance, model updating, explainability, consent, bias, and liability. Buyers evaluating the approach will need to determine whether it adds sufficient value over their current predictive infrastructure.
LLMs still trail conventional models on structured clinical prediction
A ClinicalBench study compared 14 general-purpose LLMs, eight medical LLMs, and 11 traditional machine-learning models across three clinical prediction tasks and two databases. Neither general-purpose nor medical LLMs surpassed traditional machine-learning models, even after testing different model sizes, prompting methods, and fine-tuning strategies.
The result favors established gradient-boosting, regression, and other task-specific approaches for structured clinical prediction. LLMs remain stronger in unstructured-text interfaces, summarization, and workflow assistance.
Health systems should not replace validated task-specific models because an LLM offers a more modern interface. Procurement evaluations should compare models on the buyer's own population, calibration, subgroup performance, false-positive burden, and clinical utility. LLM pilots may require separate spending for retrieval, grounding, evaluation, monitoring, and human review. Traditional models remain the lower-risk choice for narrow prediction tasks.
Google launched HIPAA-supporting Gemini for healthcare with enterprise controls
Google announced Gemini for healthcare with access controls, end-to-end audit logging, and verifiable citations for HIPAA-supporting deployments. The platform includes partner capabilities such as an ASCO Oncology Guideline Assistant and lists 16 systems integrators, including Accenture, Cognizant, Deloitte, KPMG, PwC, Tata Consultancy Services, and Wipro.
The partner ecosystem could reduce integration costs for organizations standardized on Google Cloud or its listed service partners. Audit and citation features address enterprise review requirements, but HIPAA-supporting infrastructure does not establish clinical safety or regulatory clearance.
Buyers must distinguish platform controls from product-level validation. Each clinical agent still requires assessment of intended use, evidence, monitoring, human oversight, and responsibility for incorrect outputs. Google is competing for the healthcare platform layer against Microsoft's Azure AI and Nuance ecosystem, AWS healthcare services, EHR-native AI, and specialized clinical-AI vendors.
What to watch: multi-site deployments and evidence requirements
UC San Diego Health signed a multi-year Proximie agreement covering 44 operating rooms and procedure rooms across four sites. Cedars-Sinai is expanding its Aiva Nurse Assistant beyond an initial 48-bed surgical unit. UCLA Health is deploying radiology-triage tools for potentially critical findings such as suspected strokes and received a $25 million National Institute on Aging grant to lead a national evaluation of AI tools for Alzheimer's disease and related dementias.
These deployments favor vendors that can integrate into operating-room, nursing, radiology, and research workflows—not merely demonstrate model accuracy. Multi-site agreements raise the bar for interoperability, implementation services, clinical change management, and measurable staffing or throughput benefits.
Reported deployments do not automatically establish return on investment. Buyers should request baseline metrics, deployment scope, utilization, alert volumes, override rates, patient-safety events, and realized labor or throughput savings.
The market is separating into two tiers: workflow and infrastructure platforms that can support governed deployment at scale, and clinical prediction products that must prove they outperform established methods. The clearest near-term buying criterion is no longer "does the vendor use an LLM?" but "does the system deliver validated improvement inside the buyer's existing clinical workflow?"
Technology decisions, clearly explained.
Weekly analysis of the tools, platforms, and strategies that matter to B2B technology buyers. No fluff, no vendor spin.
