Plunge in Asia’s AI shares sows doubts over world-beating rally

By Christine Ji

Inside the financial engineering of the AI build-out: Nvidia, Google and the rest of Big Tech are turning to Wall Street, vendor financing and private credit

In a rural Louisiana parish that is home to some 20,000 inhabitants, an army of bulldozers and excavators has been busy clearing a vast patch of farmland, the next frontier of the artificial-intelligence revolution. It is here where Mark Zuckerberg and Meta Platforms Inc. have chosen to build Hyperion, an AI supercomputer housed in a nine-building campus of data centers humming with the latest Nvidia semiconductor chips and equipment.

To help pay for the Manhattan-size data center, Zuckerberg has turned not to big banks like JPMorgan Chase (JPM) or Goldman Sachs (GS) that have helped finance previous American industrial expansions, but rather to the alternative-asset manager Blue Owl Capital (OWL). Funded by institutional investors such as pension funds and insurance companies instead of by customer deposits, Blue Owl's lending vehicles have boomed in recent years as regulators pushed riskier loans away from banks and toward entities not backed by U.S. government guarantees.

Hyperion will be built and owned by a $27 billion joint venture, with Blue Owl-managed funds taking an 80% stake and Meta (META) holding 20%. Blue Owl will fund a portion of its contributed capital through debt issued to Pimco and other institutional bond investors via a private securities offering. Meta will lease the facility upon its completion in 2029. It's the largest private-capital deal that's ever been done on Wall Street.

AI has the potential to completely transform the world. But first, it needs a few trillion dollars. Out of the estimated $2.9 trillion in AI capital expenditures expected by 2028, the hyperscalers building the technology - including Meta, Amazon.com Inc. (AMZN), Google parent Alphabet Inc. (GOOGL) (GOOG), Microsoft Corp. (MSFT) and Oracle Corp. (ORCL) - will only generate enough cash to cover $1.4 trillion, according to a report by Morgan Stanley strategists led by Vishwas Patkar.

To get the remaining $1.5 trillion, Big Tech companies will need to not only utilize traditional debt instruments like corporate bonds but also carry out feats of financial engineering that will fundamentally redefine their relationship with Wall Street - and pull the entire market along for the ride.

There are existing tools to fund a capital-expenditure cycle. The rise of high-yield bonds in the 1980s democratized the public debt market, allowing non-investment-grade companies to easily raise capital. In the 1990s, telecommunications companies engaged in vendor financing, lending money to their customers to help fund purchases.

But today, megacap tech names have partnered with private startups and higher-risk public companies to create a large-scale, opaque financing system that makes past strategies look rudimentary. Neoclouds that rent out graphics processing units, or GPUs, have emerged to provide fast and flexible AI computing capacity for an ecosystem where data-center inventory seems perpetually constrained. OpenAI's dealmaking spree has put the company at the center of over $1 trillion in infrastructure partnerships. Supplying the GPUs that are powering it all is chip giant Nvidia Corp. (NVDA), which has made strategic investments in its top customers.

To avoid depleting their own free cash flows - the money available to reinvest back into the business or distribute to shareholders - the hyperscalers are looking for novel ways to finance their data-center build-outs. Meta's minority stake in its Hyperion joint venture will appear as an "non-marketable equity investment" on its balance sheet, according to the company's latest quarterly filing.

That means "the debt is not consolidated on Meta's books. It sits with the special-purpose vehicle," Sean McDevitt, partner at the consulting firm Arthur D. Little, told MarketWatch. Arthur D. Little served as commercial due-diligence adviser to Meta on the Hyperion deal.

And Wall Street is eager to service these investment-grade tech behemoths, seeing them as extremely reliable streams of income.

For many of the companies in this web, AI has become an "existential" investment, Michael Green, portfolio manager and chief strategist at Simplify Asset Management, told MarketWatch. The winners could reap trillions while the losers face obsolescence.

By using borrowed money instead of cash, "you're magnifying the outcome when you win," Green said. But, he added, that comes at the cost of amplifying the downsides if you lose.

Neoclouds and vendor financing

CoreWeave, led by CEO Mike Intrator, started out as a crypto-mining company. It had its initial public offering earlier this year.

CoreWeave Inc. (CRWV), the largest neocloud company, counts Microsoft, Meta and OpenAI as its top customers. But before it was a multi-gigawatt power broker to the world's biggest tech companies, CoreWeave was a small crypto-mining company started by three commodities traders in 2017.

CoreWeave's rise from a New Jersey garage full of chips to a neocloud leader with 33 data centers across two continents wouldn't have been possible without its unique relationship with Nvidia. By 2021, CoreWeave had become Nvidia's most trusted and specialized cloud partner, giving it early access to Nvidia's latest chips. Nvidia invested $100 million in CoreWeave in 2023, when the latter was still a fledgling private company. The chip giant went on to serve as an anchor investor in CoreWeave's March 2025 initial public offering. In October, a 2023 agreement stating that Nvidia is obligated to purchase up to $6.3 billion of CoreWeave's unsold cloud capacity until 2032 was publicly disclosed in a securities filing. Nvidia currently holds a 6.6% stake in CoreWeave.

It's a symbiotic relationship: CoreWeave receives financial backing to support its capital-intensive business, and Nvidia secures an important distribution channel for its technology. CoreWeave did not respond to a request for comment.

In the last two months, a flurry of deals have further highlighted Nvidia's role at the core of the AI boom: The company has pledged to invest $100 billion in one of its top customers, OpenAI; put $2 billion in a special-purpose vehicle tied to Elon Musk's AI company xAI; and taken a $5 billion stake in chip maker Intel Corp. (INTC). For Nvidia, which recorded an impressive $72 billion of free cash flow for the 12 months ending July 27, reinvesting its cash into its customers is an effective way to drive the overall AI infrastructure build-out and, subsequently, its own sales, Ayako Yoshioka, portfolio consulting director at Wealth Enhancement, told MarketWatch.

"We do not require companies we invest in to use Nvidia technology," a Nvidia spokesperson said in a statement to MarketWatch.

"There's nothing inherently wrong with vendor financing," John Huber, founder and portfolio manager at Saber Capital Management, told MarketWatch. Customers can buy products without taking out a bank loan, and suppliers can lock down revenues and pricing power.

But these types of agreements have backfired when the demand for such purchases dries up, something that telecommunications giants such as Cisco Systems Inc. (CSCO) and Lucent Technologies experienced during the dot-com bubble, Huber pointed out. When their highly leveraged customers went bankrupt, both companies suffered severely. Cisco was forced to write off over $2 billion in inventory in 2001. Lucent was saddled with $8 billion in debt, charged with accounting fraud and eventually acquired by a French telecommunications company.

Huber cautioned, however, against drawing too many comparisons between today's AI build-out and the dot-com era. He emphasized that Nvidia, Google, Meta and Microsoft have "pristine credit" and robust free cash flows, unlike the low-quality companies that went bust during the internet bubble. But Huber added that these vendor-financing agreements could likely lead to overbuilding of AI infrastructure that eventually turns today's scarcity of data centers and computing power into a glut.

Additionally, unlike the hyperscalers, many of the companies benefiting from vendor financing don't have the revenues and free cash flows to support their levels of spending, Wealth Enhancement's Yoshioka said. CoreWeave recorded over $1 billion in net losses for the 12 months ending June 30, according to FactSet. ChatGPT creator OpenAI is a private company that recorded a $13.5 billion loss on $4.3 billion in revenue in the first half of the year, according to The Information.

Nvidia's equity investment and special purchasing agreements - such as its agreement to buy six gigawatts of Advanced Micro Devices Inc. chips in exchange for warrants to purchase 160 million of AMD's stock (AMD) - have helped OpenAI secure over $1 trillion in computing power.

A rich cosigner

Meta's Facebook data center in Eagle Mountain, Utah, shown here, is a complex of five large buildings totaling 2.4 million square feet. Meta's Hyperion data center in Richland Parish, La., will be the size of Manhattan.

Following in CoreWeave's footsteps, TeraWulf Inc. (WULF) and Cipher Mining Inc. (CIFR) were bitcoin-mining businesses that recently pivoted to building and leasing out AI data-center facilities. As it turns out, running AI workloads is a much more lucrative way to use GPUs than mining cryptocurrency.

In the last three months, both companies separately entered into 10-year agreements to host the private neocloud company Fluidstack.

TeraWulf lost $132 million and Cipher lost $71 million in the 12 months ending June 30, according to FactSet, making it difficult for them to raise money to build the infrastructure needed for their agreements with Fluidstack. But as evidenced by CoreWeave and OpenAI, companies in the age of AI need not meet profitability prerequisites before entering into financial contracts - as long as a Big Tech company is willing to cosign for them, like a wealthy parent helping a child secure a mortgage.

In what's called an equity wrap, Google, the world's fourth-biggest company, agreed to backstop $3.2 billion of Fluidstack's obligations to TeraWulf and $1.4 billion to Cipher, receiving warrants to purchase equity stakes in the miners in return, securities filings show. Under the agreement, Google will pay TeraWulf and Cipher up to the predetermined amount if Fluidstack backs out of these data-center leases, reducing the risk for the miners.

The deals offer Google two key advantages: financial upside through its warrants and, more important, future priority access to cloud capacity. The arrangement strategically avoids a capital-intensive undertaking for Google, as TeraWulf and Cipher bear the cost of owning the data centers.

"We have a long history of partnering with data-center colocation providers to bring our Cloud technologies closer to customers around the world," a Google spokesperson told MarketWatch. "Our partnerships with Cipher Mining and TeraWulf are an extension of that strategy as we focus on increasing capacity to meet customer needs."

That's exactly how TeraWulf became the first-ever crypto miner to raise debt in the high-yield corporate bond market last month. The move could kick off a trend of data centers funded by junk bonds - debt issued by companies too financially shaky to be rated investment-grade by the ratings agencies.

In the biggest junk-bond sale since the landmark 1989 leveraged buyout of RJR Nabisco, TeraWulf issued $3.2 billion worth of bonds to finance its data-center build-out in upstate New York - notably, the same amount as the Google backstop. While ratings agencies were debating the bond's placement on the junk scale, Google's guarantee de-risked the offering enough to secure a BB rating from Fitch, positioning the issuance on the most creditworthy side of the junk spectrum. Google is "ultimately stepping under the debt in its totality," Simplify's Green said.

TeraWulf and Cipher Mining did not respond to requests for comment.

If unprofitable bitcoin miners can raise public debt with the backing of the hyperscalers, then the hyperscalers themselves can undoubtedly do the same.

It's an increasingly relevant topic. Data-center projects like Meta's Hyperion or Amazon's Project Rainier come with price tags in the tens of billions of dollars. Intense investment spending led free cash flow at Google, Meta, Amazon and Oracle to decrease year over year in the first half of 2025, according to data from FactSet.

Some have indeed tapped the corporate bond market to fuel massive long-term AI spending plans. In September, Oracle issued an $18 billion bond offering for its data-center build-out, including one tranche of bonds with a 40-year maturity date. Oracle isn't the first to use such long-term debt - last year, Meta's $10.5 billion bond issuance also included a 40-year tranche.

Meta came to the credit market again last week, reporting the sale of $30 billion worth of bonds last week, which a securities filing shows also includes 40-year maturities. Alphabet announced on Monday that it is currently in the process of selling several billion dollars' worth of bonds in both the U.S. and Europe, an offering that a securities filing says includes maturities of up to 50 years.

Meta and Alphabet did not respond to a request for comment.

But corporate bonds aren't the financing tool of choice for Big Tech, which built its blue-chip status on low debt levels. The "Magnificent Seven" group of megacap tech companies may want to avoid incurring interest payments that could eat into earnings and pressure their stock valuations, according to Morgan Stanley's Patkar.

CoreWeave is a cautionary tale: It's been able to raise debt rapidly, but even with the financial support of Nvidia, the strategy has come with high interest rates. D.A. Davidson analyst Gil Luria sees CoreWeave's debt levels as a key financial risk. He highlighted in a note last month that the company's interest expenses are outpacing its operating income, meaning the business doesn't make enough from operations to pay its lenders. Luria anticipates the situation will worsen, as CoreWeave is expected to take on more debt to finance its expansion. As of June 2025, CoreWeave's weighted average interest rate on short-term debt was 12.3%, up from 9.6% six months prior, according to company filings.

As a result, Morgan Stanley anticipates that corporate bond issuances will only account for around $200 billion of the $2.9 trillion global spending on AI from 2025 to 2028.

The absence of AI-driven revenues seems to be contributing to this trend. If and when the Big Tech companies drive more AI-related revenues, "we think investors will have a clearer line of sight to facilitate unsecured debt issuance," Patkar wrote.

Private credit

Data centers could be turning into "an investable class of assets," one strategist said.

Blue Owl and Pimco fought their way through a competitive lineup of rivals including Apollo and KKR to secure the Meta private-credit deal, reportedly by proposing a more flexible financing structure that offered Meta increased control over the debt.

As major tech players tussle for scarce GPUs and computing capacity, an equally heated scramble is playing out among Wall Street asset managers. Private-credit funds - which have historically serviced smaller, riskier borrowers - are hungry for a piece of the AI trade and more than happy to oblige Big Tech's bespoke requests for raising massive amounts of capital without incurring debt and hurting free cash flows. Returns in the private-debt space have come down from their recent peak in 2021 - with the exception of infrastructure debt, which has returned 13.1% in the last year, compared with the mid-single-digit returns of other strategies, according to a PitchBook report.

The Meta and Blue Owl partnership could be "a blueprint for how funding may work in this next era," according to Arthur D. Little's McDevitt.

"You've created a data center as an investable class of assets, in a way," he said.

Blue Owl did not respond to requests for comment.

The structure of Meta's private-credit deal expands on the overall trend of Big Tech companies leasing data centers instead of building them. Microsoft and Oracle are increasingly using finance leases - debtlike long-term liabilities - for their AI infrastructure plans, securities filings show.

"From a cash-flow perspective, you're not buying the equipment," Jeffrey Foster, a finance professor at Georgetown University, said of leasing.

Unlike an upfront purchase, finance leases allow companies to spread the cost of the asset over periodic payments with the option to own the asset, similar to how someone might finance the use of a new car. Although finance leases aren't classified as traditional debt on the balance sheet, ratings agencies take them into consideration when assessing financial risk.

This sort of asset-backed financing, collateralized by a contractual stream of cash from the underlying data center or equipment, can be done by traditional Wall Street banks. But the appeal of private credit comes from its flexibility. Post-2008 regulations have imposed limits on how much banks can lend to a single client. That means a bank may already be near its regulatory threshold on credit exposure to a Big Tech company, restricting its ability to fund a new data center, according to Foster. The banking rules designed to prevent another too-big-to-fail crisis have created a large private-credit industry in which financial firms raise money from investors, as opposed to depositors.

For its Hyperion site, Meta has entered into four-year initial operating leases with the joint venture to use the facilities upon completion. Unlike finance leases, which are typically long-term agreements with no cancellation options, operating leases resemble a shorter-term rental and offer increased flexibility. But Meta has also promised to reimburse investors for potential losses if it terminates its lease early and the value of the data center falls below a predetermined level during the first 16 years of operation.

Despite the off-balance-sheet structure of the deal, "the debt is so large I imagine Meta would still have to disclose it in their footnotes," Foster said.

Asset-backed financing presents a great risk-reward trade-off for Big Tech. Leasing costs are typically an annual 7% to 10% of the upfront cost, more expensive than investment-grade corporate bond-market rates but still "quite affordable for cash-rich hyperscalers, especially if this means avoiding large capital spend," according to Patkar. The Morgan Stanley strategist anticipates that $800 billion of the total $2.9 trillion of AI capital expenditure will be funded by some form of private credit in the next three years.

US stocks snapped a three-week run of gains on Friday as investors continued to fret about stretched valuations in the tech sector, but saw a silver lining in Washington lawmakers restarting negotiations over the government shutdown.

The S&P 500 Index finished 0.1% higher in New York, paring a decline of more than 1.3% earlier in the session as traders assessed signs that the US government shutdown is nearing its end. The benchmark traded below its 50-day moving average earlier in the session. The tech-heavy Nasdaq 100 Index dropped 0.3%, also ending a three-week win streak. Block Inc., Microchip Technology Inc. and Take-Two Interactive Software Inc. were some of the biggest decliners.

Meanwhile, the Cboe Volatility Index hovered around 19. It had gone above 20 earlier in the session, a watermark that often signals mounting market stress.

“Given the froth of valuations, this is a healthy correction, more of a reversion to the mean than an indictment of the AI narrative,” said Louis Navellier, chief investment officer at Navellier & Associates.

The latest round of nonfarm payrolls data has been delayed because of the government shutdown. However, private data released earlier this week paints a picture of a softening labor market.

“We think this cooling keeps the Fed’s rate cut plans alive for December and potentially into early 2026,” said Glen Smith, chief investment officer at GDS Wealth Management. “While the labor market is softening, it’s not severe enough to suggest that any kind of recession is on the horizon anytime soon.”

Data from Challenger, Gray & Christmas Inc. on Thursday showed companies had announced the most job cuts for any October in more than two decades, as artificial intelligence reshapes industries and cost-cutting accelerates.

To Jennifer Timmerman, senior investment strategy analyst at Wells Fargo Investment Institute, recent high-profile layoff announcements “likely overstate” weakness in the job market.

“All in, the data we do have in hand remain consistent with our constructive economic outlook for 2026 from a soft patch at the turn of the year, and we look for policy-related tailwinds already in the pipeline — including tax cuts, monetary stimulus, and deregulation — to spur a reacceleration of economic growth beyond the early part of next year,” Timmerman said.

Still, the shutdown will likely impact gross domestic product growth in the US, according to President Donald Trump’s economic advisor Kevin Hassett.

“We were thinking that we have at least 3% growth in the fourth quarter. I think now we’re expecting something like half of that,” Hassett said on Fox Business.

Valuation Concerns

Tech stocks have sold off throughout the week, with beneficiaries of the artificial intelligence rally taking a step lower. Results from Palantir Technologies Inc., Super Micro Computer Inc., and Qualcomm Inc. left investors feeling underwhelmed.

The stock market’s euphoria has quickly been overtaken by a bout of skepticism, hitting mega-cap technology companies that have led the AI rally. Concerns have centered on who is going to provide all the funds needed to finance lofty ambitions of OpenAI, which sits at the center of the investment theme that has pumped billions of dollars into the equity market.

Earlier this week, hedge fund manager Michael Burry disclosed bearish wagers on Palantir and Nvidia Corp., which fueled concerns over stretched valuations and whether the tech rally could be running out of steam. However, GDS Wealth Management’s Smith sees an opportunity for investors.

“Some big tech stocks are on sale, and are presenting buying opportunities for investors, especially for investors who have missed out on the market’s strength over the past two months,” he said.

Tesla Inc. traded lower after shareholders approved a $1 trillion compensation package for Chief Executive Officer Elon Musk. It’s the largest payout ever awarded to a corporate leader.

“Everyone is getting caught up on the big number and what it means for Musk’s remuneration,” said Kyle Rodda, senior analyst at Capital.com. “But the bonus is only secured if he achieves some pretty ambitious goals and that’s far from assured.”

In single stock moves, Take-Two Interactive dropped after the video-game maker delayed the release of the highly-anticipated Grand Theft Auto VI once again. Block tumbled after the fintech platform reported an adjusted earnings and net revenue miss.

CNH Industrial NV fell after cutting guidance and noted that the August expansion of US steel and aluminium tariffs put pressure on its business. Sweetgreen Inc. plunged to a record low after the restaurant chain cut its revenue guidance for the year and received a downgrade from William Blair. Monster Beverage Corp. gained as the energy drinks company’s third-quarter results topped estimates.

By Shaina Mishkin

Is buying a house just for middle-agers? For the first time, the median age for initial home buyers has reached 40, up from 30 in 2010.

Home sales have been ailing, hit by high prices and mortgage rates. According to the National Association of Realtors' 2025 profile of home buyers and sellers, first-time buyers are becoming owners at an age when homeowners in 1997 were already trading up. "It feels like something has shifted in the housing market," says Jessica Lautz, the NAR's deputy chief economist and vice president of research. "I don't know if we can move backward [in age] unless we have a substantial change, like more-affordable housing inventory."

NAR data chronicle a stuck housing market: The share of all-cash buyers, at 26%, is tied with 2024 for highest on record, while the share of first-time buyers, at 21%, is a low. Repeat buyers put a median 23% down on their purchase, the highest since 2003. First-time buyers put down 10%, the most since 1989. And sellers' tenure in their homes grew to 11 years from 10 last year. "People just want to stay put," says Lautz.

There's a bright side. Mortgage rates are below the 6.7% one-year survey period average. Fixed 30-year rates measured by Freddie Mac are down to 6.22%, meaning those with a $400,000 loan will save more than $100 a month compared with the survey average. But, she adds, the decline is unlikely to improve demand on its own.

Write to Shaina Mishkin at shaina.mishkin@dowjones.com

Last Week

Markets

Oil prices rose as OPEC+ paused plans for increasing output. The government shutdown continued, U.S. manufacturing shrank for the eighth straight month, and stocks flagged on artificial-intelligence worries, sending shares of Palantir Technologies down despite record revenue. Democrats swept state and local elections, and the Supreme Court appeared skeptical of President Trump's tariffs. On Thursday, the S&P 500 index fell 1.1% after private-sector data showed big layoffs. On the week, the Dow industrials were down 1.2%; the S&P, 1.6%; and the Nasdaq Composite, 3%.

Companies

Tesla shareholders rewarded Elon Musk with a trillion-dollar pay package. The White House said Nvidia has to reserve its most advanced chips for U.S. customers — not Chinese — and sealed a deal with Novo Nordisk and Eli Lilly to cut the cost of anti-obesity drugs. The Federal Aviation Administration reduced air traffic by 10% because of the shutdown.

Deals

A Delaware judge denied Pfizer's attempt to block Novo Nordisk's Metsera bid...Shale producers SM Energy and Civitas Resources agreed to a $12.8 billion merger, including debt...Eaton acquired Boyd Thermal's liquid cooling unit for $9.5 billion... Coeur Mining bagged New Gold for $7 billion.... Starbucks sold 60% of its China unit to Boyu Capital for $4 billion... Kimberly-Clark agreed to buy Tylenol maker Kenvue for $49 billion, including debt... Charles Schwab is buying private-shares platform Forge Global for $660 million.

Next Week

Monday 11/10

Earnings reports from artificial-intelligence players will continue on Monday with "neocloud" company Coreweave reporting, followed by nuclear upstart Oklo on Tuesday and networking firm Cisco Systems on Wednesday. Noteworthy AI firms have reported mixed results in recent weeks, as some on Wall Street express concerns about surging valuations among stocks that could benefit from the AI buildout.

Tuesday 11/11

The major U.S. stock exchanges will remain open on Tuesday; bond markets will be closed for Veterans Day.

Thursday 11/12

Walt Disney will headline this week's earnings reports. Analysts polled by FactSet expect the firm to report September-quarter non-GAAP earnings of $1.05 a share on revenue of $22.76 billion. Shares have traded sideways this year.

Friday 11/13

Wall Street was slated to get key updates on inflation this week, but the shutdown has halted government data releases. The producer price index for October was slated for Friday, which would have followed the consumer price index on Thursday.

The Numbers

4

The number of months on average that regulators are taking to approve bank mergers, the lowest since 1990.

2

Estimate in percentage points of fourth-quarter U.S. growth lost if the shutdown lasts for eight weeks.

66%

The size of China's gross domestic product in 2024 compared with the U.S., down from 75% in 2021.

1.1 M

Challenger, Gray & Christmas estimate of U.S. layoffs this year so far, on par with the 2008-09 recession.

Write to Robert Teitelman at bob.teitelman@dowjones.com

This content was created by Barron's, which is operated by Dow Jones & Co. Barron's is published independently from Dow Jones Newswires and The Wall Street Journal.