by Oscar Valdés Viera, Senior Policy Analyst, Private Equity & Capital Markets
In the run up to the 2008 financial crisis, credit rating agencies inappropriately awarded investment-grade ratings for risky subprime mortgages securities that fueled the predatory mortgage crisis and precipitated the Great Recession. Are we about to repeat this mistake with AI?
The companies driving the AI and data center boom are now reportedly pushing credit rating agencies to grant them top tier ratings for their risky investments. This should set off alarm bells. Regulators—including the Securities and Exchange Commission, which has authority over the credit rating agencies—should recognize a familiar warning sign and use their authority to dig into what is happening here.
According to the Financial Times, Morgan Stanley and Goldman Sachs have been lobbying credit rating agencies on behalf of OpenAI and Anthropic for investment-grade ratings following their anticipated public offerings. Morgan and Goldman are the top investment banks for the Anthropic IPO and stand to rake in millions of dollars in fees for the deal. The investment banks and the AI companies have a lot at stake in higher ratings: they would unlock cheaper borrowing and a much larger pool of investors to finance their investments. It would also further inflate the mother of all bubbles and help further spread the investment risks related to the AI buildout across the financial system—including to retirement savings accounts and pensions.
Rating agencies awarding high, investment grade ratings that are unwarranted is not a new problem—and it is a serious one. It was a significant dynamic leading up to the financial crisis, but the Dodd-Frank reform law failed to fully address this failure and regulators only watered down—and failed to enforce—these imperfect measures. Now, even as warning signs flash that the AI stock surge is a bubble that could pop and spread financial contagion across the economy—as documented in our recent report with Open Markets Institute, Brace for Impact—there is reason to fear that the rating agencies will hand out inflated ratings that will pump the bubble still further pumping up this bubble even more.
Ratings agencies pumped up the subprime mortgage bubble, precipitating the crisis
We have all suffered the consequences of overhyped credit ratings before. When the rating agencies awarded a safe investment seal of approval on risky subprime mortgage securities in the early 2000s, it gave investors and the financial system reassurance while the mortgage-backed securities (MBS) and collateralized debt obligations became ticking time bombs. Money plowed into MBS, encouraging even more, even riskier predatory subprime mortgage lending and when these mortgages—and the underlying securities—blew up, it triggered the financial crisis.
The Financial Crisis Inquiry Commission concluded that “the failures of credit rating agencies were essential cogs in the wheel of financial destruction.” And a U.S. Senate report and the Securities and Exchange Commission staff each reached similar conclusions: Rating agencies played a key role in the expansion of the mortgage securitization market and the accumulation of excessive leverage in the financial system. Favorable ratings allowed securities backed by risky mortgages to be packaged, marketed as safe investments, and distributed throughout the financial system. When the agencies eventually issued sweeping downgrades, the resulting losses, collateral demands, fire sales, and uncertainty intensified the panic.
Millions of people lost their homes and the financial crisis cratered working people’s retirement savings—$17 trillion in household wealth in homes and savings evaporated. In 2016, Moody’s agreed to pay CalPERS $130 million to settle allegations that negligent ratings had led the public pension fund into investments that later soured. S&P and Moody’s also reached federal and state settlements totaling approximately $2.2 billion over claims related to fraudulent ratings of mortgage-backed securities and collateralized debt obligations in the lead up to the crisis.
Credit rating agency business model marked by conflicts of interest
Companies pay the credit rating agencies to assign investment ratings, so there is an incentive for them to puff up their findings and give good grades even when the underlying financials do not warrant investment-grade ratings in order to continue to attract business. According to the Financial Crisis Inquiry Commission final report, when one rating agency staffperson was asked “if the investment banks frequently threatened to withdraw their business if they didn’t get their desired rating, [he] replied, ‘Oh God, are you kidding? All the time. I mean, that’s routine. I mean, they would threaten you all of the time.’” The ‘issuer pays’ model that enables this dynamic continues in place today and it appears to be playing out again with the investment banks and AI companies.
The Dodd-Frank Act did not make fundamental changes to the conflicted credit rating business model. It did require the SEC to watch over the ratings agencies and write rules to address the lack of transparency, consistency, and conflicts of interest that can fuel bubbles. But the rules the agency wrote are weak and weakly enforced and have allowed the rating agencies to continue to operate much as they did before the crisis. If it chose to use it, however, the SEC does have oversight authority.
Favorable ratings agency assessment could boost pending AI company IPOs
A lot rides on whether the ratings agencies fairly evaluate and assess the risk and creditworthiness of the AI companies. An investment-grade designation could make a firm’s bonds eligible for purchase by pension funds, insurers, and other institutions whose policies restrict exposure to speculative debt. Those institutions manage money that workers and retirees depend on. Favorable ratings could help move the risks of an uncertain AI business model into portfolios that are intended to provide lasting financial security. Despite the demand for triple A ratings from Morgan Stanley and Goldman Sachs, some are predicting lower than investment grade ratings, closer to the top end of the junk bond scale.
But unwarranted high ratings would enable further inflation of a borrowing spree that fuels excessive speculation and leverage across the financial system. If AI borrowers disappoint, downgrades could trigger fire sales by investors with investment-grade restrictions, higher capital requirements for insurers, and collateral demands where contracts require them. Simultaneous efforts to cut exposure could spread losses to balance sheets beyond the original borrowers. This is exactly the kind of cascading contagion that precipitated the last financial crisis.
Before they give out investment grade ratings the rating agencies should be examining whether future revenues can support the debt AI companies have taken on and the expensive commitments they have made to acquire chips, computing capacity, electricity, and data centers. What happens if customers pay less than projected, competition compresses margins, or equipment becomes obsolete before the associated debts mature? How dependent is repayment on repeatedly rolling over debt or raising fresh money?
OpenAI has yet to turn a profit, while Anthropic has only had one profitable quarter. The differences between their revenues and expenses are enormous. According to its IPO prospectus, Anthropic lost $42 billion in 2025, and OpenAI reportedly projects nearly $280 billion in negative free cash flow from 2026 through 2030. Bain & Co. estimates that the industry as a whole must find approximately $4.2 trillion in additional annual revenue by 2031—far beyond projected revenues from existing customers and applications—to sustain the AI buildout.
Regulators should act now to examine whether bankers are presenting evidence or pushing agencies to accommodate a desired outcome, whether the agencies are rigorously evaluating risk or being swayed by the compensation dynamics of the issuer pays model, and whether firms are shopping for ratings and unfavorable preliminary assessments are being discarded.