The Horizontal Enabling Layer: Structural Disruption and the Reconfiguration of Healthcare AI Moats

Nelson Advisors
2 hours ago
14 min read

The healthcare technology landscape has entered a period of profound architectural transition, characterised by the emergence of a horizontal enabling layer that is systematically dismantling the traditional moats of vertical software providers.

Historically, the healthcare sector was defined by highly fragmented, task-specific "point solutions" that relied on proprietary data silos and steep switching costs to maintain market dominance.

However, the advent of foundation models and agentic AI platforms has introduced a new substrate of generalisable intelligence that treats legacy systems of record as mere infrastructure. This "quiet shift" is not merely a technical evolution but a fundamental redistribution of economic power within the software stack, where value is migrating from information storage to workflow orchestration.

As healthcare organisations adopt AI at 2.2 times the rate of the broader economy, the industry is witnessing the industrialisation of innovation, where the ability to "build a bridge" across established moats is becoming more valuable than the moats themselves.

The Emergence of the Horizontal Substrate: Foundation Models and Clinical Generalisation

At the core of this transition is the shift from narrow, supervised machine learning to large-scale foundation models. Traditional clinical AI was built as a series of specialists, individual models trained on highly labeled datasets to perform singular tasks, such as identifying a specific fracture or a pulmonary nodule. This approach created a fragmented ecosystem where each new clinical use case required a ground-up development effort, resulting in high costs and limited scalability.

The new horizontal layer, exemplified by multimodal foundation models, learns general data representations across diverse sources including text, images and structured electronic health records (EHRs). This enables a single model to adapt to many clinical tasks through fine-tuning or prompting, effectively turning AI from a collection of narrow tools into a system-wide capability.

The technical mechanism driving this horizontality is self-supervised learning, which allows models to train on vast amounts of real-world, unlabeled data by solving pretext tasks. This approach eliminates the manual labeling bottleneck that previously restricted clinical AI to well-defined niches. For instance, Aidoc’s Clinical AI Reasoning Engine (CARE) functions as a horizontal base layer that can be adapted across multiple medical domains, radiology, cardiology and beyond, without rebuilding the intelligence architecture from scratch.

This versatility is critical for health systems that are increasingly wary of managing hundreds of disparate point solutions. By moving AI from passive alerting to active decision support, foundation models accelerate care delivery and reduce the delays associated with siloed information.

Model Generation	Architecture Strategy	Primary Training Data	Scaling Mechanism	Clinical Scope
Traditional AI	Vertical/Task-Specific	Small, Manually Labeled	Linear (New Task = New Model)	Narrow (Nodules, Fractures)
Foundation Models	Horizontal/Enabling Layer	Massive, Multimodal/Unlabeled	Exponential (Fine-tuning)	Broad (Diagnostic Reasoning)
Agentic AI (2026+)	Orchestration/Goal-Driven	Real-time Contextual Streams	Autonomous (Multi-agent chains)	End-to-End Workflows

The transition toward horizontal foundation models is also reshaping the competitive dynamics between Big Tech and medical startups. While general-purpose models like GPT-4 demonstrate impressive breadth, they often struggle with the precision required for clinical practice, where the cost of error is uniquely high.

This has led to the rise of healthcare-specific foundation models, such as NYU’s clinical LLMs or Tempus’s multimodal pipelines, which are trained exclusively on high-quality medical data. These domain specific platforms combine the generalisable power of horizontal AI with the accuracy of vertical expertise, creating a formidable "clinical-grade" intelligence layer that generalist assistants cannot easily replicate.

The Erosion of Traditional Moats: Analyzing the Displacement of Vertical SaaS

The most significant implication of the horizontal enabling layer is the erosion of the "moats" that have protected incumbent healthcare software for decades. These moats were primarily built on four pillars: high switching costs, proprietary data advantages, regulatory complexity and deep workflow embedding. As AI capabilities expand, each of these pillars is facing unprecedented pressure from AI-native competitors and the commoditisation of raw intelligence.

Switching Costs and the "Invisible Plumbing" Risk

For years, the Electronic Health Record (EHR) was the ultimate moat. The cost of migrating tens of millions of records and retraining thousands of clinical staff made "rip and replace" strategies almost unthinkable for major hospital systems.

However, horizontal AI agents are beginning to disintermediate these products by operating as an "overlay" layer. By using APIs and advanced mapping techniques to interact with messy legacy schemas, AI can abstract the user interface away from the underlying system of record. When the clinician interacts primarily with an AI assistant that handles documentation, scheduling, and billing, the EHR is reduced to "headless SaaS", a back-end database that can theoretically be swapped out with far less friction than in the past.

This shift represents a move from "friction-based stickiness" to "value-based stickiness." Incumbent software that relies solely on being a repository of information is seeing its "fade rate", the speed at which excess profits shrink, accelerate as competitors use AI to build "bridges" across their historical barriers. Investors are increasingly penalising companies whose moats are driven by user inertia rather than mission-critical value.

The Data Moat Paradox: Proprietary vs. Synthetic Context

Proprietary data was once considered the unreplicable asset of healthcare tech. Companies that spent a decade collecting genomic sequences or radiology images held a significant lead. However, the horizontal AI shift has introduced two counter-forces: synthetic data and federated learning.

Synthetic data generation, utilising GANs and diffusion models, allows new entrants to create high-fidelity artificial datasets that mirror the statistical structure of real patient data without the privacy constraints of HIPAA or GDPR. This allows startups to "bootstrap" their models and overcome the data scarcity problem that previously favored incumbents.

Data Type	Moat Durability	AI Erosion Risk	Strategic Response
Static Repositories	Low	High (Replicable by Synthetic Data)	Move to Real-time Orchestration
Longitudinal/Outcome-Linked	High	Moderate (Hard to simulate temporal depth)	Secure Exclusive Hospital Pipelines
High-Dimensional Interactive	High	Low (Messy real-world context is hard to fake)	Focus on Exception Handling Logs
Synthetic/Generated	N/A	Catalyst for new entrants	Use for Edge Case Training

Furthermore, federated learning enables models to train across decentralised datasets without moving sensitive information, allowing multiple institutions to collaborate on a shared intelligence layer while maintaining data sovereignty. This democratises access to high-quality training signals, effectively neutralising the advantage of having a single, massive silo.

The "data moat" is therefore shifting from simple ownership of information to the ownership of the "feedback loop", knowing not just what data exists, but how a clinical decision actually impacted a patient outcome over time.

Market Growth and Economic Drivers (2026–2031)

The economic impact of the horizontal AI shift is immense, with the healthcare AI market projected to reach hundreds of billions of dollars by the early 2030s. This growth is driven by the urgent need to address systemic inefficiencies, including a $200 billion to $360 billion potential for annual savings in the U.S. alone. AI adoption in healthcare reached 85% by the end of 2024, and the industry is now spending real capital, $1.4 billion in 2025, to industrialise these solutions.

Segments of Rapid Expansion

Growth is concentrated in segments where the ROI is immediate and measurable: administrative efficiency, medical imaging, and drug discovery. The software solution segment, which dominated the market with a 46% share in 2025, is being supplemented by a rapidly growing services segment as organisations seek help integrating these complex horizontal layers into their existing workflows.

Market Segment	2025-2026 Base (USD)	2031-2033 Projection (USD)	CAGR (%)	Primary Driver
Global AI in Healthcare	36.67 Billion (2025)	505.59 Billion (2033)	38.90%	Efficiency & Precision
North America AI Healthcare	5.83 Billion (2025)	23.07 Billion (2031)	25.72%	Labor Shortage
Healthcare Analytics	64.49 Billion (2025)	369.66 Billion (2034)	21.41%	Unstructured Data Mining
AI Precision Medicine	2.70 Billion (2025)	14.85 Billion (2034)	20.87%	Targeted Oncology
Healthcare Gen AI (U.S.)	518.4 Million (2023)	10.17 Billion (2030)	36.40%+	Workflow Automation

Administrative AI currently captures the largest share of investment (roughly 60%) because it targets "rule-based, mechanical" work where AI excels. Medical coding, for instance, has been transformed by platforms like Hathr.AI, which can automate CPT and ICD-10 suggestions with nearly 100% accuracy, saving large practices hundreds of thousands of dollars in annual labor costs while eliminating backlogs. The clinical decision support segment remains at a lower maturity rate (6.8%) due to regulatory and liability concerns, but it is expected to accelerate as foundation models prove their reliability.

Productivity and the New Economic Map

The shift toward horizontal AI is splitting the healthcare professional class into two radically different economic populations. "Lower Level 4" workers, such as junior lawyers, medical coders and financial analysts, are seeing their tasks automated, leading to a "slow squeeze" on their market value. Conversely, "Upper Level 4" stakeholders, CEOs, founders and venture investors, are capturing the gains as capital replaces labor.

This is evidenced by the "ARR per FTE" (Annual Recurring Revenue per Full-Time Employee) metrics: traditional healthcare services generate $100K–$200K per FTE, while AI-native healthcare companies are hitting $500K–$1M+, a efficiency gain that is fundamentally reconfiguring hospital balance sheets.

The Rise of Orchestration: Agentic AI as the New Control Plane

As raw intelligence becomes a horizontal commodity, the new "battleground" for moats is the orchestration layer. It is no longer enough to have a model that can read an X-ray; the value lies in a system that can take that reading, autonomously coordinate with the pathology department, update the patient’s EHR, schedule a follow-up, and submit the insurance claim without human intervention. This is the promise of Agentic AI, a sub-domain of AI capable of autonomous operation and goal-driven behaviour.

The Technical Composition of Agentic Orchestration

Agentic orchestration platforms function as the "brain" of the distributed clinical process. They manage four interrelated phases: perception (ingesting multimodal data), reasoning/planning (determining the clinical path), action (executing orders or updates) and learning/feedback (refining behaviour based on outcomes). These systems rely on specialised "worker agents" for domain-specific tasks and "orchestration agents" to manage dependencies and execution order.

For health systems, this modular architecture provides a way to act locally but think enterprise-wide. Instead of having siloed point solutions, they can build a "living network" of intelligence where different models, whether from Microsoft, NVIDIA, or a specialised startup, are unified by common security and data frameworks. This orchestration tier enforces priority, governs policy, and arbitrates decisions, ensuring that the autonomous actions of AI remain aligned with clinical standards and institutional goals.

Orchestration Capability	Functional Impact	Clinical Example
Multi-Model Chaining	Combines different AI outputs	Linking radiology find with genomic risk
State Management	Tracks patient context over time	Managing a 6-month oncology pathway
Policy Enforcement	Ensures regulatory compliance	Blocking non-HIPAA data transmissions
Dynamic Agent Discovery	Finds best model for task	Routing rare disease case to specialist AI

The strategic value of this layer is immense. Organizations that own the orchestration layer control the "who, what, when, and how" of healthcare. This allows them to capture the "labor spend", the $740 billion annually flowing into administrative tasks, rather than just competing for the $63 billion IT software budget.

Case Study: The "Quiet Shift" in Action

The tension between horizontal generalism and vertical specialization is best illustrated by recent implementations in hospital settings. A major Boston hospital initially attempted to use a general-purpose AI chatbot to assist radiologists with chest X-ray analysis. The results were suboptimal; the generalist AI frequently misidentified anatomical terms and required constant human correction, leading to clinician frustration. The hospital subsequently switched to Aidoc, a vertical AI platform purpose-built for radiology. Within weeks, the system was detecting brain bleeds and pulmonary embolisms with 95% accuracy and crucially, flagging them for immediate attention within the existing workflow.

This case study highlights the "Customization Burden" of horizontal AI. While tools like ChatGPT are cheaper to license initially, the cost of training, integration and ongoing maintenance to make them clinically safe often exceeds the cost of a specialised vertical solution. This suggests that while foundation models provide the enabling layer, the "durable value" will come from domain-specific moats that make automation reliable and auditable.

The Role of Hyperscalers: Infrastructure as a Service

The "Big Tech" players, Microsoft, NVIDIA, Google, and Amazon, are positioning themselves as the ultimate horizontal layer for healthcare AI. Their strategy is to provide the massive compute, the foundational data stores, and the "AI factories" that power the rest of the industry.

NVIDIA and the Industrialisation of Biology

NVIDIA has transitioned from a chip designer to an integrated stack provider. Their BioNeMo platform acts as a "foundry" for biology and drug discovery, enabling companies like Eli Lilly to build their own frontier models for chemistry. The Lilly-NVIDIA "AI factory" project aims to build the pharmaceutical industry's largest supercomputer, focusing on "physical AI" and robotics to accelerate medicine production. This partnership exemplifies how horizontal infrastructure providers are teaming up with vertical giants to redefine the R&D landscape.

Microsoft and the Cloud Migration of Clinical Data

Microsoft’s Azure Health Data Services is another horizontal pillar, designed to unify multimodal health data (imaging, clinical, device) in a FHIR-native cloud environment. By moving mission-critical EHR systems like Epic to Azure, Microsoft not only reduces hardware refresh costs for hospitals but also provides a "launchpad" for AI innovation. Once the data is unified in the cloud, hospitals can instantly spin up Azure AI models or third-party solutions without the traditional data silo constraints.

Hyperscaler	Key Healthcare AI Offering	Primary Strategy
Microsoft	Azure Health Data / DAX Copilot	Deep EHR integration & cloud-native clinical data
NVIDIA	BioNeMo / DGX Cloud	AI Factories for drug discovery and physical AI
Google	Vertex AI Search for Healthcare	Unstructured data mining & MedLM models
AWS	HealthLake / Bedrock	FHIR-native storage & modular foundation models

Regulatory Darwinism: The New Infrastructure Advantage

In the AI era, regulatory and compliance moats are strengthening even as technical moats erode. The implementation of the EU Medical Device Regulation (MDR) and the upcoming AI Act has created a massive, capital-intensive barrier to entry. Regulatory approvals are now treated as "tradable financial assets" because they are too expensive for small startups to obtain independently.

This "Regulatory Darwinism" favors established players who have already built the infrastructure for transparency, auditability, and clinical validation. For instance, the EU AI Act’s categorisation of many medical AI tools as "high-risk" necessitates robust data governance that early-stage entrants often lack. This creates a bifurcated market where established firms can rapidly integrate AI capabilities through their existing compliance pipelines, while new entrants face multi-year validation timelines.

However, this regulatory moat also creates an opportunity for horizontal "Compliance-as-a-Service" platforms. Companies like AirgapAI or Drata provide automated HIPAA and SOC 2 evidence collection, helping developers navigate the "compliance gauntlet" more quickly. By automating 100% of local PHI processing and offering air-gapped security, these platforms allow AI developers to focus on intelligence while the horizontal layer handles the regulatory risk.

The Transformation of Clinical Validation: In-Silico Trials and Digital Twins

One of the most profound shifts in the next five years will be the transition from physical to virtual validation. The high cost of patient recruitment and statistically powered trials, often in the tens of millions of dollars, has historically limited medtech innovation to the largest players. AI-generated synthetic data and "In-Silico Twins" (ISTs) are now lowering these barriers.

By creating high-fidelity, artificial replicas of biological systems, ISTs allow researchers to simulate drug responses and assess risks without exposing real patients to harm. This is particularly transformative for orphan diseases and pediatric studies, where physical patient cohorts are too small for traditional validation.

Digital twins have already been shown to reduce clinical trial enrolment by up to one-third, cutting months off development timelines and millions from R&D budgets. This democratises the medical device industry, allowing smaller, innovative companies to compete with established giants by proving efficacy in virtual groups before moving to targeted physical trials.

Implementation Challenges: The AI Velocity Gap and Shadow AI

Despite the technical potential, the transition to a horizontal AI ecosystem is fraught with operational friction. Organisations are discovering that the hardest part of AI is not building the model, but building the trust and the workflows around it.

The AI Velocity Gap

Research identifies a significant "AI Velocity Gap" where individuals are adopting AI tools far faster than enterprises can implement governance. Only 10% of enterprises have a cross-departmental AI rollout, while the rest remain gridlocked in governance committees and "change management debt". This gap between adoption and readiness is the defining challenge of the 2026–2031 period. While 85% of healthcare organisations are exploring AI, only 18% are actually operationally ready to deploy it in care delivery.

The Shadow AI Crisis

Frustrated by this institutional inertia, clinicians are increasingly turning to "Shadow AI", unauthorised tools used without IT approval. Shadow AI is now present in 40% of hospitals, adding significantly to breach costs and compliance risks. Over 80% of stolen patient records now come from third-party vendors rather than hospitals directly, highlighting the vulnerability of the increasingly modular software stack.

Implementation Barrier	Severity	Impact	Mitigation Strategy
Data Quality/Silos	High	Model Hallucination	Unified FHIR-native platforms
Trust/Explainability	High	Clinician Rejection	Explainable AI (XAI) & audit trails
Cybersecurity	Moderate	$7.4M avg breach cost	Zero-Trust Architecture & Air-gapping
Org Inertia	Moderate	"Execution Paralysis"	CEO-sponsored AI transformation squads

The Geopolitical and Macroeconomic Context: 2030 Outlook

By 2030, the healthcare industry will have moved from augmenting existing systems with AI to becoming "AI-first." In this vision, care is proactive, automated, and robot-enabled. The "digital dividend" promised by the federal EHR incentive programs will finally be realised as interoperable APIs and horizontal enabling layers streamline the flow of data across the entire patient journey.

The geopolitical balance of power is also being reshaped by AI. Nations with significant capital and clear national AI strategies, such as the US and certain Middle Eastern nations with "Vision 2030" plans, are emerging as leaders in AI innovation and commercialisation. Conversely, the market is expected to consolidate as the "AI bubble" matures, with hyperscalers and deeply embedded vertical platforms surviving while "thin wrapper" startups are weeded out.

Strategic Imperatives for the Next 5 Years

For industry stakeholders, the next five years will be a race to own the "context" and the "orchestration" rather than the raw intelligence.

For Incumbent Software Providers

Incumbents must recognize that their historical moats are melting. The strategic move is to pivot from being a repository of information to becoming the "control unit" for clinical workflows. This requires adopting modular, API-first architectures and integrating generative moats that compound value through expert-labeled outcomes and real-world feedback loops. Those who remain "invisible plumbing" will face relentless pricing pressure and eventual displacement by AI-native overlays.

For Healthcare Providers and Systems

Health systems must "act local but think enterprise-wide." This involves laying the groundwork for a modular, connected AI architecture that can link workflows across multiple domains—radiology, pathology, and administration. By treating AI as a "capacity multiplier," leaders can refocus human labor on "exceptions, judgment, and complex care," while the horizontal AI layer handles information synthesis and routine coordination.

For AI Startups and Innovators

The "Thin Wrapper" era is over. To survive, startups must build deeper, not wider. They must anchor their AI in proprietary clinical context and tie it directly to patient outcomes. The most durable companies will combine defensive moats (regulatory certification and compliance) with generative moats (compounding data signals) to widen the gap over competitors who only have access to shared horizontal models.

The horizontal enabling layer is the most significant structural change to healthcare technology in a generation. By melting traditional barriers to entry and eroding once-impenetrable moats, it is creating a more modular, efficient, and ultimately proactive healthcare ecosystem. The organisations that thrive in this environment will be those that embrace the shift from intelligence-as-a-product to orchestration-as-a-platform.

Nelson Advisors > European MedTech and HealthTech Investment Banking

Nelson Advisors specialise in Mergers and Acquisitions, Partnerships and Investments for Digital Health, HealthTech, Health IT, Consumer HealthTech, Healthcare Cybersecurity, Healthcare AI companies. www.nelsonadvisors.co.uk

Nelson Advisors regularly publish Thought Leadership articles covering market insights, trends, analysis & predictions @ https://www.healthcare.digital

Nelson Advisors publish Europe’s leading HealthTech and MedTech M&A Newsletter every week, subscribe today! https://lnkd.in/e5hTp_xb

Nelson Advisors pride ourselves on our DNA as ‘Founders advising Founders.’ We partner with entrepreneurs, boards and investors to maximise shareholder value and investment returns. www.nelsonadvisors.co.uk

#NelsonAdvisors #HealthTech #DigitalHealth #HealthIT #Cybersecurity #HealthcareAI #ConsumerHealthTech #Mergers #Acquisitions #Partnerships #Growth #Strategy #NHS #UK #Europe #USA #VentureCapital #PrivateEquity #Founders #SeriesA #SeriesB #Founders #SellSide #TechAssets #Fundraising #BuildBuyPartner #GoToMarket #PharmaTech #BioTech #Genomics #MedTech