The Model Is Not the Asset

The Exception

One sector never got the productivity dividend.

Technology has lowered the unit cost of almost everything it has touched. Healthcare is the conspicuous exception. US health spending reached $5.3 trillion in 2024, growing 7.2% and rising to 18.0% of GDP — and, as CMS records, growth again outpaced the expansion of the wider economy. Decades of clinical software arrived without bending that line.

The conventional explanation is that care is irreducibly human. The more precise explanation is that healthcare has bought tools, and tools do not change cost structures. A better scribe, a faster diagnostic aid, a smarter administrative layer — each makes an existing process cheaper per unit while leaving the shape of the process intact. Costs fall at the margin. The production function does not move.

The Evidence

A different proposition, and what it does not yet prove.

On 30 July 2026, Akido announced what it describes as the first AI-native health system in the United States: its ScopeAI technology deployed across roughly 100 clinics, virtual care and field medicine, with licensed clinicians retaining responsibility for clinical decisions. The company also retired “Labs” from its name, introducing Akido Medical as the clinical brand.

This is a company announcement, not an audited result, and it should be read as one. No independent evaluation of the system’s cost or clinical performance is yet in the public domain. What makes it worth an institutional reader’s attention is not the claim but the structural posture behind it: intelligence positioned inside the operating system through which care is delivered, rather than alongside it.

The significance is not the branding. It is where the intelligence sits.

The Mechanism

Tools lower the cost of a task. Architecture changes the cost of scale.

The economic signature of an AI-native system is a shift in the composition of cost, from variable to fixed. Conventional care scales linearly: each additional patient requires a proportional increment of clinical labour, the scarcest and least elastic input in the system. That linearity is the reason the cost curve has never bent.

Software does not scale that way. Once built, the marginal cost of delivering an AI-mediated function can fall sharply relative to clinician-led delivery. Where intake, triage, documentation and routine workflow are absorbed into the architecture, scarce clinical judgement concentrates on the cases that genuinely require it. Output per clinician rises not because anyone works faster, but because lower-value work leaves the clinician entirely.

This is also where access and economics converge. The populations conventional systems serve worst are those where delivery cost is highest relative to reimbursement. An architecture that lowers delivery cost changes the arithmetic of who can be served viably — an economic claim before it is a social one.

The Allocation Question

Underwrite the system, not the model.

Model performance is a filter, not an investment thesis. Benchmark advantages narrow as capability diffuses, and in a field where frontier performance is contested quarterly, an edge measured in benchmark points has a short half-life.

Durable value is more likely to sit in the system built around the model: distribution, clinical workflow, proprietary data, patient relationships, reimbursement pathways, regulatory integration and demonstrated outcomes. These compound, and they resist replication in ways model performance does not. Capital that prices only the algorithm risks mispricing the enterprise.

The broader signal is structural. When operators begin defining themselves as systems rather than tools, competition moves from model capability to delivery architecture — and the returns move with it.