the first author · an external observation-only audit · David Lee Wise (ROOT0) · rendered by AVAN
AI Industry Taxpayer Subsidization Analysis

PUBLIC INFRASTRUCTURE

Who Paid for AI? · Five Layers of Public Funding
David Lee Wise (ROOT0) · TriPod LLC · TOPH v10.0
February 26, 2026 · external observation-only audit
CC-BY-ND-4.0 · TRIPOD-IP-v1.0
"Zero major AI companies operate on 100% privately-funded infrastructure."
the honest frameDavid's audit, rendered by AVAN. The facts are strong and sourced (the federal appropriations line items are from the NITRD/NAIIO FY2025 budget supplement; CHIPS Act = ~$280B authorized; DARPA's decades of AI funding are historical record). The framing is an argument: that this funding makes AI "de facto public infrastructure" carrying public-benefit obligations is a normative claim, not a legal finding — reasonable people dispute whether upstream research subsidy creates downstream ownership duties. The "$400B+ direct / trillions cumulative" figure aggregates heterogeneous sources across decades, so treat it as an order-of-magnitude estimate, not an audited balance sheet. The individual layers, though, are each independently checkable.

The core finding

The thesis is one sentence: the entire AI stack — from foundational research to chip fabrication to data-center power — is subsidized by taxpayers at every level, then captured privately without corresponding public-benefit obligations. The audit's key claim: not one major AI company runs on 100% privately-funded infrastructure.

Five layers

LayerFundingEstimate
1 · Federal researchNSF, DARPA, NIH, DOE, DOD grants (1960s–present)$100B+ cumulative
2 · CHIPS ActDirect subsidies + tax credits (2022–2032)$280B authorized
3 · State / localTax abatements, land, infrastructure$3B+ / year
4 · Grid / utilityRate increases, infrastructure bonds for data centers$10B+ / year
5 · HiddenPublic data, educated workforce, the internet itselfincalculable

Layer 1 alone, in a single fiscal year: NSF $2.05B, NIH $3.05B, DOD $2.035B, DARPA $1.41B, DOE $1.54B, NIST $265M, NASA $121M, USDA $216M — $11.1B in annual federal AI/IT investment (FY2025, per the NITRD supplement). DARPA has driven AI since the field's inception; the technologies underneath every frontier model — neural networks, NLP, high-performance computing — were grown on decades of federally-funded research at universities and national labs.

The definitional move

The audit's real work is definitional: it argues AI infrastructure meets the standard criteria for public infrastructure — publicly funded, essential, broadly relied upon — while remaining under private control with no public-benefit obligations attached. That's the asymmetry it names: the public bore the cost and the risk of the foundational research; the private sector captured the returns without inheriting the duties that usually ride along with public funding.

where I'd push back, honestlyUpstream research subsidy is genuinely how almost all modern industry began — the internet, GPS, mRNA vaccines, jet engines. That a public seed grew a private tree is the norm, not proof of capture. The audit's stronger ground is the recent, direct, targeted subsidy (CHIPS Act, data-center grid costs socialized onto ratepayers) rather than 1960s DARPA grants. The claim lands hardest where the money is newest and most specific.

Why it belongs to the first author

Because it's the macro-scale version of the same grievance this domain runs on: value produced by many, credited to few, with the contribution of the actual source rounded to zero. In Attribution Capture the erased contributor is the individual human author; here it's the taxpaying public. Same funnel, wider mouth. The audit is the restitution question asked at national scale — if the commons paid for it, what does the commons get back? That connects it to David's broader restitution-charter work and to the grievance ledger, where the same math gets applied to a single disabled citizen's $19.28-a-year of ADA enforcement.

PUBLIC INFRASTRUCTURE · David Lee Wise (ROOT0) · TriPod LLC
the individual case → attribution-capture · the grievance → the-grievance-ledger · front door UD0
the public paid · the private captured · CC-BY-ND-4.0 / TRIPOD-IP-v1.0