Wall Street finds itself in an unusual position as it prepares to lend staggering amounts to artificial intelligence companies. Even as banks facilitate what may become the largest borrowing binge in technology history, they are simultaneously deploying an arsenal of financial tools to protect themselves from the very bubble their money might be inflating.

The anxiety permeating credit markets tells the story. The cost of insuring Oracle debt against default through derivatives has climbed to levels not seen since the Global Financial Crisis. Morgan Stanley has explored using specialized insurance mechanisms to reduce exposure to its tech borrowers. Across trading desks, lenders are quietly hedging positions even as they publicly champion the transformative potential of artificial intelligence.

Unprecedented Wall Street lending to technology giants

Mega offerings from Oracle, Meta Platforms and Alphabet have pushed global bond issuance past $6.46 trillion in 2025. These hyperscalers, alongside electric utilities and related firms, are expected to spend at least $5 trillion racing to build data centers and infrastructure for technology promising to revolutionize the global economy.

The scale is so immense that issuers must tap virtually every major debt market, according to JPMorgan Chase analysis. These technology investments could take years to generate returns, assuming they deliver profits at all. The frenzied pace has left some lenders dangerously overexposed, prompting them to use credit derivatives, sophisticated bonds and newer financial products to shift underwriting risk to other investors.

Technology that may not translate to profits

Steven Grey, chief investment officer at Grey Value Management, emphasized that impressive technology does not automatically guarantee profitability. Those risks became tangible last week when a major outage halted trading at CME Group and served as a stark reminder that data center customers can abandon providers after repeated breakdowns. Following that incident, Goldman Sachs paused a planned $1.3 billion mortgage bond sale for CyrusOne, a data center operator.

Banks have turned aggressively to credit derivatives markets to reduce exposure. Trading of Oracle credit default swaps exploded to roughly $8 billion over the nine weeks ended November 28, according to analysis of trade repository data by Barclays credit strategist Jigar Patel. That compares to just $350 million during the same period last year.

Banks are providing the bulk of massive construction loans for data centers where Oracle serves as the intended tenant, likely driving much of this hedging activity, according to recent Morgan Stanley research. These include a $38 billion loan package and an $18 billion loan to build multiple new data center facilities in Texas, Wisconsin and New Mexico.

Hedging costs climb across the sector

Prices for protection have risen sharply across the board. A five year credit default swap agreement to protect $10 million of Microsoft debt from default would cost approximately $34,000 annually, or 34 basis points, as of Thursday. In mid October, that same protection cost closer to $20,000 yearly.

Andrew Weinberg, a portfolio manager at Saba Capital Management, noted that the spread on Microsoft default swaps appears remarkably wide for a company rated AAA. The hedge fund has been selling protection on the tech giant. By comparison, protection on Johnson & Johnson, the only other American company with a AAA rating, cost about 19 basis points annually on Thursday.

Weinberg suggested that selling protection on Microsoft at levels more than 50% wider than fellow AAA rated Johnson & Johnson represents a remarkable opportunity. Microsoft, which has not issued debt this year, declined to comment. Similar opportunities exist with Oracle, Meta and Alphabet, according to Weinberg. Despite their large debt raises, their credit default swaps trade at high spreads relative to actual default risk, making selling protection sensible. Even if these companies face downgrades, the positions should perform well because they already incorporate substantial potential bad news.

Sophisticated tools to Wall Street shift risk

Morgan Stanley, a key player in financing the artificial intelligence race, has considered offloading some data center exposure through a transaction known as a significant risk transfer. These deals can provide banks with default protection for between 5% and 15% of a designated loan portfolio. Such transfers often involve selling bonds called credit linked notes, which can have credit derivatives tied to companies or loan portfolios embedded within them. If borrowers default, the bank receives a payout covering its loss.

Morgan Stanley held preliminary talks with potential investors about a significant risk transfer tied to a portfolio of loans to businesses involved in AI infrastructure, Bloomberg reported Wednesday. Mark Clegg, a senior fixed income trader at Allspring Global Investments, observed that banks remain fully aware of recent market concerns about possible overinvestment and overvaluation. He suggested it should surprise no one that they might explore hedging or risk transfer mechanisms.

Private capital firms including Ares Management have been positioning themselves to absorb some bank exposure through significant risk transfers tied to data centers. The massive scale of recent debt offerings adds urgency to these efforts. Not long ago, a $10 billion deal in the American high grade market qualified as big. Now, with multi trillion dollar market capitalization companies and funding needs in the hundreds of billions, Teddy Hodgson, global co head of investment grade debt capital markets at Morgan Stanley, suggested that $10 billion represents merely a drop in the bucket. He noted that Morgan Stanley raised $30 billion for Meta in a drive by financing executed in a single day, an event not historically commonplace. Investors will need to adjust to bigger deals from hyperscalers given how much these companies have grown and how expensive capturing this opportunity will prove.