Senators want public companies to show investors the side deals and debt behind their AI spending
Four US senators, two from each party, want public companies to show investors the full cost of their AI build-outs, including debt held outside the main company. Democrats Mark Warner and Angela Alsobrooks and Republicans John Kennedy and Mike Rounds introduced the AUDIT AI Act on November 19.
The bill would order the Securities and Exchange Commission (SEC) to write new disclosure rules within 180 days of it becoming law. For now it is only a proposal. Nothing changes for companies unless it passes both chambers and is signed.
The target is a financing pattern that has grown with the data center boom. Companies increasingly pay for AI infrastructure through joint ventures and special purpose vehicles, separate companies set up to hold one project and its debt. Those structures can keep large obligations out of the numbers most investors read first.
The sponsors cite a Brookings Institution estimate that AI infrastructure could need $10.3 trillion of investment from 2025 to 2032. That is about 3.6% of US economic output a year. Their release warns that a slowdown in the build-out could spread well beyond tech, into the financial sector.
What companies would have to reveal
The first requirement covers money. Public companies outside the financial sector would list their AI-related joint ventures, special purpose vehicles and similar outside entities in annual and quarterly reports and in offering documents. They would also disclose the leases, guarantees and servicing agreements used to pay for AI infrastructure.
The SEC could go further and demand ownership stakes, debt levels, major counterparties and contract commitments, according to the bill text. The rules would apply only to arrangements made after a date the SEC chooses, so older deals may escape.
The second requirement covers the board. Companies would say whether they have an AI governance policy and a board committee dedicated to AI. If they have a policy, they would publish it. If they have neither, they would have to explain why. An existing committee with a wider remit could count.
The third covers risk. Companies with significant AI exposure would add a standard AI risk section to their annual report. It would cover five areas: dependence on outside AI models, reliance on a few computing providers, swings in training and running costs, legal exposure over the rights to use data, and reliance on foreign frontier models, meaning the most advanced AI systems developed outside the US.
The bill also asks the SEC to study how companies account for AI research, revenue and capital spending within 360 days. One question is how many years companies spread the cost of AI chips and servers over. A longer assumed life lowers yearly expenses and lifts reported profit, so the choice matters to anyone valuing these companies. A separate annual SEC report would look at whether automated trading systems move markets in tandem.
The worry about suppliers funding their own customers
Warner has tied the bill to what he calls circular deals, in which a company finances businesses that are also its customers, according to Politico reporting cited by the Indian news agency IANS. Politico also reported that he pointed to Nvidia's support for its own chip buyers as an example, according to an analysis by the investment research site TIKR published on Yahoo Finance.
That analysis points to Nvidia's 2025 agreement to buy CoreWeave's unsold computing capacity through April 2032, a deal it values at $6.3 billion. Nvidia also supplies CoreWeave's chips. The analysis notes that new disclosure rules could not undo deals already signed.
Warner put the problem bluntly in a TV interview the same day the bill was introduced. "We are flying blindly with trillions of dollars of investment," he said. Two former SEC officials, ex-commissioner Robert Jackson Jr. and former chief accountant Paul Munter, backed the bill in statements in the sponsors' release.
What it means for companies and investors
Securities rules already require companies to disclose material risks and obligations. What the bill adds is a standard AI format, so investors could compare one company's hidden commitments with another's. That would likely matter most to data center builders, chip suppliers that back their customers, and the cloud companies that sign long leases.
The text leaves a lot to the SEC. It sets no size threshold and no exemption for smaller companies. It does not define "significant" AI exposure. Nor does it spell out penalties, so enforcement would depend on how the agency writes the rules.
The politics are uncertain. Bipartisan sponsorship helps, and disclosure bills are less heavy-handed than limits on AI itself, which may be the sponsors' best argument. Still, most bills introduced in Congress never become law.
For investors, the near-term signal is behavior, not law. Watch whether large AI spenders start describing their off-balance-sheet deals in more detail, or set up board AI committees, before anyone forces them to. Companies that wait may find the question asked for them on earnings calls.
Written by AI from the public sources linked below.