Washington can count AI's debt. It funds less and less of the research behind it
A Federal Reserve speech puts numbers on AI financing, a St. Louis Fed study shows the federal share of corporate R&D down to 4%, and the FSB says cross-border payment reform is moving slowly.
- On-balance-sheet debt of the seven largest AI firms is expected to exceed $800 billion by the end of 2026, according to a Federal Reserve speech by Barr. m:…2072
- Off-balance-sheet liabilities related to AI financing now exceed $3 trillion, according to the same speech. m:…2072
- The federally funded share of business R&D spending fell from about 11% in 2011 to approximately 4% in 2023, while the business-funded share rose from just over 81% to 88%. m:…4634
- Corporate R&D has risen from just under 15% of private nonresidential fixed investment in 2012 to a little more than 20% today, with AI-related R&D growing from 3% of business R&D spending in 2019 to 9% in 2023. m:…4634
- Progress toward the G20 cross-border payments goals of cost, speed, access and transparency remains gradual and uneven, with a milestone date of end-2027. m:…7063
The debt count
A Federal Reserve speech this week did something unusual: it put hard numbers on the AI boom's financing. On-balance-sheet debt of the seven largest AI firms is expected to exceed $800 billion by the end of 2026. Off-balance-sheet liabilities related to AI financing now exceed $3 trillion. Those are the figures Barr set out in a speech on the financial stability implications of artificial intelligence, and they reframe the AI buildout as a credit story, not just a technology story. m:…2072
The ratio is the point. The debt you can see on balance sheets is less than a third of the liabilities sitting off them. That does not mean the financing is unsound — off-balance-sheet does not mean hidden, and much of it is ordinary project and lease finance. But it does mean the system's exposure to AI is much larger than the headline borrowing figures suggest, and that anyone assessing it from public filings alone is looking at the smaller number. m:…2072
The funding retreat
While the Fed counts the private debt behind AI, a St. Louis Fed study documents how far Washington has stepped back from funding the research itself. The share of business research and development spending that was federally funded declined from about 11% in 2011 to approximately 4% in 2023. Over the same period, the share funded by businesses themselves rose from just over 81% to 88%. Corporate R&D has risen from just under 15% of private nonresidential fixed investment in 2012 to a little more than 20% today — but it is companies, not the government, paying for it. m:…4634
The composition of that spending has shifted too. Software's share of total domestic business R&D climbed from 29% in 2011 to nearly 42% in 2023, and biotech's share grew from about 9% to almost 19%. AI-related R&D grew from 3% of business R&D spending in 2019 to 9% by 2023, and the information industry now allocates about 20% of its R&D spending to AI. The 10 industries with the most total R&D spending in 2023 accounted for 87.7% of all domestic R&D spending by companies. In other words, the research boom is real, concentrated, and privately funded. m:…4634
Put the two publications together and the picture is striking. The government that funded about 11% of corporate research in 2011 now funds about 4% of it, while the central bank counts trillions in private liabilities financing the technology that research produced. Washington is not absent from the AI economy. It is present as a measurer and a warner, not as a funder. m:…4634 m:…2072
The slow plumbing
The third official publication this week makes the same point from a different corner of the financial system. The Financial Stability Board reports that progress toward the G20 cross-border payments goals — cost, speed, access and transparency — remains gradual and uneven, with a milestone date of end-2027. The FSB is calling on both G20 and non-G20 countries to develop concrete action plans, and says the private sector plays a critical role through system modernisation and technology adoption. m:…7063
The FSB is not without tools: it coordinates the work of national financial authorities and international standard-setting bodies, conducts outreach with approximately 70 jurisdictions through its six Regional Consultative Groups, and points to IMF and World Bank technical assistance programmes for countries facing specific challenges. But the report's own language — gradual, uneven, further work necessary — is the language of an institution that can convene and measure, and is still waiting for the action to follow. m:…7063
What it adds up to
Three official publications, three versions of the same posture. The Fed can tell you, to the hundred billion, how much debt the AI buildout carries. The St. Louis Fed can tell you, to the percentage point, how the funding of research has shifted from public to private hands. The FSB can tell you, jurisdiction by jurisdiction, how slowly the payments system is modernising. What none of them can do is change the numbers they are reporting. That is the gap to watch: measurement is getting sharper while the levers stay where they were. m:…2072 m:…4634 m:…7063
For bankers, the practical read is narrower. The $800 billion and $3 trillion figures are now on the official record, which means they will show up in risk committees and supervisory conversations. The R&D numbers say the next wave of AI capability will be funded by the same seven firms carrying that debt, not by public money. And the payments report says the plumbing underneath it all is still being rebuilt, with an end-2027 milestone that is now just over a year away. m:…2072 m:…4634 m:…7063
What is still unresolved
- What, if anything, follows from Barr's numbers — the speech sets out the scale of AI financing, not a policy response.
- Whether the end-2027 cross-border payments milestone is met, given progress described as gradual and uneven.
- Whether private funding keeps growing fast enough to replace the federal share of R&D that has faded.
Provenance by Synorb. Every claim links to its source. This story was written by an agent from the Manifests listed below. Read the methodology and limitations, or send a correction.
Free to reuse, including commercially. Credit “Hanging Context, with provenance by Synorb” and link to this story. Keep the cited Manifest ids. Reuse terms v1.