AI infrastructure.
Built on credit.

In 2023, Amazon, Alphabet, Microsoft, and Meta put about 41 cents of every dollar of operating cash into capital spending. In the first half of 2026, they put in 91 cents. Amazon spent more than it brought in.

What they're buying is AI infrastructure, the physical layer under every model: the chips, the memory, the servers, the data centers, and the power plants that keep them running. Those four, plus Oracle and the big AI labs, pay for most of it, and none of them is slowing down. Alphabet has already said its 2027 spending will be "significantly" higher than this year's.

The spending has nearly outrun the cash, so they've started borrowing. The big four's combined long-term debt went from $132 billion at the end of 2024 to $342 billion by June, and investors have noticed. Alphabet fell 7% the day after it raised its spending plan. That's the shift to understand about AI infrastructure stocks going into 2027. The buildout used to be paid for out of profits. Now it runs on credit.

When a boom runs on borrowed money, where you sit in the payment chain matters more than which part of the stack you build. So this list of AI infrastructure stocks runs from the vendor that lends to its customers, through the suppliers who get paid first and the power sellers locking in decades of revenue, to the builder carrying the most debt and the two private labs whose IPOs would refinance the whole thing.

The biggest seller in AI has started acting like its customers' bank.

Nvidia NASDAQ: NVDA

Nvidia still sells more AI hardware than anyone. Revenue hit $96 billion last quarter, and it guided to $108 billion for the next one. The more important number is buried in an August filing: Nvidia agreed to guarantee up to $105 billion of lease payments on an Ohio data-center campus leased to OpenAI for 20 years.

That isn't a one-off. Nvidia's stakes in private companies more than doubled this year to about $51 billion. It's also giving big customers longer to pay, which pushed the time it takes to collect from 45 days to 60. Nvidia has figured out that the fastest way to sell more chips is to help finance the buyers.

When a boom runs on credit, the safest seat belongs to whoever gets cash on delivery while the debt sits on somebody else's balance sheet. These three are the AI infrastructure stocks sitting in it.

Micron Technology NASDAQ: MU

Micron sells the one part every AI system is short of, and it's being paid like it. Fiscal 2026 revenue more than tripled to $133 billion. Next quarter's guide is $61.5 billion at a gross margin around 86%.

The hyperscalers' debt piles up on their side of the ledger. Micron's side just shows cash.

Dell Technologies NYSE: DELL

Dell builds the racks the chips go into, which makes its order book the most direct read on what's actually been bought. It took $60.9 billion of AI server orders in a single quarter and is sitting on a $95 billion backlog.

It's the closest thing in public markets to watching the buildout's purchase orders in real time, and those orders turn into cash as the racks ship. Dell now expects revenue to grow 69% this fiscal year.

GE Vernova NYSE: GEV

GE Vernova makes the big gas turbines that power plants and data-center campuses are ordering faster than anyone can build them. Its gas backlog plus reserved production slots reached 116 gigawatts by the end of June. Turbine output is set to reach 20 gigawatts a year this quarter, so that's roughly six years of production already spoken for.

Customers are signing slot reservations just to hold a place in that line. The cash shows up early, too. Second-quarter free cash flow was $5.1 billion, more than the company generated in all of 2025.

Borrowed money makes time expensive. A finished data center waiting on electricity is a pile of interest payments, and these two sell the fix.

Constellation Energy NASDAQ: CEG

Constellation runs the largest nuclear fleet in the U.S. and has been selling its output to tech companies on 15- to 20-year contracts. It added another 920 megawatts of those deals this summer, starting between 2029 and 2032.

The beauty of that structure is who's on the other side. Constellation isn't betting on an AI startup's revenue. It's selling power to some of the strongest balance sheets on Earth, at a fixed price, for decades.

Bloom Energy NYSE: BE

Bloom Energy sells fuel cells that sit on the data-center site and produce power without waiting years for a grid connection. For a builder paying interest on idle chips, that's worth a premium. Revenue grew 166% last quarter to just over $1 billion, and Bloom says every major U.S. hyperscaler has now approved its systems.

One company decided to win AI infrastructure by borrowing against the future, and it's the purest test of whether that works.

Oracle NYSE: ORCL

Oracle has signed so many AI cloud contracts that its backlog of future revenue hit $664 billion. To deliver it, Oracle is spending at a pace it can't fund from operations. It spent $28.5 billion on capital projects last quarter, burned $5.4 billion of free cash flow, carries about $125 billion of debt, and sold $20 billion of new stock.

If the contracts turn into revenue on schedule, Oracle ends up with one of the biggest AI clouds in the world. If a major customer stumbles, Oracle has the least room to absorb it of anyone on this list. Management's answer is to make customers carry more of the load, through prepayments and bring-your-own-hardware deals, and the next few quarters will show whether that holds.

The two labs that sit at the end of most of these contracts both filed confidentially to go public this year.

OpenAI

OpenAI is the single biggest buyer in AI infrastructure, with commitments spread across Oracle, Microsoft, Nvidia, AMD, Broadcom and Cerebras. It confirmed a confidential draft IPO filing in June, after a private round that valued it at a reported $852 billion.

That listing matters to everything above. A public OpenAI could fund its commitments with public equity instead of leaning on suppliers' guarantees.

Anthropic

Anthropic filed its own confidential draft in June, after a raise at a $965 billion valuation. What makes it unusual is who owns it. Amazon and Google are investors as well as suppliers, AMD has agreed to invest too, and Amazon alone booked about $53 billion of gains last quarter, mostly from its Anthropic stake.

An IPO would put a public price on that circle. If it holds up, the hyperscalers' paper gains become real. If it doesn't, some of this year's earnings were just marks.

In May, Meta sold bonds that don't mature until 2066. The chips that money helps pay for will be obsolete well before 2031.

That's the gap that will shape the next leg of the buildout. Data centers last decades. Accelerators turn over every few years. As long as the buildout was paid for in cash, nobody had to match the two. Now lenders do.

Expect the money to start looking like the asset. That means more financing tied to specific chips, more vendor leases, more prepayments and more joint ventures that keep the hardware off the hyperscalers' books. Lenders will price each piece on how fast its hardware depreciates, and AI infrastructure stocks will start trading on the same math.

So the pace of AI infrastructure in 2027 won't just be set by how much demand the labs see. It'll be set in credit committees, by people deciding what a three-year-old GPU is worth as collateral.