Skip NavigationMarketsBusinessInvestingTechPolitics & PolicyVideoWatchlistInvesting Club
PRO
LivestreamMenuThe artificial intelligence spending boom is testing one of Big Tech’s greatest strengths: pristine balance sheets. Club holdings Meta Platforms , Amazon , Alphabet and Microsoft entered the AI race as some of the world’s healthiest companies and most desirable borrowers, armed with enormous cash on hand and little debt relative to their equity, known as leverage. These hyperscalers’ rock-solid balance sheets made their stocks safe havens during periods of economic uncertainty and market volatility, such as the Silicon Valley Bank crisis in 2023. But after years of accelerating investment in data centers and AI servers, those once-enviable financial profiles are beginning to show some strain. Capital spending has climbed above 100% of operating cash flow at some of these hyperscalers, pushing them to outside financing, including traditional debt sales and more creative “off-balance sheet” arrangements. Amazon and Alphabet have tapped debt markets around the globe; Alphabet also issued equity. Meta has also issued bonds and is leaning into strategic ventures with alternative asset managers to help fund data center projects. This evolution in AI investment raises an important question for investors: How long can the hyperscalers maintain this pace of spending without weakening their credit ratings and risking higher borrowing costs in the future? “As leverage gets worse, their credit quality comes down,” said Naveen Sarma, analyst at S & P Global. While deteriorating credit quality doesn’t necessarily mean a company will default, “it becomes more expensive for them to finance debt,” Sarma said. That’s a key reason why stock market investors need to care about their portfolio companies’ credit ratings, which are issued by S & P Global and two other independent firms, Moody’s and Fitch. The goal of credit ratings is to provide a relative ranking system that reflects the likelihood that a company will pay its debt on time and in full. The agencies generally look at a range of metrics to inform their evaluations, including cash flow and leverage ratios, which measure debt relative to a company’s equity and operating earnings, among others. S & P Global, in particular, also uses qualitative factors, such as a company’s competitive position and management track records. As the AI buildout progresses, S & P Global is watching the stability of these metrics, and Sarma said some companies are beginning to move closer to thresholds that could eventually pressure their ratings. Moody’s struck a similar tone in a recent report , arguing that the hyperscalers’ AI buildouts “threaten credit quality,” due to “unprecedented levels of investment and capital raising.” To be sure, Alphabet, Amazon, Meta and Microsoft still maintain very high credit ratings, and their underlying businesses are performing well, as evidenced by strong growth in operating cash flow during their respective June quarters. That’s also when Amazon CEO Andy Jassy expressed confidence that the company’s aggressive investments will pay off, eventually driving long-term revenue and free cash flow. “We would prefer to see our hyperscalers get back to positive free cash flow and restart share repurchases, but if Jassy is correct on the cash flow timeline, the payoff should be significant in the future,” said Jeff Marks, the Club’s director of portfolio analysis. These hyperscalers still have “really good balance sheets,” added S & P Global’s Sarma. “Even with all the debt that they’ve issued, including all of our adjustments, leverage isn’t blowing up,” he said. Their balance sheets are on much stronger ground than another tech giant spending aggressively on AI infrastructure: Oracle . S & P Global downgraded Oracle’s credit rating in July to the lowest tier of investment grade. “Oracle’s AI strategy carries material credit risks given the aggressive spending and uncertain path to positive cash flow,” the firm wrote at the time. Cash flow crunch For the Big Four hyperscalers, this past earnings season did amplify concerns about their balance sheets. That’s because management teams’ commentary suggested the AI investment spree is not about to end, despite existing pressure on their free cash flow (FCF) — the money remaining after a company covers its daily expenses and invests in new equipment and property (capital expenditures). That leftover money can be used to pay down debt, though the figure already accounts for interest expenses. In the second quarter, Meta’s FCF plunged 91% from a year earlier as capex jumped 88% to $31.1 billion. After its capex doubled to $44.9 billion, Google parent Alphabet posted its first-ever quarter of cash outflows since going public in 2004. Meanwhile, Amazon joined Alphabet in negative FCF territory during the three months ended in June, posting a cash outflow of $8.8 billion after capex jumped 68% year over year to $54.21 billion. The relative standout is Microsoft, which saw FCF decline 23% to $19.6 billion, despite capex more than doubling. These companies continue to argue that the enormous demand for AI computing capacity justifies the investments, as cloud growth has accelerated across several major platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure, underscoring that customer demand for AI tools and services continues to exceed the available AI infrastructure. Meta doesn’t have a traditional cloud business like its competitors, but is considering selling excess computing power to outside customers. Nevertheless, at this stage of the AI boom, the sheer scale of the buildout and the pressure it’s putting on cash flows have created the need for outside funding. Amazon, Alphabet, Meta and Microsoft are estimated to spend roughly $960 billion on capex in calendar 2027, according to FactSet’s consensus estimates. Their combined operating cash flow, on the other hand, is projected to be around $905 billion. Only Microsoft is projected to have positive free cash flow in 2027. Borrowing increases The hyperscalers have indeed become major borrowers in the bond market this year — so much so that the flood of issuance is cited as one reason for the recent rise in government bond yields, which move inversely to prices. The belief is that some money that historically would’ve been used to buy sovereign debt is being redirected toward high-quality corporate bonds, as government yields rise to stay competitive. Take Alphabet, which has issued debt on multiple occasions and in multiple currencies. Last week, the company sought $3.6 billion in its first Australian bond sale. And earlier in August, it closed a $25 billion senior note sale . In February, the company also issued a rare 100-year bond , denominated in sterling, the UK currency, with a maturity in 2126. The company’s long-term debt — anything that takes more than a year to pay off — stood at roughly $115 billion at the end of June, more than tripling from a year earlier. Alphabet has also sold equity this year, unlike Meta, Microsoft and Amazon thus far. For its part, Amazon tapped the bond market in July for roughly $25 billion after raising about $64 billion earlier in the year across the U.S., Europe and Canada. The e-commerce and cloud giant ended June with $222 billion in long-term debt, up 67% from a year earlier. But CNBC’s David Faber, citing sources close to the matter, reported that Amazon told its underwriters it won’t issue any additional debt in 2026. In April, Meta sold investment-grade bonds worth $25 billion , which came on top of a roughly $30 billion debt offering in the fall of 2025. At the end of June, Meta had long-term debt of roughly $110 billion, up 131% year over year. Microsoft is the outlier of the bunch. The company has not sold any bonds since 2024, according to FactSet data. However, Microsoft has signed significant long-term data center leases — and those are considered liabilities under accounting rules, given the company is obligated to make regular lease payments. Microsoft’s long-term debt was up 9% year over year at the end of June, to $110 billion. Other financing approaches Another method to finance additional capacity is through strategic partnerships. Meta has leaned heavily on these alternative structures. Its new venture with BlackRock to develop a roughly $14 billion data center campus in El Paso, Texas, allows the company to secure additional computing capacity without funding the entire project itself. BlackRock will own 80% of the venture and Meta 20%, with Meta leasing the campus. The structure reduces Meta’s upfront capital burden, which illustrates why credit analysts are looking beyond traditional debt to long-term leases and other commitments when assessing financial health. Meta also has a joint venture with Blue Owl Capital for its sprawling “Hyperion” data center in Louisiana. Even AI kingpin Nvidia partnered with Wall Street heavyweights to establish financing platforms to mobilize $500 billion in third-party capital for the AI infrastructure buildout. In June, fellow Club chipmaker Broadcom partnered with asset managers in Apollo and Blackstone on a funding platform for AI infrastructure. CNBC’s Faber reported last week that Broadcom was looking to the bond market for additional money to finance chip purchases. In these instances, Nvidia and Broadcom appear to be backstopping some portion of the loans, rather than issuing the debt outright themselves. The developments show the enormous amount of capital being directed toward AI. “The hyperscaler industry has natural limits around debt levels and power agreements, and we seem to be approaching those limits in ’27,” analysts at Barclays wrote in an Aug. 13 note to clients. That has opened the door to ventures like Nvidia’s wide-reaching Wall Street partnership, and Broadcom’s June venture with Apollo and Blackstone. Not all the same The pressure isn’t equal across Big Tech. S & P Global’s Sarma sees Oracle as the most vulnerable of the five major hyperscalers because the company has significant leverage while burning through cash. Oracle has had negative free cash flow in five straight quarters, including for the three months ended in May . On July 9, S & P Global cut Oracle’s credit rating and sees the company’s credit as low investment grade. S & P now has a BBB- rating on Oracle; AAA represents the highest quality and lowest risk. Of the four Club-owned hyperscalers, S & P Global has the lowest rating on Meta, at AA-. That’s still considered very strong and investment grade. Sarma told CNBC that something to watch with Meta is its future lease commitments, which ballooned from roughly $180 billion last quarter to about $280 billion at the end of June. Meta’s leases for future data centers are “going to be a liability in a couple of years,” Sarma said. “That’s a concern.” Meanwhile, Microsoft and Alphabet have considerably more financial capacity and therefore have far less concern about their credit quality, according to Sarma, who gave Microsoft an AAA rating and Alphabet an AA+. S & P Global has an AA rating on Amazon. Despite deteriorating hyperscaler cash flows, some Wall Street analysts aren’t sounding the alarm just yet. Lloyd Walmsley, an internet equity analyst at Mizuho Securities, said he feels good about Meta’s balance sheet. He argued the Facebook and Instagram parent has several ways to generate returns on the enormous amount of computing capacity it’s building. That includes renting out some of the capacity to other companies, something Jim Cramer has urged the company to do. “We are not concerned right now,” Walmsley said. “They have a lot of optionality where they can convert it into operating cash generation effectively,” he said. Walmsley said the fundamental problem for investors is “not how they’re financing, as much as what is the return on investment on the spend.” He said that what would make investors more comfortable is Meta striking deals to rent out capacity to other AI labs, such as Anthropic or OpenAI. “It’s critical that they show investors they can generate a return on some of this capacity,” in the short term, Walmsley said. S & P Global analysts agreed that the ultimate question is whether AI investments generate sufficient returns, and what happens to their businesses before they arrive. Assessing credit risk requires analysts to look out years ahead, said fellow S & P Global analyst David Tsui. “We are tracking the qualitative part of the business and quantitative credit metrics, which is clearly deteriorating,” Tsui said. “Maybe they have a cushion today, but if you look out two to three years, maybe they’ve breached their downgrade threshold. That’s when we start signaling there’s a threat of a downgrade,” Tsui said. What “gets us nervous,” Tsui continued, is the pace of capex outpacing revenue growth and profitability , and the pressure it puts on free cash flow. “That will have an impact on how much cushion they have within the current rating to determine whether or not we’re closer to a rating downgrade,” Tsui explained. The real test may come over the next two years. Sarma said Wall Street broadly expects an inflection in returns on AI investments around 2028, leading to an uplift in revenue, earnings and cash flow. But if that inflection doesn’t happen and Sarma observes that “these guys are still spending as much as they’re doing, then we’re going to have much more serious credit issues.” (Jim Cramer’s Charitable Trust is long AMZN, META, MSFT, GOOGL. See here for a full list of the stocks.) As a subscriber to the CNBC Investing Club with Jim Cramer, you will receive a trade alert before Jim makes a trade. Jim waits 45 minutes after sending a trade alert before buying or selling a stock in his charitable trust’s portfolio. If Jim has talked about a stock on CNBC TV, he waits 72 hours after issuing the trade alert before executing the trade. THE ABOVE INVESTING CLUB INFORMATION IS SUBJECT TO OUR TERMS AND CONDITIONS AND PRIVACY POLICY , TOGETHER WITH OUR DISCLAIMER . NO FIDUCIARY OBLIGATION OR DUTY EXISTS, OR IS CREATED, BY VIRTUE OF YOUR RECEIPT OF ANY INFORMATION PROVIDED IN CONNECTION WITH THE INVESTING CLUB. NO SPECIFIC OUTCOME OR PROFIT IS GUARANTEED.Read More
About The Author
