AI computing power is becoming a tradable commodity. What does that mean for healthcare?

Most of the conversation about AI in healthcare is about what the models can do: better predictions, more efficient clinical workflows, earlier outreach to patients at risk. But a development outside healthcare may matter just as much for how the industry adopts and pays for AI.

Computing power itself is becoming a financial commodity.

CME Group scheduled the first regulated futures contracts tied to the rental price of Nvidia's H100 and B200 processors to begin trading this month, and Intercontinental Exchange has announced contracts of its own. Specialized infrastructure companies, known as neoclouds, are investing billions in the processors needed to power AI, while new marketplaces are making that capacity easier to buy, sell, and hedge. These markets are appearing for a reason: the price moves. One index tracked the hourly rental price of Nvidia's newest chips rising 48% between February and April of this year.

Much like electricity or natural gas, computing power is developing its own financial ecosystem.

For healthcare executives, particularly those managing populations under financial risk, this raises an important question: As intelligence becomes more accessible and its underlying costs become more transparent, are we evaluating AI investments against the right measures of value?

From Computing Power to the Economics of Healthcare Intelligence

The emerging market for AI computing changes how intelligence is produced and purchased.

AI requires substantial processing power, both to develop models and to run them. The latter process, known as inference, occurs whenever an existing AI model produces a prediction, recommendation, or response.

For healthcare organizations, inference is becoming an important operating expense, and one whose price can move. When a vendor charges by usage, an organization's cost is tied, directly or indirectly, to a commodity price it doesn't control.

Consider a risk-bearing organization managing 100,000 patients. AI could be deployed to identify patients at risk of hospitalization, predict readmissions, detect gaps in care, and support clinical decision-making.

Each application consumes computing resources, software costs, clinical oversight, and operational capacity. Yet their financial and clinical value may differ dramatically.

A sophisticated algorithm that predicts readmissions with impressive accuracy may generate little value if the organization cannot effectively intervene. A relatively simple assessment that identifies the patients whose readmissions can actually be prevented may produce substantial savings.

The distinction is between prediction and impact.

In my experience leading population health operations within a full-risk healthcare organization, the challenge was rarely a shortage of data or predictive scores. It was determining which interventions would meaningfully change patient outcomes, and then executing them consistently at scale.

The most valuable intelligence is not necessarily the most computationally sophisticated. It is the intelligence that changes what happens to a patient.

Why Value-Based Care Changes the AI Investment Equation

Value-based care provides a particularly useful framework for evaluating AI because financial performance is tied to outcomes, utilization, and the efficient deployment of resources.

Under traditional fee-for-service reimbursement, technology may create revenue by supporting additional services. Under financial risk, technology must demonstrate that it improves outcomes, reduces avoidable utilization, or generates operational efficiencies sufficient to justify its cost.

Consider readmission prevention.

An AI platform might successfully identify thousands of patients at elevated risk. But risk identification alone does not prevent hospitalization.

The greater opportunity lies in determining which patients are most likely to benefit from intervention, understanding the barriers driving utilization, and directing care teams toward actions that can change outcomes.

The same principle applies to chronic disease management, care transitions, and administrative automation.

Healthcare organizations should evaluate AI against three questions:

Is the intelligence actionable? Does it identify an opportunity that can realistically be changed?

Can the organization execute? Are the workflows, clinical resources, and operational infrastructure in place to act on the insight?

Does it produce measurable value? Can improvements in outcomes, utilization, or productivity be credibly attributed to the intervention, net of its total cost?

These questions should be central to AI procurement, particularly as the number of available applications expands.

What Value-Based Care Already Knows About Derivatives

The development of compute derivatives introduces an interesting possibility. It also holds up a mirror, because value-based care has been built on similar instruments for years. We just call them something else.

A capitation contract works much like a swap. A medical group gives up variable fee-for-service revenue in exchange for a fixed payment per member per month, and from that point its margin depends on what care actually costs. Stop-loss coverage works like an option: the organization pays a premium so a handful of catastrophic cases can't sink the year. Risk corridors set a floor and a ceiling on gains and losses. Finance teams in risk-bearing organizations already think like traders. The commodity they trade is medical cost.

AI adds a second price to that position. When AI is bought on usage, by the transaction, the token or the compute hour, an organization carries two exposures at once: the cost of care, and the cost of the intelligence used to manage it. Over the long run, the cost of a single prediction is falling quickly. In the short run, the price of the computing behind it can swing sharply within a single quarter.

Large healthcare organizations may eventually face sufficient AI infrastructure costs to benefit from pricing arrangements that protect against those swings. For most, the practical step today is simpler: know who carries the price risk in each AI contract. If a vendor's computing costs rise next year, does your price follow?

Two other ideas from these markets translate directly.

The first is basis risk, the gap between the price a hedge tracks and the price you actually pay. Healthcare AI has its own version. A model validated on another organization's patients is a contract written on someone else's index. It may perform well in the published study and drift once it meets your members, your network, and your community. Testing a model against your own population before scaling it is the equivalent of checking the basis before you hedge.

The second is how a contract settles. The compute futures ICE has proposed settle on the average price across the full contract term rather than a single day, so one volatile week can't distort the result. AI in healthcare deserves the same discipline. A pilot that looks strong for one quarter in one clinic is a single day's price. The fair measure is performance across a full year and every site.

The most immediate opportunity, however, is contractual rather than financial.

What if healthcare AI companies increasingly structured commercial agreements around verified improvements in clinical outcomes, reductions in avoidable utilization, or operational savings?

Rather than paying exclusively for software licenses, transactions, or computing capacity, healthcare organizations could negotiate agreements that align vendor economics with measurable performance. Paying an AI vendor by the token is much like paying a specialist by the visit. It rewards volume.

Such arrangements would require credible measurement, appropriate risk adjustment, patient safety safeguards, and clear accountability. But the underlying concept is already familiar.

Value-based care has spent years developing financial models that reward outcomes rather than activity. Applying similar principles to AI procurement would represent a logical evolution.

The Risk of Repeating Healthcare's Technology Mistakes

Healthcare has invested billions of dollars in information technology without consistently realizing the operational transformation those investments promised.

Electronic health records, population health platforms, and predictive analytics have dramatically increased access to information. Yet many organizations still struggle with fragmented workflows, clinician burden, and difficulty translating insights into measurable improvements.

AI could accelerate progress, or it could magnify the problems we already have.

As inference becomes less expensive, organizations may generate substantially more predictions, alerts, and recommendations than their clinical teams can realistically act upon. When predictions become cheap, the scarce resource is the time and attention of nurses, pharmacists, social workers, and physicians. Every alert spends some of it.

Lower computing costs do not automatically create lower medical costs.

Increasing the volume of algorithmic output without redesigning care delivery could create additional operational burden.

The objective should not be to introduce AI into as many workflows as possible. It should be to identify where intelligence can change clinical decisions, improve resource allocation, and produce measurable results.

Technology implementation is not the same as operational transformation.

The Future Belongs to Organizations That Understand Both Technology and Risk

The emergence of financial markets for AI computing signals a broader transformation.

Artificial intelligence is becoming an accessible input into business operations. Over time, the cost of generating many forms of intelligence may decline, even as demand grows.

That will change how healthcare organizations evaluate technology investments.

Competitive advantage will not come simply from having access to powerful algorithms. Many organizations will have access to similar tools.

The differentiator will be the ability to deploy those tools against the right problems, integrate them into effective clinical workflows, and measure their contribution to patient outcomes and financial performance.

For value-based care organizations, the next phase of AI adoption should be defined by disciplined investment decisions, operational accountability, and demonstrated return on investment.

The organizations that succeed will be those that understand not only what AI can predict, but what it costs to act on those predictions, and whether doing so meaningfully changes outcomes.

The next phase of healthcare innovation will not be defined solely by how much intelligence we can generate. It will be defined by how effectively we convert intelligence into better outcomes, lower costs, and a more sustainable healthcare system.

Risk-bearing organizations already know how to price risk, share it, and hedge it. As intelligence becomes a commodity, that may be the most valuable expertise healthcare brings to AI.

Jennifer Dunphy, DrPH, MBA, MPH, is Founder and Managing Principal at The Value Care Group, advising risk-bearing healthcare organizations and their partners on medical cost performance, population health, care delivery, and operational strategy.