Wall Street banks ride AI super cycle to record quarters as financing demands reshape markets

Big Tech's AI backstops risk ignominy

Five years ago, TeraWulf Inc went public via a blank cheque merger as a bet on bitcoin mining. When crypto prices crashed shortly thereafter, the Maryland-based company pivoted into another real estate trend that also required massive electricity consumption: AI data centres. That bet is looking surprisingly solid. Last year, TeraWulf sold more than $3bn of privately placed bonds to finance data centre construction with a coupon payment of just above 7 per cent, despite the company itself still having scant revenue. The reason TeraWulf was able to secure such cheap financing was that it has a handy benefactor: Google, the Silicon Valley digital advertising giant that wants to be a big player across the automation value chain. TeraWulf secured a backstop agreement with Google to ensure that its data centres would receive contracted lease payments. Both parties benefit from such agreements. Google is just one of several Big Tech "hyperscalers" offering their balance sheet -- not quite in upfront cash but rather as an emergency deep pocket -- to give credit investors the comfort to lend at relatively low rates to unproven AI companies. Meta, too, has guaranteed data centre revenues. The hyperscalers do that because they have an interest in the data centres being built. Take Google's case, for instance. TeraWulf's big direct customer is a start-up called Fluidstack which sells the actual computing capacity inside a TeraWulf data centre. The frontier lab Anthropic is a major client further downstream. Fluidstack has pledged TeraWulf $3.7bn in lease payments over 10 years, with Google guaranteeing around half of that. At the time, the Silicon Valley stalwart took warrants in TeraWulf for nearly a tenth of the company. Google's chip segment makes so-called "tensor processing units", or TPUs, that are the backbone of AI computations, meaning that the success of TeraWulf and Fluidstack is in Google's own pecuniary interests. For those raising debt to build data centres, meanwhile, the attraction is clear to see. Despite the fact that the TeraWulf bond is officially junk-rated at BB, it has traded above par. Without Google's support, its borrowing costs would probably be in the double digits. Big Wall Street banks and private capital firms are beneficiaries too. Such guarantees allow them to raise and allocate hundreds of billions of dollars to loans that are marketed as safe and investment grade. This is clever financial engineering. But the risk, of course, is that the revenue that AI technologies can actually generate turns out to be disappointing, affecting all of the parties involved. Analysts say TeraWulf's revenue will go from about $300mn this year to 10 times that by 2029. Hit those numbers and any Google backstop will merely be academic. Miss them badly, and -- while Google's $126bn cash pile will barely be dented -- its ego will be sorely knocked.

The AI boom just found two new winners: Goldman Sachs and JPMorgan Chase

American megabanks on Tuesday gave evidence that the global artificial intelligence boom isn't just benefiting tech giants and chip makers. Goldman Sachs and JPMorgan Chase each posted record quarterly revenue hauls, fueled by massive gains in equities trading and investment banking. Behind the surge in activity -- Goldman revenue jumped 39% to $20.3 billion, while JPMorgan saw it rise 27% to $58 billion -- is the fact that AI is "everywhere in financial markets," JPMorgan CFO Jeremy Barnum told reporters. "These are booming environments with a ton of activity, big IPOs, big index rebalancing, a lot of activity in Asia," Barnum said Tuesday. "A lot of it is downstream of the AI theme, writ large on a global basis. It's just a very, very, very active environment." The quarter showed that the AI boom is creating winners far beyond Silicon Valley. While Nvidia and hyperscalers including Alphabet have captured many of the headlines, Goldman, JPMorgan and other banks are profiting from the massive flows of capital into AI. They are advising on AI-related deals, financing data centers and power infrastructure, underwriting debt and equity offerings, and facilitating the surge in trading that has accompanied the global race to deploy the technology. That is creating "a ripple effect" across the American economy and giving banks a flood of new opportunities to provide financing and trading solutions across public and private markets, Goldman CEO David Solomon told analysts Tuesday. "We are in the middle of an AI capex super cycle where there are demands on financing in every single financing instrument, in every region of the world and across every single industry," Solomon said. Capex is short for capital expenditures, or investments made by a business for physical assets like factories. Goldman is preparing for a three-to-five year investment cycle that is still in its early stages, he told analysts. Goldman shares jumped 8% in afternoon trading, while JPMorgan rose 2%.

Wall Street just had its best investment banking quarter in years, and it is calling AI a super cycle

Goldman booked record fees. Morgan Stanley thinks the market is 10% of the way through. JPMorgan's CFO is quietly walking away from data centre deals. Goldman Sachs booked $3.4bn in investment banking fees in the second quarter, a record, up 55% on a year earlier. Its chief executive has a name for what is driving it. "We are in the middle of an AI CapEx super cycle where there are demands on financing into every single financing instrument, in every region of the world and across every single industry," David Solomon told analysts on the firm's 14 July earnings call. Within the total, equity underwriting rose 130% to $985m and debt underwriting rose 75% to a record $1.03bn. It is the same demand that pulled eight banks into SoftBank's $40bn OpenAI loan. The pattern held across the street. JPMorgan reported $3.3bn in investment banking fees, up 30% and its highest since 2021. Morgan Stanley was up 58% to $2.44bn, Bank of America up 50% to $2.14bn, and Citigroup up 44% to $1.55bn, though Citi switched from reporting fees to reporting revenues this quarter, so its line is no longer like-for-like with peers. Ted Pick, Morgan Stanley's chief executive, put a number on where the cycle sits. "You're basically looking at us being around 10%-15% of the way through the investment cycle," he said on 15 July, citing his own firm's research forecasting data centre capital expenditure of roughly $850bn this year, $1.3trn in 2027, and possibly $1.5trn in 2028. He hedged it immediately. "It's really early, and I'm not sure we altogether know because of the known unknown element of this," Pick said. "One has to have humility in all of this." Jamie Dimon credited AI in JPMorgan's written results, citing "AI-driven capital investment, fiscal stimulus and the benefits of more efficient regulation" among the tailwinds behind a resilient US economy. On the call he was less lyrical. "It's getting close to as good as it gets," he said. "We just don't know how long it's going to last." Jane Fraser described the demand side at Citigroup. "AI is dominating a lot of the conversations. Tech, data center, energy, defense, CapEx is accelerating," she said. "Wherever there's a bottleneck in that whole energy power compute memory ecosystem, we're seeing a lot of activity." She pointed to SK hynix, which priced a $26.5bn American depositary receipt offering on 9 July. The most interesting remark of the week came from JPMorgan's chief financial officer, and it was about the deals the bank did not do. "We passed on some deals," Jeremy Barnum said. "When you look at the data center stuff, the key question is what happens with power supply? What happens with tenants? We saw some deals come through where we were just like, 'Yeah, we're not doing that.'" No bank disclosed a dollar figure for its data centre lending or AI infrastructure exposure. Not JPMorgan, not Goldman, not Morgan Stanley, Citi, or Bank of America. Asked directly to size AI's contribution to Goldman's results, Solomon declined. "I'm not sure that I can do that in a way where I give you a good answer," he said. Combined net income across the five largest US banks came to roughly $49bn, up about 39% year on year. Advisory work, the business that actually involves bankers talking to clients, grew far more slowly than underwriting did: Goldman's advisory line rose 17%, and Citi's fell 4%. The money is in placing paper. He also supplied the quarter's most useful corrective, twice. IPO volumes, he noted, were "kind of at or below the 10-year average" despite the record fee line. And on where this ends: "Ultimately, you will have a recalibration, a reset, a drawdown, and then a further acceleration." The financing structures underneath are getting more exotic as the volumes grow. Meta raised $27.3bn in a private placement with Blue Owl and Pimco for a single Louisiana campus, part of more than $40bn placed in that market since November. ICE is building futures contracts on compute, and CoreWeave is trying to hedge memory chips, an asset with no market. Goldman reports again in October. Barnum's two questions, power and tenants, are the ones to carry into it.

Morgan Stanley becomes Wall Street's top bank for AI debt deals

Morgan Stanley has emerged as Wall Street's chief architect of the financing structures underpinning the AI boom, devising new debt and equity models that are funnelling tens of billions of dollars into the massive build-out of data centres. According to industry executives, the bank has become the dominant adviser putting together the biggest and most inventive AI infrastructure financings since last year. This includes a $3.2bn bond for data centre developer TeraWulf backed by Google, a $27bn debt package for Meta's Hyperion data centre tie-up with Blue Owl and, more recently, advising Broadcom on a $35bn chip financing deal. The bank's surge in AI dealmaking helped it edge past longtime rival Goldman Sachs in the first half of the year in debt and equity capital market fees, which grew to $2.3bn, up from $1.4bn a year earlier, according to data from LSEG. That put Morgan Stanley in second place globally for capital markets fees after JPMorgan Chase, up from fourth a year earlier. The deals illustrate how AI is reshaping not only technology but also capital markets. Rather than relying solely on traditional project finance or corporate borrowing, bankers are increasingly designing structures that package long-term computing contracts and Big Tech's balance sheets into securities that can be sold to mainstream investors. The result has dramatically expanded the pool of capital available to fund AI infrastructure, while tying more of the financial system to continued demand for AI computing. "Dollar amounts that used to be $1bn, $2bn, $5bn are now $10bn or $20bn and higher," said Mo Assomull, Morgan Stanley's co-head of investment banking. Silicon Valley's tech giants have said they cannot meet customer orders and have continued to raise their spending. Morgan Stanley itself expects the AI build-out will consume $10tn of spending over coming years. Key to securing low rates and billions in capital has been tying in hyperscalers -- Google, Amazon, Meta and Microsoft -- which entered the AI boom with pristine balance sheets. When one of them guarantees the leases on a data centre, the cost of financing roughly halves. "Where do you want to raise money, at mid to high single digits or double that?" said Assomull. Morgan Stanley's leveraged finance co-head, William Graham, was behind the deal that has become the template for AI infrastructure financing. His team designed a bond for data centre developer TeraWulf, a security that can be sold widely. But they added some protections of a project loan, creating a hybrid instrument "wrapped" -- or backstopped -- by Google. Most of the data centre's capacity will be for Anthropic, people familiar with the matter said. The structure brought a new pool of credit investors such as insurers, asset managers and pension funds into financing data centre builds, allowing TeraWulf to raise $3.2bn at a 7.75 per cent yield. Patrick Fleury, chief financial officer of TeraWulf, said the novel construction bond let them skip the slow, stage-by-stage lender reviews that come with drawing down traditional project-finance loans from banks -- their main alternative -- while still borrowing cheaply enough to make the economics of its business work. "Effectively, we were able to borrow against the strength of Google's balance sheet," Fleury said. To further reassure investors, Morgan Stanley adopted project-finance features such as "lockboxes" that ringfence the lease payments and direct them straight to bondholders, and added other extra collateral. Since the TeraWulf deal, Morgan Stanley has sold more than $40bn of the construction bond products and is taking the structure to new markets across Asia and Europe. "We'll get to a place where AI infrastructure bonds are the majority of the annual supply of new-money non-investment-grade debt," Graham predicted. "It is the fastest-growing segment of the market and the first time to have a new segment in the market in the last 20 years." Some rival bankers said they were reluctant to be seen as the number one player in the space given the unpopularity of data centres in communities across the US. JPMorgan's chief financial officer Jeremy Barnum also cautioned this week that the bank has looked at the lending terms of some data centre financing deals "where we were just like, 'yes, we're not doing that'". Morgan Stanley has helped push the model beyond financing data centres themselves. In May, the bank and MUFG arranged a $3.1bn loan for neocloud CoreWeave to buy and install Nvidia graphics processing units, the chips needed to power advanced AI. It was the first GPU financing done as a broadly syndicated term loan, bringing in a wider pool of capital in -- this time to finance the chips themselves. The loan attracted almost $20bn of investor demand, Graham said, underscoring appetite for a financing model that hinges less on the value of the chips than on the contracts to use them. Under the structure, the GPUs and the data centre are financed separately: the building is backed by a lease, while the chips are financed against long-term take-or-pay agreements. "You may have a car payment and you may have a garage bill," Graham said. "The chip is the Ferrari . . . it needs a place to be parked. And so you need a data centre to place that chip." While the chips are added collateral, Graham said credit investors focused more on who signed the contracts to rent the chips. Morgan Stanley in March helped CoreWeave price an $8.5bn chip loan backed by a hyperscaler contract at 2.25 per cent above the benchmark rate. The May chip loan was backed by weaker credits -- two AI labs -- and priced at 4.5 per cent above the benchmark. The wider spread also illustrates a key risk underpinning the AI financing boom. The further the lending moves away from hyperscalers' balance sheets towards the AI labs consuming much of the computing power, the thinner the underlying credit becomes. Raj Joshi, a senior vice-president at Moody's Ratings, said the financial health of Anthropic and OpenAI was a risk he watched closely. "This is a huge capex investment cycle, you don't have parallels to it in history," Joshi said. "There is no playbook for this."

Wall Street banks see AI 'super cycle' set to boost deals, financing

Investment banks have reaped strong fees from AI-related deals, including SK Hynix's $26.5 billion ADR offering and SpaceX's record $86 billion initial public offering, as well as debt issuance. A rush by technology companies to fund AI infrastructure is boosting dealmaking and financing activity for Wall Street, bankers said on Tuesday, generating lucrative fees from capital raising and loans. Investment banks have reaped strong fees from AI-related deals, including SK Hynix's $26.5 billion ADR offering and SpaceX's record $86 billion initial public offering, as well as debt issuance. But July has been rough for technology stocks, especially microchip makers, as investors wrestle with high valuations and question the longevity of the AI capex boom. 'Multi-year cycle' "The build-out of ⁠AI infrastructure ⁠remains in its early stages, and we believe this multi-year investment cycle will continue to drive elevated levels of strategic activity, financing, and capital formation across markets," Goldman Sachs CEO David Solomon said during an earnings call. Solomon added that the industry "is in the middle of an AI capex super cycle" where there are demands to utilize every single financing instrument. Goldman Sachs was the lead left underwriter on the SpaceX IPO, and is also poised to play a major role alongside Morgan Stanley in the upcoming listing of Anthropic, as investors seek exposure to the AI boom. Rival OpenAI has also filed ⁠for a U.S. IPO. Citigroup, which was a joint global co-ordinator on the SK Hynix sale, earned over $70 million from the deal. AI dominating conversations: Citi Citi CEO Jane Fraser told investors on its conference call that AI was "dominating ⁠a lot of the conversation" with spending on technology, data centers, energy and defense accelerating. Bank of America recently extended a $520 million credit line to OpenAI, its first loan to the AI company, a person familiar with the matter told Reuters. "Overall, the U.S. economy has proved more durable than expected, supported by the strong consumer, ongoing AI-driven investments across the board and easing energy costs, though inflation and tighter monetary policy remain key risks," Bank of America CEO Brian Moynihan said on a conference call. BofA has helped raise nearly $500 billion for AI-related companies since 2025, accounting for 60% of such fundraising across investment-grade debt, leveraged finance and equity capital markets, according to internal data seen by Reuters. "The AI-driven capex super cycle has benefited equity issuance, M&A activity and debt financing," said Stephen Biggar, director of financial services research at Argus Research. Larger rival JPMorgan Chase has also been involved with AI-related ⁠companies on fundraising and is financing data centers. Meta Platforms is working with Morgan Stanley and JPMorgan Chase on a roughly $13 billion financing package for a data center in El Paso, Texas, Reuters reported in May, citing a source familiar with the matter. JPMorgan Chief Financial Officer Jeremy Barnum said the firm is seeing decent capital expenditure and loan demand from companies that may not be AI-related, but have an indirect link. "It's like the comments about data centers wind up creating a lot of demand for plumbers and electricians, so you wind up seeing it in sort of slightly non-obvious places", he said. Barnum added that it was hard to say if such demand was AI-related.