If you asked ChatGPT a question this week, the reply probably landed in seconds. Behind it sat racks of chips, cooling systems, and power lines someone paid for long before you typed.
Oracle Corporation (ORCL) is one of the companies that pays first and collects later.
On Oct. 27, Oracle lined up more chips for that business. Its cloud will offer processors from Alphabet’s (GOOGL) Google, Nvidia Corporation (NVDA), Intel Corporation (INTC), and Cerebras Systems (CBRS). It also pulled OpenAI deeper into its business software.
Yet none of these deals answers the question investors bring to Oracle’s Investor Day on Oct. 28: how Oracle pays for everything it has promised to build.
Also read: Oracle CEO admits its software can take years to get running
Evercore ISI estimates that Oracle needs $50 billion to $60 billion in additional funding, according to Seeking Alpha. The firm sees that need lasting until free cash flow turns positive in fiscal 2029.
Oracle’s building plans are running ahead of the cash its business brings in.
The new chips come with new bills
The Google deal shows the tradeoff. Oracle said its cloud will be the first large provider outside Google to offer Google’s custom chips, called TPUs, starting in 2027. The same press release says Oracle will supply the power, cooling, and data center networking to run them.
The release does not say who pays for the chips themselves. Either way, Oracle must house and power several kinds of hardware.
Oracle also plans clusters measured in gigawatts on Nvidia’s Vera Rubin systems, plus a multiyear Intel Xeon deal, according to Seeking Alpha. Nvidia’s Ian Buck said the aim is to help customers “put more of their infrastructure to work today and sustain its value as AI evolves.”
Meanwhile, CNBC reported that Oracle’s cloud will also host OpenAI’s Jalapeño chip. Mahesh Thiagarajan, president of Oracle Cloud Infrastructure, told the network that chips remain scarce and customers like OpenAI benefit from having options.
I read that as Oracle protecting its supply, which is sensible.
Yet Thiagarajan also called Nvidia’s chips “the best, most fungible,” meaning the easiest to shift between models and workloads.
Hardware must stay useful long enough to pay for itself. By his own measure, the other accelerators joining Oracle’s cloud are less flexible bets.
Cristi Croitoru / Getty Images
The market already values the AI cloud below zero
Those bets come while investors already doubt Oracle’s AI cloud. Evercore valued Oracle’s software and non-AI cloud businesses at about $197 per share after accounting for debt. By Evercore’s math, the market assigns Oracle’s AI cloud a negative value.
If investors see the AI cloud as a drag, more of it cannot fix that. Evercore said investors need “a clearer picture of the remaining funding needs.”
Oracle has already leaned on equity investors. The company sold $20 billion of common stock during its fiscal first quarter, according to its earnings release. New shares spread future profits across more owners.
Oracle has forecast $90 billion to $95 billion in capital spending for fiscal 2027, according to Seeking Alpha.
The spending plan roughly equals the $90 billion or more in revenue Oracle expects this fiscal year, according to its earnings release. In other words, Oracle plans to spend about as much on building as it collects.
Oracle says it is structuring new deals to ease that strain. Its earnings release said more than $30 billion in AI contracts booked last quarter added nothing to its capital-raising plans. That covers new contracts, not those already driving the bill.
OpenAI moves into Oracle’s safer business
The OpenAI announcement matters most to me because it reaches the businesses Evercore says carry Oracle’s value. OpenAI said Oracle’s applications and data are coming into ChatGPT, and OpenAI models will power new Oracle products.
Oracle’s own recruiting team now prepares some market research in 15 to 20 minutes instead of two to four days, OpenAI said. Results like that become Oracle’s sales pitch.
Thiagarajan told CNBC that using ChatGPT “allows anybody to build an application, deploy it, manage it.” Software sales do not need a new data center for each customer, making this the cheapest growth Oracle announced on Oct. 27.
However, the catch is concentration. OpenAI is now Oracle’s compute customer, the designer of a chip headed to its cloud, a partner in its apps, and a vendor to its staff. Evercore said investors want clarity on Oracle’s contracted backlog excluding OpenAI, a sign the market treats that link as a risk.
Reports in April that OpenAI fell short of several sales targets showed how quickly that link can turn. Tying Oracle’s steadier software business to the same partner raises the cost of any OpenAI stumble.
More Oracle:
- CoreWeave, Oracle have an interest rate problem
- Oracle boss Larry Ellison just suffered a historic wealth collapse
- Oracle’s ‘force majeure’ warning lands as AI power gets scarce
Investors want a funding plan, not more chips
The recent deals also hint at who will carry the AI buildout’s cost. Google placing its chips in a rival hyperscaler’s cloud for the first time suggests chip designers want reach without building every data center. Someone still owns the power, the cooling, and the debt.
Oracle has volunteered for that role. It works if customers keep paying and help fund the build. Evercore listed customer funding, alternative financing, and possible data center delays as details investors will watch.
The Oct. 28 test is simple. Oracle needs to put its own number on the funding gap and show who fills it.
The next phase of the AI boom may reward companies that finance data centers more than those that fill them with the most chips.
More Stocks News
- Celestica sees 2027 sales nearly doubling to $40 billion
- What Jim Cramer is watching as Anthropic heads to Nasdaq
- ServiceNow admits its help desk bots are after payroll dollars
- Deutsche Bank just put a hard number on AI productivity
