Credit rating agency S&P Global and renowned investment bank Morgan Stanley have separately released reports, pointing out from the perspectives of credit rating and accounting disclosure that future AI (Artificial Intelligence) spending by tech giants could reach as high as $70 trillion, while traditional financial statements may significantly underestimate the real risks.
S&P Global Ratings issued a credit outlook report on September 3, causing a stir in the industry and on Wall Street. The report bluntly stated that by 2030, the top six hyperscale tech giants globally could spend up to $70 trillion on AI capital expenditures, but they are under credit pressure and facing the pressure of credit rating downgrade.
These six companies are Microsoft, Alphabet (Google’s parent company), Amazon, Meta (Facebook’s parent company), Oracle, and SpaceX. As a debt credit rating agency, S&P screened these six companies based on their significant debt issuance in the bond market and large off-balance sheet long-term agreements.
The report mentioned that as investment scales continue to expand, the relationship between AI activities and the growth of the U.S. private sector economy is becoming increasingly close, and potential credit risks are coming under scrutiny. Two core issues were highlighted; one being the lack of quantitative data on the return on investment (ROI) for AI by tech giants, and the other being the increasingly complex financing models behind AI infrastructure, involving debt, lease obligations, and power purchase agreements among other arrangements.
While the return on investment in AI is crucial for credit rating assessment, the top six tech giants have not quantitatively disclosed the returns on AI investments. Amazon faces the risk of pressure on its AA rating; as for Oracle, if its credit rating is further downgraded, it could directly fall into the junk bond category, which considering Oracle has around $117 billion of index-related US dollar bonds, a downgrade would be a significant event in the bond market.
S&P believes that the relevant data may be difficult to calculate or that companies may choose not to disclose it publicly. Lack of this key information forces rating agencies to rely on benchmark assumptions, assuming that tech companies generally have a clear direction for their investments and that current and future AI investments will eventually yield returns. However, even investment-grade companies cannot increase debt burdens indefinitely, hence critical thresholds triggering rating downgrades have been set for each company.
S&P also warned that an increasingly complex “cyclical financing” ecosystem has formed among hyperscale tech giants, involving arrangements such as residual value guarantees, take-or-pay agreements, chip financing, lease liabilities, power purchase agreements, backstop guarantees, and direct equity investments. While these financing arrangements may currently be within companies’ manageable scope when viewed individually, if AI demand were to significantly decrease, the previously dispersed risks could simultaneously erupt due to their high interconnections. Even the financially strongest and most profitable companies could face significant pressure during an economic downturn.
Morgan Stanley, in a report in early September, for the first time dissected the “footnotes” in tech giants’ financial reports, revealing the massive off-balance sheet “shadow debt” tech giants have incurred for AI investments.
The report estimated that seven tech giants such as Google, Microsoft, and Nvidia have committed to spend around $30 trillion on AI-related infrastructure, but these expenditure commitments are not included in the balance sheet. The capital burdens of the AI race are extending to off-balance sheet financing, and traditional financial statements may underestimate the actual risk exposure of tech giants.
This includes around $1.1 to $1.2 trillion in “infrastructure leasing contracts not yet initiated” (i.e., not recorded on the balance sheet until data centers are operational), and $1.7 to $1.9 trillion in chip and server long-term procurement agreements.
The report also unveiled a crucial four-layer financial chain, where tech giants leverage by providing long-term lease endorsements, credit guarantees, and customer prepayments, allowing downstream startups or data center developers to obtain leveraged bank loans.
The report cautioned that the capital expenditures of tech giants have begun to surpass their own cash flow generation capabilities, with companies like Google and Amazon recently having negative free cash flows, necessitating continuous bond issuances to sustain spending.
The commentary from The Economist suggests that the AI frenzy indeed exhibits some characteristics reminiscent of the early dot-com bubble, including the rapid rise of new technologies, historic bull markets in tech stocks, and the sharp increase in valuations of tech giants. The scale of “shadow debt” for these tech giants is considerable, and it may continue to rise.