A public-policy simulationThe American AI Healthcare Compact
Interactive · Policy simulation · November 2026

What if every AI data center built in America helped pay for healthcare for those without it?

America is starting the biggest buildout of physical infrastructure since the interstate highways. Gigawatt-scale campuses are going up in Texas, Virginia, Louisiana, Ohio and Arizona. At the same time, tens of millions of Americans have no health insurance. This page tests one possible grand bargain between those two facts: faster AI infrastructure in exchange for a permanent contribution to their care.

28.0million

Americans projected to be uninsured in 2027. Each dot below is 1 million people. The filled dots are the people the Compact would cover in 2035 under the default scenario.

● covered in 2035 · ○ still uninsured
§ 01 — THE BARGAIN

Speed for the builders. Care for the uninsured.

The idea is a simple trade. Washington offers what the AI industry says it needs most, which is speed: permits, power and places to build. In return, every facility pays into a trust fund that covers uninsured Americans and builds hospitals where care is scarce.

Diagram of the Compact On the left, the government offers fast-track permits, grid priority, federal sites and regulatory certainty. On the right, AI facilities pay a power levy, a build contribution and a compute revenue share. Both flow into the Compact Trust Fund, which pays for coverage for the uninsured and for community hospitals. AMERICA OFFERS THE AI INDUSTRY CONTRIBUTES THE Compact TRUST FUND PAYS FOR Coverage for the uninsured BUILDS Community hospitals

America offers

  • Fast-track federal permits
  • Priority grid interconnection
  • Federal land and sites
  • Twenty years of regulatory certainty
↓ in exchange for ↓

The AI industry contributes

  • A levy on every megawatt-hour
  • A share of construction cost
  • A share of compute revenue
↓ Compact Trust Fund ↓

Which pays for

  • Coverage for the uninsured
  • New community hospitals

Why data centers?

These are among the largest and most location-bound investments ever made. A single campus can draw as much power as a mid-sized city, and you can’t offshore a substation that’s already built. That makes the revenue unusually predictable.

Why healthcare?

The uninsured already get care, often late and in emergency rooms, and hospitals absorb tens of billions of dollars a year in unpaid bills. Coverage moves that cost into prevention and moves the bill off of everyone else’s premiums.

Why a bargain?

Permitting and grid queues are the industry’s binding constraints. A trade that speeds them up gives both sides a reason to stay in the deal, which a one-time tax doesn’t.

§ 02 — SIMULATE

Write your own Compact.

Choose how fast America builds and how much each facility pays. The model projects the Trust Fund from 2027 to 2040 and compares it with the rising cost of covering everyone who lacks insurance. Start from a preset or move any slider.

The buildout
Gigawatts of new AI data-center load added each year, starting from about 20 GW in 2026.
What each facility pays
Per megawatt-hour consumed. Industrial power costs roughly $80/MWh.
A one-time share of construction cost, at about $40B per gigawatt including chips.
A share of the revenue each gigawatt earns, assumed at about $10B a year.
The healthcare side
26.4M in 2023. Projections rise after the 2025 Medicaid and subsidy changes.
A Medicaid-style benefit for a mostly working-age group. Grows 4% a year.
About $250M per 100-bed community hospital.

Projected outcome

0%
of uninsured Americans covered in . That’s people.
050%Universal
$0B
Trust Fund that year
$0B
Cost of full coverage
0
Hospitals funded, cumulative
0%
Of AI compute revenue paid in
0 GW
Installed AI capacity
0%
Of U.S. electricity use that year
Compute revenue share Power levy Build contribution Redirected unpaid care Cost of covering everyone
How this is calculated
§ 03 — ONE GIGAWATT

What a single gigawatt pays for.

A gigawatt is the scale of a frontier AI campus. It runs about as much power as 750,000 homes. Here is what one gigawatt contributes in a year under the settings you chose above.

01 / ENERGY
6.6M MWh

Consumed per year at 75% average utilization.

02 / CONTRIBUTION
$0M

Paid into the Trust Fund each year: power levy plus revenue share, with the build contribution spread over 10 years.

03 / COVERAGE
0 people

Uninsured Americans fully covered for a year by that contribution.

04 / OR, IN CONCRETE
0 hospitals

100-bed community hospitals that one year of contributions could build instead.

§ 04 — WHERE IT LANDS

The data centers and the uninsured are often in the same places.

Many of the states drawing the most AI investment, including Texas, Georgia, Louisiana and Arizona, also have some of the highest uninsured rates. In this model, hospital funding goes to states based on how many uninsured residents they have, with an extra share for states that host the data centers.

Uninsured rate 3% → 16%+ Major AI data-center hub +n = hospitals funded
§ 05 — THE HARD PARTS

What the simulation can’t settle.

A model is only as honest as what it leaves out. Here are the strongest arguments on each side.

For

The revenue base is enormous and fixed in place.

Hyperscalers are spending hundreds of billions of dollars a year on U.S. infrastructure. A contribution that’s small next to that spending can still be large compared with the cost of coverage. Facilities that are already built can’t move to another country.

Against

Costs get passed on.

Levies can show up in cloud prices, slower hiring or smaller buildouts. If the fees make U.S. sites less competitive than Canadian, Nordic or Gulf sites, some new capacity may be built abroad instead.

For

It recovers money the system already spends.

Hospitals already absorb billions in unpaid care, and that cost is shifted into everyone else’s premiums. Coverage turns emergency-room spending into primary care and moves it out of hidden subsidies.

Against

Earmarked taxes are fragile.

Tying a healthcare entitlement to one industry’s revenue makes coverage depend on that industry. If AI demand shrinks or efficiency gains cut power use, the fund shrinks too, and the people it covers still need care.

For

The grid is the bottleneck either way.

Interconnection queues run for years. If the Compact really speeds up permits and power, the industry may come out ahead even after paying in. That’s what makes it a bargain rather than just a tax.

Against

Fast-tracking has costs of its own.

Faster permits can mean weaker environmental review, more strain on local water and higher residential power rates. The communities that host the facilities carry those costs, and they aren’t always the ones who benefit.

§ 06 — METHODOLOGY

How the numbers work.

This is a simple, transparent model meant for exploration, not a budget score. Each year it projects installed AI capacity, applies the three contributions and compares the result with the cost of covering every uninsured person.

AssumptionValueBasis
AI data-center load, 202620 GWMidpoint of utility and industry estimates for AI-dedicated U.S. load
Utilization75%Training and inference fleets running near capacity
All-in build cost$40B / GWFacility, power and accelerators. Industry estimates range from $35B to $60B
Compute revenue$10B / GW / yrAbout 1M accelerator-equivalents at blended cloud rates
Uninsured, 202326.4MU.S. Census Bureau, CPS ASEC
Unpaid care for the uninsured~$42B / yrKFF estimate of uncompensated care
Cost growth4% / yrIn line with CMS national health expenditure projections
Community hospital$250MAbout 100 beds. Rural critical-access hospitals cost far less
U.S. electricity use~4,200 TWhU.S. Energy Information Administration, 2026. Grows 2% a year
State uninsured ratesACS 2023Rounded. Data-center hubs are illustrative
  1. Capacity. GW(y) = 20 + build × (1.25 if fast-track) × (y − 2026)
  2. Power levy. GW × 8,760 h × 75% × $/MWh
  3. Build contribution. new GW × $40B × share, counted in the year built
  4. Revenue share. GW × $10B × share
  5. Hospitals. The hospital share of the fund ÷ $250M. Assigned to states by number of uninsured residents, with a 50% bonus per data-center tier.
  6. Coverage. covered = min(1, coverage fund ÷ (full cost − redirected unpaid care)). Unpaid-care savings grow with the share of people covered.
  7. Grid share. Annual AI consumption ÷ U.S. consumption (4,200 TWh in 2026, growing 2% a year). Above 25%, the KPI turns red as a feasibility warning.

Not modeled: behavioral responses to the levy, capacity moving offshore, federal and state Medicaid interactions, administrative costs, hospital operating costs, or the value of faster permitting to the industry. Treat the outputs as orders of magnitude.