Tl;dr – Oracle remains the x-axis of AI/cloud computing (but their revenue forecasts are geometric)
Update: Oracle reported earnings yesterday, missing on both revenue and profit expectations and reporting slower growth than the much larger Amazon, Google, Meta and Microsoft. Free cash flow dropped deeper into the red. Oracle’s CAPEX spend dropped from $9.2 billion to $8.5 billion. But they announced they have signed contracts for nearly geometric increases in cloud revenue over the next five years, going from $10.3 billion in the last fiscal year (FY25) to $144 billion in FY30. The stock is up insanely as a result.
The discussion of the CAPEX required to realize these revenue forecasts was minimal. Oracle guided to $35 billion in CAPEX for FY26 and threw around some words about how they’re “asset pretty light”. Oracle’s belief in magic beans abides. But no infrastructure, no revenue remains an iron law of the cloud world. Oracle still needs to come up with a lot of capital for CAPEX, amidst other demands for capital (dividends, buybacks to raise Larry’s stake, servicing debt).
Oracle’s future is tied to Open AI. The vast majority of the forecasted revenue would come from Open AI. It isn’t clear how Open AI will find hundreds of billions to pay for infrastructure and all their other ambitions. The other customers referenced are neither big nor reliable revenue. xAI has no revenue, no prospects for real revenue, and is already turning to debt funding (plus borrowing from SpaceX). Meta has no business plan to pay for their ($600 billion and counting!) AI mulligan, but likely promised some very expensive employees they’d have GPUs to play with. NVIDIA and AMD paying Oracle for infrastructure is simply revenue recycling.
Oracle still has to come up with the cash to play this game. They’re promised the revenue. Now they need to show us the CAPEX to realize that revenue.
Update 2: WSJ confirms the vast majority of Oracle’s promised future revenue ($300 billion) would come from Open AI. Oracle stock is now a one-way bet on Open AI’s ability to raise hundreds of billions in more capital.
Update 3: I just went ahead and wrote another post about Remaining CAPEX Obligation.
Absolute bestest most glorious company ever in entire universe of all time prepare awesome-sauce victory lap. With tomorrow’s earnings announcement the database vampire will permit us to bask in their greatness and invite us to contemplate why other companies even exist. Things are always so fabulous that the Oxford English Dictionary is considering replacing the word “superlative” with “Oracle”.
Beyond bombastically scripted earnings calls, deploying CAPEX effectively and efficiently at scale takes place in a cold, hard reality. We’ve been chronicling Oracle’s difficulties building data centers at bonsai scale, never mind at hyperscale:
- Why Can’t Oracle Build Data Centers?
- Oracle Still Can’t Build Data Centers
- Oracle’s Data Center Difficulties
- Oracle’s Data Center Difficulties: FY25 Q1
- Oracle’s Data Center Difficulties: FY25 Q2
- ClownWatch™: Oracle FY25 Q3
- Why Can’t Oracle Build Data Centers? FY25 Q4 Edition
- Words are Cheap, CAPEX is Expensive: Oracle Edition
- Why Can’t Oracle Build Data Centers? (Or Subcontract Them?)
- Follow the CAPEX: The Clown Car Race Checkered Flag
Oracle (company slogan: “that’s not just a rack, it’s an entire region!”) has slowly increased its CAPEX spend, but at a pace that falls further and further behind the hyperCAPEX companies.
Oracle’s first decent-sized data center, the oft-touted yet somehow still under construction project in Abilene, is well on its way to becoming the picture accompanying the dictionary definition of “construction site”. Partly it earns that status because it has taken so long, but also because that data center has so many mouths to feed (Blue Owl, Primary Digital Infrastructure, Lancium, Crusoe, Oracle, and Open AI are all laying claim to those construction photos). Oracle’s primary contribution seems to consist of yelling at the contractors.
Oracle is “all-in” on their data center plans, and has expended enormous energy to get themselves referred to as a “hyperscaler”. But are they really in the hyperscaler club?
What Does Hyperscale Look Like?
Let’s see how Oracle stacks up with actual hyperscalers. This is a high-stakes financial game requiring gargantuan resources and a broad array of competencies. When it comes to the metrics, Oracle always seems to play the role of the x-axis (a joke I never tire of and Oracle never manages to break away from that gravitational pull). Let’s look at some proxies for financial prowess:
Oracle is the revenue runt of their aspirational litter. By a lot.
Oracle’s cloud infrastructure business is the runt, except for Meta who don’t have a cloud business.
(One of the great mysteries of the universe is where Meta sees the return coming from its rapidly escalating infrastructure investment promises — $600 billion and counting!. Unlike our other contestants, they have no cloud services business and a checkered history serving developers. As with Elon, taking a dependency on Meta is high risk. Maybe Meta alone has glimpsed some new zero-to-one geyser of revenue that will pay for all those data centers? Regardless, and reluctantly returning to today’s topic at hand, Meta still has vastly more cash to play this game than Oracle).
Three-for-three on runt status.
Oracle does show leadership on one metric: the debt to equity ratio. They’re already carrying a much bigger debt load than their desired peer group, while also the company that could most use some credit.
For free cash flow (cash flow minus CAPEX), Oracle is yet still again the runt. Oracle’s competitors are both spending much more on CAPEX and still have significant, positive free cash flow.
And when we look at the most recent quarter, Oracle’s free cash flow went negative (-$394 million) as they ramped their CAPEX spend to (merely) $9 billion in the quarter.
Now let’s look at CAPEX spending, that ultimate separator of clouds and clowns.
Emphatically runty.
Oracle is already more stressed with their (modest) CAPEX spend relative to revenue.
Cumulative CAPEX spending gives us a sense of how much infrastructure a company has deployed. Amazon, AWS, Google, Meta and Microsoft all spent more on CAPEX in the last year than Oracle has in the last decade, and all but Meta spent more last year on CAPEX than Oracle has in its entire fifty year existence.
All of which is to say Oracle is both way behind in deployed infrastructure and playing with the smallest stack of chips in this game.
Show Us The (CAPEX) Money
Which brings me (at long last!) to my point: Oracle has hyped big revenue growth forecasts for the next couple years. In particular, they tout a deal with Open AI to bring in as much as $30 billion in incremental revenue by FY28 (we’ll ignore the question of why Microsoft decided it didn’t want this business and discarded it to Oracle).
If you’re excited about the revenue opportunity in front of Oracle, you also have to grapple with the costs of the infrastructure necessary to deliver that revenue. No infrastructure, no revenue. No CAPEX, no infrastructure.
So where does the money for data center CAPEX come from and what are the implications for Oracle’s free cash flow, debt load and/or margins? They need to step up from their current (modest) levels of spending which are already stressing the company’s financials.
It is pretty clear Oracle is not very efficient in its CAPEX spend. They’ve been publicly demonstrating they are early on the learning curve, struggling to forecast CAPEX spending a quarter out, much less years out. They’re also in a big hurry, and there is a YOLO infrastructure cost premium. Some hardware is scarce (GPUs, gas turbines). Costs and wait times for both interconnection and hardware delivery have widened, while the hyperCAPEX companies are both bigger buyers and got in the queues earlier. End-to-end, Oracle lacks a manufacturing machine that turns dollars into data centers efficiently and effectively at scale.
Over the last three years, Oracle has spent between $1 and $2 of CAPEX per dollar of cloud infrastructure revenue. AWS CAPEX historically has been between 25 to 60% of revenue (and is ramping back up as they deploy AI infrastructure).
We could also model CAPEX spend based on the incremental revenue it drives (new capacity allows new revenue), which makes that $30 billion revenue dream even more expensive (and remember those GPUs are short-lived, especially if you’re chasing the pre-training market).
It looks like Oracle is going to lean more on OPM (Other People’s Money or perhaps Oracle Pleads for Money) for its CAPEX spending. The data center build team was hit in the recent Oracle layoffs. The Abilene data center is already a layer cake of financial participants.
Stargate was briefly the OPM salvation, but it has “struggled to get out of the gate”. Open AI finds itself doing party rounds for both cash and electricity, raising billions of dollars instead of trillions, and scavenging for megawatts rather than gigawatts.
Maybe those notoriously charitable financial actors will accept lower returns just to partner with notoriously charitable Oracle? Going the OPM route negatively impacts Oracle margins, architectural control, and time-to-market.
How do Oracle’s hyperscaler hopes and dreams get bankrolled?
