Morgan Stanley Earns $2.3B in Capital-Markets Fees as AI Debt Boom Lifts Its Rank

Morgan Stanley (MS ) collected $2.3 billion in debt and equity capital-markets fees in the first half of 2026, up from $1.4 billion a year earlier — a jump that pushed the bank to second place globally behind JPMorgan Chase and past its longtime rival Goldman Sachs, according to LSEG data reported by the Financial Times. A year ago it ranked fourth. Almost all of the increase traces to a single source: the financing behind the AI infrastructure buildout.

That haul is the payoff on a deliberate bet. Over the past year Morgan Stanley has positioned itself as Wall Street’s busiest arranger of the debt and equity packages funding data centers, the chips inside them, and the power to run them. Rather than lean on traditional project finance or plain corporate borrowing, its bankers have been bundling multiyear cloud-computing contracts and the balance sheets of Big Tech into bonds that ordinary institutional investors can buy — a shift that has widened the pool of capital available for the buildout while tying more of the financial system to continued AI demand.

The template was a $3.2 billion bond for data-center developer TeraWulf (WULF ), a tradable security the bank backstopped with a guarantee from Google and sold to insurers, asset managers and pension funds. TeraWulf priced it at a 7.75% yield, with most of the capacity earmarked for Anthropic. Since that deal, Morgan Stanley has sold more than $40 billion of similar “construction bonds” and is carrying the structure into Asia and Europe, where issuers are already testing the same well — Mistral raised $830 million in debt to build a Paris data center earlier this year.

Bigger deals followed: a $27 billion debt package for Meta’s Hyperion data-center venture with Blue Owl, and, more recently, an advisory role on a $35 billion chip-financing deal for Broadcom. “Dollar amounts that used to be $1 billion, $2 billion, $5 billion are now $10 billion or $20 billion and higher,” said Mo Assomull, the bank’s co-head of investment banking.

Borrowing against Big Tech’s balance sheet

The engineering that makes these deals work is the hyperscaler guarantee. When Google, Amazon (AMZN ), Meta or Microsoft (MSFT ) backstops the leases on a data center, the cost of borrowing drops by about half, because credit investors are lending against a blue-chip balance sheet rather than an unproven developer. The hyperscalers are leaning on debt in the first place because their capital spending is on track to consume nearly all of their operating cash flow this year, leaving bond markets to fund the gap. Morgan Stanley layered on project-finance protections — “lockboxes” that ring-fence lease payments for bondholders, plus extra collateral — to reassure the new buyers.

The bank has since pushed the model onto the chips themselves. In March 2026 it priced an $8.5 billion loan for neocloud CoreWeave (CRWV ), backed by a hyperscaler contract, at 2.25 percentage points over the benchmark rate. Two months later it teamed with Japan’s MUFG to arrange a $3.1 billion loan — the first GPU financing sold as a broadly syndicated term loan — to buy and install Nvidia chips. That deal drew roughly $20 billion in demand, but it was backed by weaker credits, two AI labs, and priced at 4.5 points over the benchmark.

The gap between those two spreads is the entire story. The further a deal sits from a hyperscaler’s balance sheet, and the closer it sits to the AI labs actually consuming the compute, the thinner the underlying credit — and the more investors demand to hold it.

Where the risk sits

That is also where the risk concentrates. The ultimate tenants for much of this capacity are OpenAI and Anthropic, companies burning cash at extraordinary rates even as they raise it just as fast; Anthropic has weighed a raise valuing it at $900 billion. Their financial health underwrites the newest, thinnest layers of AI debt. Raj Joshi, a senior vice-president at Moody’s (MCO ) Ratings, said he watches both closely. “This is a huge capex investment cycle, you don’t have parallels to it in history,” he said. “There is no playbook for this.”

Execution risk is real too — OpenAI and Oracle scrapped a planned Stargate expansion in Texas (ORCL ) earlier this year — and some rivals are keeping their distance. JPMorgan’s finance chief said the bank walked away from data-center deals whose lending terms it found unpalatable, and skeptics continue to question whether the infrastructure spending will pay off.

For now, the volume is rewriting Wall Street’s income statements. Morgan Stanley reported record second-quarter revenue of $21.35 billion in July, with investment-banking fees up 58% year over year, and the bank estimates the broader buildout will consume some $10 trillion in spending over the coming years. Its leveraged-finance team expects AI infrastructure bonds to become the majority of new non-investment-grade debt issued each year — the first genuinely new segment to appear in that market in two decades.