Healthcare’s First AI Success Is Raising Its Costs

The first AI tool to deploy at real scale inside American clinics is doing something curious: it is raising costs.

The tool is ambient documentation software, meaning AI-powered scribes that listen to patient visits and automatically generate medical notes. According to STAT News, insurers and providers have privately agreed in recent months that these tools are increasing how much doctors bill, because more detailed documentation lets them code at higher complexity levels. In a fee-for-service system, that means higher costs, not lower ones.

This is not what Andreessen Horowitz promised when the firm published its Infinite Healthcare thesis in February. The thesis argues that AI will finally break healthcare's cost curve, making care so abundant and cheap that using more of it paradoxically drives spending down. The most-cited example: a company called Function Health, which the firm said hit $100 million in annual recurring revenue in under two years, a pace that historically took a decade or more in healthcare.

The velocity story is real. Health tech companies founded since 2022 are reaching that $100 million ARR threshold at roughly the same pace as the best software companies in any sector, according to Bessemer's State of Health AI 2026 report. But velocity and cost reduction are not the same thing. And the ambient scribing experience, the first AI tool to achieve genuine clinical deployment at scale, suggests the path from adoption to savings is longer and more tangled than the compression thesis implies.

"The tension is that more utilization in healthcare has always meant more cost," said Julie Yoo, a general partner at Andreessen Horowitz who leads the firm's healthcare investments, in a podcast appearance this year. "The question is whether making care cheaper and more abundant breaks that link."

The honest answer, at least for now, is: the payment model does not exist to find out.

Healthcare is still structured around fee-for-service, which means paying for encounters, procedures, and coded complexity. That is why ambient scribing increases costs in the near term. The CMMI ACCESS program, a federal initiative, is running experiments in alternative payment frameworks for AI-enabled care: per-task, per-workflow, per-episode, and per-outcome models designed to reward efficiency rather than volume. If those models scale, the compression thesis has a plausible path to reality. If they do not, companies that scaled fast on venture capital will eventually hit a reimbursement cliff when growth slows and margins matter.

What founders and investors are actually betting on is the sequence. AI will eventually compress costs. That is the theory. But the infrastructure to pay for that outcome has not been built yet, and the first widely deployed AI tool in clinical settings is a reminder that the default path from adoption to savings runs straight through the existing billing incentives, not around them.

The companies that navigate the payment model transition fastest, not the ones that grow the fastest, will determine whether Infinite Healthcare becomes a category or a cautionary tale.