The Cost of AI: The Debt Frenzy of Tech Giants and Hidden bubble Concerns

Recently, American tech giants have launched an unprecedented bond issuance frenzy. Meta issued $30 billion in corporate bonds, setting a record for the largest corporate bond transaction since 2023. Alphabet, the parent company of Google, issued $25 billion. Oracle raised 18 billion US dollars. Amazon also launched its first bond issuance in three years, raising 15 billion US dollars. The total scale of this bond issuance feast has exceeded 200 billion US dollars. The bond issuance volume of the five major cloud computing giants alone has reached 121 billion US dollars, which is four times the average level of the past five years.

Behind this debt frenzy lies the frenzied investment of tech giants in AI infrastructure. Oracle has made it clear that the $18 billion bond will mainly be used for the construction of data centers. Even more astonishingly, Meta's $30 billion bond received an oversubscription of $125 billion, setting a new record for the highest subscription in the history of US corporate bond issuance. Jpmorgan Chase predicts that this wave of AI investment will drive the US high-rated bond market to a record size of 1.81 trillion US dollars next year.

Ironically, on the day Meta issued its bonds, its share price plunged by 11.33%, yet the bonds were highly sought after. This split scene in the stock and bond markets reveals investors' contradictory attitude towards AI: the bond market values the stable cash flow of tech giants at present, while the stock market casts a vote of no confidence in their long-term profitability. Bank of America data shows that the spread of such bonds has climbed to 78 basis points, significantly widening from September, indicating that the risk premium demanded by the market is on the rise.

Risks are secretly accumulating. Barclays Bank's report indicates that Oracle's capital expenditure to fulfill AI contracts has far exceeded its cash flow support capacity. The bank has downgraded its debt rating to "underweight" and warned that it may slide towards the brink of BBB- junk debt. What is even more alarming is that Oracle's credit default swaps rose sharply in September, with the 5-year CDS rate reaching 87.7 basis points, nearly doubling compared to August. This signal is reminiscent of the eve of the 2008 subprime mortgage crisis.

The more fundamental question is whether massive investment can bring corresponding returns? Reality is depressing. A KPMG survey shows that although 93% of Canadian enterprises use AI, only 2% of leaders report a return on investment. The report from the Massachusetts Institute of Technology also points out that 95% of enterprises have failed to see results in their investments in generative AI. DA Davidson, an expert, warns that AI companies still need hundreds of billions of dollars in funds. If they continue to invest in assets with unclear returns, it may trigger systemic risks.

Facing financing pressure, technology companies have also resorted to financial magic. Meta has completed nearly 30 billion US dollars in indirect financing by setting up a special purpose vehicle (SPV). As it only held a 20% stake in the SPV, this huge debt was able to "disappear" from the balance sheet. This kind of "small-share operation" has enabled Meta to repay its principal and interest in the form of rent over the next 16 years, which not only beautifies its financial reports but also sows regulatory risks.

Goldman Sachs analysis shows that since the release of ChatGPT, the market value of AI-related companies has soared by more than 19 trillion US dollars, approaching the upper limit of their long-term contribution expectations to AI. Looking back at the dot-com bubble in 2000, investors were once wild with illusory concepts. Nowadays, although tech giants have a more solid foundation, it remains to be seen whether this AI race built on debt will bear fruit or turn into bitter wine. On this path leading to the future of AI, tech giants are building tomorrow's dreams with today's debts, and dreams and bubbles are often separated by only a thin line.