The Patent Bargain

The problem

The United States government is the largest funder of basic research on earth. The National Institutes of Health alone spends roughly $48 billion a year, and total federal research and development runs to about $200 billion.

This money buys the part of invention that no private investor will touch: the decade of unglamorous mechanism work before anybody knows whether there is a product. A well-known study in the Proceedings of the National Academy of Sciences traced the drugs approved by the FDA over a recent seven-year span and found federally funded research behind every single one.

Then the invention is patented, licensed to a firm, and sold back to the public that paid for the science, at whatever the market will bear.

Under the Bayh-Dole Act of 1980, universities and contractors keep title to inventions made with federal money. The government retains two things: a royalty-free licence for its own use, and march-in rights under 35 U.S.C. § 203, which allow it to require additional licensing when the invention is not being made available on reasonable terms.

March-in rights have been petitioned for repeatedly, over drugs including ritonavir and enzalutamide. They have been exercised exactly zero times in forty-five years.

So the public's stake in an enterprise it funds at $200 billion a year is: a licence it rarely uses, and a right it has never once exercised.

Meanwhile universities collect around $3 billion a year in licensing revenue from this pipeline. That is not scandalous, it is the system working as designed, and the design has real successes to its name. The scandal, if there is one, is the party that is missing from the cap table. The entity that funded the research holds nothing.

There is a second problem sitting on top, which is that the patent being licensed is often not one patent. Adalimumab accumulated something on the order of 130 patents and held off American biosimilar competition until 2023, five years after European patients got it. That is not a reward for invention. That is a reward for filing.

The proposal

Where federal money funds an invention, the United States takes a small royalty and a warrant, deposited in the American Permanent Fund. Bayh-Dole's ownership rules stay exactly as they are.

Let that second sentence carry weight. This proposal does not propose taking title back. That experiment was run before 1980 and it failed badly.

How it would work

1. A standing royalty. Inventions arising from federally funded research carry a 0.5 to 1 percent royalty on net sales above a revenue threshold, payable by the licensee, running for the life of the patent. Small enough not to change a licensing decision. Large enough to compound.

1. A warrant, not a share. Where a startup licenses federally funded IP, the funding agency receives a non-voting warrant for a small equity position, exercisable only on liquidity, assigned immediately to the Fund. This mirrors № 020 and № 002. Non-voting is essential: the government funds the science and does not sit in the boardroom.

1. Receipts to principal. All of it flows to the American Permanent Fund under the two locks of № 023. None of it returns to the funding agency's budget, for the reason set out at length in № 035: an agency that profits from its own grants is an agency with a corrupted judgment.

1. A threshold, so it never touches the small. No royalty below a meaningful revenue floor, and none at all on university licensing income, non-profits, or defence-unique technology. This is a claim on commercial success, not a tax on tech transfer.

1. March-in, made real or made honest. Publish a binding framework specifying the circumstances in which § 203 will and will not be used, and then use it when those circumstances are met. A right that is never exercised is not a safeguard, it is decoration, and its existence has been used for four decades to argue that the public interest is already protected. Note that this proposal deliberately does not make price the trigger. Pricing is the subject of № 043, and using march-in as a price-control tool is a fight that would swallow everything else here.

1. Patent quality, separately. Fund the USPTO to give examiners materially more time per application, and end terminal-disclaimer stacking that lets a thicket of obvious variants extend an exclusivity beyond the invention that earned it. A patent system that grants too easily produces the thicket described above, and no royalty design fixes that.

The numbers

American pharmaceutical revenue attributable in part to federally funded research is enormous, and any estimate involves a judgment about attribution. Take a deliberately conservative base: $150 billion a year in US net sales of products whose foundational research was federally funded.

A 0.75 percent royalty on that base is roughly $1.1 billion a year.

Extend the same mechanism across federally funded inventions outside biomedicine, materials, semiconductors, software, agriculture, and call the total $1.5 billion a year at maturity, phased in over a decade as the patent estate turns over. Warrants add an irregular tail that could be much larger in a good year and nothing in a bad one, exactly like № 020.

At a 5 percent real return, $1.5 billion a year compounds to about $20 billion of corpus in a decade, paying $1 billion a year, permanently. Per citizen, roughly $3 a year.

Three dollars, and the accounting confession this catalog keeps making applies again.

What makes this one worth the page is the asymmetry. The public currently carries the entire cost of the failures, since most federally funded research leads nowhere, which is what basic research is, and captures none of the successes. Every venture investor on earth would recognise that portfolio as insane. Taking a fractional share of the winners is not a new burden on industry; it is the public finally holding the upside of a bet it has been placing, unhedged, since 1945.

The honest objections

"Bayh-Dole works. Before 1980 the government held 28,000 patents and licensed under 5 percent of them. Do not touch it." The correct starting point, and the reason this proposal leaves title alone. The pre-1980 regime, in which Washington held inventions and nobody could get an exclusive licence to commercialise them, left an enormous amount of publicly funded science sitting in a drawer. Bayh-Dole fixed that and the fix worked. A royalty and a non-voting warrant do not restore federal title, do not affect who may licence, and do not give the government a say in commercialisation. If any part of the drafting starts to, that part is wrong.

"A royalty will make firms refuse federal funding, or refuse to license federal IP." The scale argument answers this and it is the whole reason for the rate in item 1. A firm walking away from a licence over 0.75 percent of net sales is a firm that had no viable product. Compare the royalties universities themselves routinely charge, which run in the low single digits and sometimes far higher, and which have not collapsed tech transfer. If the rate ever rises to a level that changes licensing behaviour, it has stopped doing its job, which is the same test applied to compute in № 031.

"Universities will simply pass this cost through, or lose licensing revenue they depend on." Item 4 exempts university licensing income specifically, so the charge sits with the commercial licensee. Some of it will be negotiated back into licence terms, which means universities bear part of it indirectly. That is a real cost to institutions already under financial strain, and the honest response is that the exemption limits it rather than eliminating it.

"Attribution is a nightmare. Nearly every invention touches some federal research somewhere." The most practically difficult objection. If the trigger is drawn broadly, the royalty becomes a general tax on innovation. The line has to be the existing one: inventions with a Bayh-Dole disclosure obligation, meaning those conceived or reduced to practice under a federal funding agreement. That is a bright, administrable, already-tracked category, and it deliberately excludes the vast penumbra of research that merely benefited from the general stock of public knowledge. It also means the estimate above understates the public's real contribution, which we accept in exchange for a rule that can actually be applied.

"This is a tax on medicine that will land on patients." The incidence question is genuine, and a 0.75 percent royalty on a monopoly-priced product is very likely absorbed in margin rather than passed through, because a monopolist is already charging the profit-maximising price and a small per-unit cost does not move it much. That is a theoretical argument and reasonable economists would want to see it tested. It is also why the pricing question belongs in № 043, where it can be addressed directly, rather than being smuggled in here.

Sources

• NIH appropriation of approximately $48 billion; total federal R&D obligations near $200 billion (nih.gov; NSF National Center for Science and Engineering Statistics)
• Cleary, Beierlein, Khanuja, McNamee and Ledley, "Contribution of NIH funding to new drug approvals 2010–2016," PNAS (2018)
• Bayh-Dole Act, Pub. L. 96-517 (1980); march-in rights at 35 U.S.C. § 203; petitions concerning ritonavir (2004) and enzalutamide (2016, 2023), all declined
• University licensing income of roughly $3 billion annually; AUTM Licensing Activity Survey (autm.net)
• Adalimumab patent estate and delayed US biosimilar entry to 2023 versus European entry in 2018; I-MAK and FTC filings
• USPTO examiner time-per-application analyses; Frakes and Wasserman on examination time and grant quality
• Proposals № 002 (The Founders' Pledge Act); № 020 (The Sovereign Equity Act); № 023 (The Debt Covenant); № 031 (The Machine Dividend); № 035 (The Antitrust Dividend); № 043 (The Formulary)