The US science funding overhaul, read from Sri Lanka

The US science funding overhaul now moving through Washington is not really an argument about budgets. It is an argument about who holds the money and, downstream of that, who is allowed to read the results. Michael Kratsios, the White House science chief, made the case for it in The Economist's By Invitation section on 13 August 2026.

I read this from the far end of the pipe: the end where you download a PDF for free at 11pm and build something with it by Sunday.

The op-ed sits behind The Economist's paywall. The report it argues from — "Science: A New Golden Age" — is public on whitehouse.gov, published July 2026. That document is what I'm working from here, not the paywalled piece.

---

🔍 What is actually being proposed

Strip the framing away and the report makes one structural bet: move money out of institutions and into people and missions. It positions itself as the first serious rethink of the American research system since Vannevar Bush's Science: The Endless Frontier in 1945, roughly 80 years of accumulated default.

Dimension	Current default	What the report pushes for
Where the money lands	Research universities, carrying large indirect-cost overheads	Individual scientists and new mission-driven organisations
How proposals get picked	Traditional peer-review panels	More flexible grant mechanisms, deliberate room for unconventional work
Field emphasis	Life-sciences heavy	Physical sciences and AI up, life sciences down
Pace	Multi-year drift	Federal R&D agencies asked for action plans within 90 days

The implementation lever is a FY2028 R&D Priorities Memorandum from OMB director Russ Vought and Kratsios, which pushes agencies to act rather than deliberate. Critics, including Representative Zoe Lofgren, warn it weakens fundamental research and hands politics a bigger role in who gets funded. Both readings can be correct at once. Slow systems are genuinely slow; the thing that makes them slow is often the thing that makes them impartial.

---

🌐 Open publication is infrastructure, and it is not on anyone's balance sheet

Here is the part of this debate that gets almost no airtime, and it is the part that decides whether a student in Kandy can still teach themselves modern machine learning for the price of electricity.

The university system has one property that matters enormously to people outside it, and it has almost nothing to do with quality: its output is publishable by default. Careers there run on citations. A paper that nobody can read is worth nothing to the person who wrote it, so it gets posted, preprinted, and given away.

Company research runs on the opposite incentive. It runs on NDAs, on release timing, and on whether publishing helps or hurts a product roadmap.

I cannot give you a percentage of my own work that stands on freely published research, because I have never tracked it. The honest answer is close to all of it. Every AI tool on this site sits on top of transformer literature that arrived as a free arXiv preprint, evaluation methods someone published rather than sold, and datasets released because releasing them was a professional asset.

Key takeaway: The real export of American universities to the developing world is not graduates or grants. It is free, readable, citable output. Route more of that research through private companies and the free tier of global knowledge quietly gets smaller, with no announcement and no bill.

That is not an argument that the current system is good. It is an argument that "less overhead" and "less publishing" are the same lever pulled from different ends, and only one of them appears in the report's cost column.

---

⚡ What a field-emphasis shift means for what you can learn free

Funding priorities are a leading indicator of what is free to build on about two years later. If physical sciences and AI go up while life sciences come down, the downstream effects reach us as changes in supply.

• More AI research money does not automatically mean more open AI research. Recent history runs the other way: the better-funded a model becomes, the more likely its weights and training details stay private.
• Mission-driven organisations publish on mission timelines, not academic ones. Useful work, later access.
• Physical sciences have a strong open-preprint culture already. That part is likely fine.
• Life sciences pullback hits the open dataset supply hardest, because biology's public datasets are disproportionately grant-funded rather than product-funded.

The practical version for a small builder: assume that over the next few years, more of what you need will be behind an API key and less of it will be a PDF you can read at 2am. Budget accordingly, in money and in time.

---

🛠️ Five things I would actually do about this

None of this is in our control. The response is therefore boring and defensive, which is usually what good responses look like.

1. Mirror what you depend on. If a paper, dataset, or model card matters to a project you maintain, keep a local copy with the retrieval date. Links rot; policy shifts accelerate the rot.
2. Prefer open weights where the quality gap is small enough to live with. Not on principle. On supply risk.
3. Price the paid path before you need it. If your fallback for a disappearing open model is a commercial API, know the number in advance. Our AI model comparison tool exists for exactly this kind of "what does the alternative cost me" question.
4. Cite with dates, not just URLs. Anything sourced from a government or institutional site in a policy-volatile period should carry the date you read it. This site's calculator modules carry a LAST_VERIFIED constant for the same reason.
5. Publish your own work openly anyway. A Sri Lankan engineer publishing a working method is a rounding error in the global literature and a meaningful contribution to the local one. The cost is near zero and the compounding is real.

---

💡 What this means for you

If you are a student or a small-team builder here, the American funding fight is not your fight and you have no vote in it. What you have is exposure. Your ability to learn expensive things cheaply has been subsidised for decades by a system that pays researchers in reputation, and reputation requires publishing.

Any reform that pays researchers in equity or contracts instead will produce results. It will just produce fewer of them in a form you can read. Watch for that specific signal over the next two years: not the funding totals, but whether the interesting work still shows up as a preprint you can open without logging in.

Plan for a slightly more expensive world and be pleasantly surprised if it does not arrive.

Sources: The Economist, By Invitation, 13 August 2026 · "Science: A New Golden Age", OSTP, July 2026 · Hacker News discussion