What Healthcare AI Can Learn From Vertical AI in Construction

There is a recurring debate in health AI about the relative merits of horizontal platforms — tools that work across many clinical contexts — versus vertical specialists that go deep in a single domain. The debate is not new, but the emergence of powerful general-purpose foundation models has given it new urgency. If GPT-5 or Gemini Ultra can answer clinical questions, interpret radiology images, and generate clinical notes, what is the competitive advantage of a company that does only one of those things?

The answer, increasingly, is everything. The history of AI adoption in adjacent industries — construction, legal, financial services — suggests that the most durable competitive advantages in AI-native businesses come not from the underlying model, but from the domain-specific data, workflow integrations, and institutional trust that vertical specialists build over years of deep engagement with a single customer segment.

The Construction Parallel

Consider Procore, the construction management software company, which went public in 2021 at a $9 billion valuation. Procore did not build the most sophisticated software in the world — it built software that was specifically designed for the workflows of general contractors, subcontractors, and project owners in the construction industry. The company spent years learning the specific terminology, regulatory requirements, and workflow patterns of construction projects, and built a product that fit those workflows precisely. When general-purpose project management tools like Asana and Monday.com entered the market, they could not displace Procore because they lacked this domain depth.^[1]

The same dynamic is playing out in health AI. Companies like Recursion Pharmaceuticals in drug discovery and Tempus AI in oncology diagnostics have built competitive advantages that are not primarily about the quality of their underlying AI models — they are about the proprietary datasets, the regulatory clearances, the clinical validation studies, and the deep workflow integrations that took years to build and cannot be replicated quickly by a general-purpose AI company entering the market.^[2]

The Data Moat

The most important source of competitive advantage for vertical health AI companies is proprietary data. General-purpose foundation models are trained on internet-scale text data, which includes a significant amount of medical literature and clinical content. But they are not trained on the de-identified EHR data, imaging studies, genomic profiles, and clinical outcomes data that vertical health AI companies have assembled through years of partnerships with health systems and research institutions. This proprietary clinical data is the raw material for training models that perform better in clinical settings than general-purpose models — and it is not available to competitors at any price.^[3]

The Implication for Investors and Founders

For investors evaluating health AI companies, the key question is not "how good is the model?" but "how deep is the domain moat?" The companies that will generate the most durable value are those that have assembled proprietary clinical datasets, built deep EHR integrations, obtained FDA clearances for specific clinical indications, and established the institutional trust with health system customers that makes switching costs high. These are slow, expensive things to build — which is precisely why they are valuable.

References

1. a16z. "The Case for Vertical AI." 2025. https://a16z.com/the-case-for-vertical-ai/
2. Rock Health. "2025 Year-End Funding Report." January 2026. https://rockhealth.com/insights/2025-year-end-digital-health-funding-report/
3. Stanford HAI. "AI Index Report 2026." 2026. https://aiindex.stanford.edu/report/