The Jim Rickards $236 billion AI debt story started with a $200 billion number. In 2025, artificial intelligence companies raised roughly $200 billion through debt markets. That was the figure Rickards built his June 22, 2026 presentation around. By July 13, the number had been updated. AI companies issued $236 billion in new debt during the first five months of 2026 alone.
The pace quadrupled. What took twelve months last year took five months this year. Morgan Stanley projects AI-linked debt issuance will reach $570 billion by the end of 2026, nearly double the 2025 total.
Rickards released the updated figures through a July 13, 2026 GlobeNewswire press release. His framing has not changed. The technology is real, the borrowing is real, and the gap between the two is where the risk lives. “Technology changes quickly,” he says in the presentation. “Balance sheets don’t.”
What the $236 billion AI debt numbers actually show
The $236 billion figure covers debt issued by AI-related companies globally through May 31, 2026. Morgan Stanley tracks the data. The composition spans investment-grade bonds, high-yield debt, and private credit arrangements, with banks and private lenders continually building new financing structures designed specifically for AI infrastructure projects.
The accumulation over a longer horizon is more striking. By October 2025, AI-linked debt had reached approximately $1.2 trillion, according to M&G Investments. That made it the single largest segment of the U.S. investment-grade credit market, surpassing U.S. banks as the biggest sector in the JPMorgan U.S. Liquid Index. Rickards has been building this AI debt warning since March, and the $236 billion update sharpens the case.
This is where the story leaves Silicon Valley. When AI bonds become the largest position in the investment-grade market, they stop being a tech sector story and become a retirement account story. Index funds, target-date funds, and pension portfolios hold investment-grade bonds as their conservative backbone. The composition of that backbone has shifted toward AI infrastructure debt without most account holders knowing it happened.
Why the pace matters more than the total
A large debt stock is one thing. A quadrupling of issuance pace is another. The acceleration tells you something the total does not: the borrowing is compounding quarter over quarter.
Rickards draws a line between the borrowing pace and the earnings schedule. Companies raise debt against future revenue. When the debt issuance rate outpaces the revenue growth rate, the gap between what was borrowed and what is being earned widens every quarter. The $236 billion issued in five months was raised against the assumption that AI revenue would follow the spending within a predictable window.
The July 29 earnings cycle is the first clean test of whether that assumption holds. Meta, Microsoft, Alphabet, Amazon, and Oracle are all scheduled to report quarterly results in the window Rickards has been pointing at since March. His July 29 prediction centers on Meta as the bellwether — the company whose AI spending is most visible and whose debt profile is most exposed. The reports will show updated figures on AI spending, customer adoption, operating costs, and forward guidance.
If revenue from AI products is growing but operating expenses are growing faster, the market reads that as the adoption tax. If AI revenue is accelerating and operating expense growth is decelerating, the thesis weakens. Rickards expects the first pattern. The earnings will provide the first measurable read.
The financing innovation angle
Part of the acceleration comes from financial engineering. Rickards has been documenting the structures since April. Data center leases bundled into tranches and sold as bonds, echoing the mortgage-backed security architecture that preceded 2008. Special-purpose vehicles keeping debt off primary balance sheets, the same technique Enron used. Private credit arrangements outside public markets entirely, where pricing and risk disclosure are opaque.
Banks and private lenders keep inventing new vehicles to keep the money flowing. The July 13 release notes that fresh financing arrangements are being built specifically to fund AI projects, meaning the structures themselves are evolving to accommodate demand that traditional debt channels cannot fully absorb. When lenders start designing bespoke products for a single sector, the sector has become the market.
The JPMorgan credit default swap basket launched in 2026, covering Alphabet, Amazon, Meta, Microsoft, and Oracle, gives institutional investors a structured way to hedge against exactly this risk. When the biggest bank on Wall Street builds a product to price AI debt risk, the thesis is no longer a fringe view. Wall Street does not build hedging products for sectors it considers safe.
Where Rickards draws the line
Rickards has been explicit that he is not betting against artificial intelligence. The July 13 release states he thinks AI could reshape the economy in real ways. His concern is the distance between borrowing and earning, and the timeline over which the debt has to be repaid. A fact-check of the $200 billion baseline confirmed the underlying numbers are real — the issue is not whether the debt exists but whether the revenue follows fast enough.
Raising billions has not been the hard part. Companies have shown they can do that. The harder part comes when the debt matures and has to be repaid out of actual profits. Rickards believes many AI companies have not yet proven they can clear that second hurdle, and that investors may start demanding proof sooner than anyone expects.
The $570 billion Morgan Stanley projection for full-year 2026 tells you where the borrowing is heading. The July 29 earnings reports will tell you whether the revenue is heading there at the same speed. If the two lines are diverging, the $236 billion issued in five months becomes the number everyone suddenly wants to understand.
Rickards has been pointing at that gap since March. The earnings cycle is when it gets measured.