What a heavy way to start this week. I took that selfie last week: with my daughter at her workplace, Bondi Beach Surffish. We’re enjoying the setting sun, enjoying the quiet tables by the beach and enjoying being together. A week later, other families were crouched under the same tables, hiding from gunfire from terrorist anti-semites. My daughter is fine, as are her coworkers and friends. But dozens of other families are not so lucky, with 15 dead and 27 casualties.

As an American, it's all too imaginable.

The Bondi community is in our prayers. And I have to wonder at the role that technology and algorithmic-social media may have played in urging on this hateful mindset. Our community must hold our tech leaders to a higher standard — or find new leaders and less pernicious technologies.

This week Micron Technology (MU: NASDAQ) reports. Memory is cyclical, capital-intensive and suddenly central to AI economics. Micron’s results won’t just close a quarter; they’ll frame 2026 expectations for supply discipline, pricing power, and whether AI demand is finally translating into durable margins rather than promises.

Last week’s corporate results were weird: they exposed a mismatch between performance and expectations. Broadcom (AVGO: NASDAQ), Ciena (CIEN: NYSE) and Oracle (ORCL: NYSE) delivered evidence of real AI demand, but also timelines and capital requirements the market appears unwilling to accept. That battle between those numbers and fear of a bubble could make for an interesting week ahead.

Have you checked out our website? https://epistrophy.beehiiv.com is a nice repository of these notes, with lots of perhaps-surprisingly still relevant research.

As always, I’m focused on three things:
1) Technology-driven change;
2) the latest in innovation and startup trends, and;
3) stock fraud.

Companies Discussed

In This Note:

1. Companies Discussed

2. Fear Of An Oracle Planet

   1. The Control of Cap Ex: the Oracle Way 

3. Epistrophy In The News

4. 📆 of Epistrophy Events

5. Availability This Week

Fear Of An Oracle Planet

What the Market doesn’t understand — how Oracle’s capital expenditures actually work.

Oracle (ORCL: NASDAQ) spending has increased materially. And the market doesn’t like it.

In the three months ended November 30, 2025, Oracle spent $12 billion on capital expenditures, up from $4 billion in the prior-year period. During the quarter, Oracle signed an incredible $68.3 billion in new contracts, bringing total remaining performance obligations to $523.3 billion.

That should be good news. It’s more business than any company has ever claimed in the history of time (check it out.) But the market treated the contract announcements as bad news, focused less on demand than on the cost of meeting it.

The concern is familiar. Large infrastructure investments made before revenue is recognized have historically produced long periods of underutilization and weak returns. In prior cycles, capital was deployed early, capacity arrived ahead of demand and returns never materialized.

The Dot Com crash still hurts. In the late 1990s, Enron Broadband Services wrote down billions of dollars in fiber and switching assets. WorldCom capitalized network spending that led to fraud. Global Crossing filed for bankruptcy in 2002 after building transoceanic capacity no one needed. Teligent collapsed after raising over $2.5 billion for a metropolitan fiber network that was never built. 

Those losses left an indelible mark on psyche. So today, the question is not whether Oracle can sign large customers. It is whether Oracle can afford it.

The Control of Cap Ex: the Oracle Way 

Here’s how Oracle’s capital expenditures work.

Oracle spends capital on equipment. This includes servers, accelerators, networking gear and storage. But under just retired-CEO Safra Catz – a sharp-witted one-time Donaldson, Lufkin & Jenrette investment banker and Oracle CFO – Oracle set up a clever scheme for its data center business. 

Oracle waits until the very last minute to buy equipment, close to the time the data center is ready and a customer is prepared to accept it. The goal is to reduce the time between paying for equipment – recorded as cap ex – and putting it to work. Revenue starts to flow then, as customers use the cloud services or as contracted capacity is delivered over time.

Oracle does not buy the land or the buildings. Those are leased. The leases typically cover the data center shell and the power infrastructure. Lease payments generally begin only when the facility is delivered and usable. So, again, they never show up in cap ex costs until they’re about to be deployed.

In this view, Oracle cap ex is a short fuse to an burst of revenue.

Governing the cap ex spending and the revenue are signed contracts. Oracle calls these “remaining performance obligations” – RPO. They represent work the company has agreed to perform in the future. The contracts come first. The equipment follows. The revenue arrives last.

Catz’s genius was to shorten the riskiest time of capital exposure, the period after the equipment is paid for and the customers start paying for it.  

This approach does not reduce the amount of capital required to build data centers. It is meant to reduce how long that capital sits idle. Whether it succeeds depends on how quickly customers begin using what has been built.

Oracle’s filings show how even these short-term expenditures are growing. 

In its fiscal 2025 Form 10-K, Oracle states that construction in progress “primarily consist[s] of computer equipment to be built and deployed at our data centers.”

Oracle also revised depreciation assumptions, increasing the estimated useful lives of servers and networking equipment from five years to six years, effective at the beginning of fiscal 2025. The change affects reported earnings, not cash outlays. That is, the income statement costs will be lesser as they’re be divided over more periods. 

This is a controversial change. But comments from NVIDIA (NVDA: NASDAQ) supported the longer life of older expensive chips, when CEO Jensen Hwang told the audiences at the 2025 GTC conference in March said: “There are circumstances where Hopper is fine. Not many.” Many customers have since admitted that, like a four-year-old Honda, older chips are perfectly fine for many tasks, underscoring that prior generations continue to see use even after successors arrive.

To be sure, there are concerns. 

Lease commitments aren’t their problem, yet.

But they do loom off-balance sheet. On balance sheet there are just $18.7 billion in operating lease liabilities. But in a 10-Q released after the close of the market on Thursday, Dec. 11, Oracle disclosed $248 billion of additional lease commitments, “substantially all related to data centers and cloud capacity arrangements,” with terms of fifteen to nineteen years. These commitments are generally expected to kick off no later than fiscal 2028.

That freaked out the market.

But if Oracle recognizes the lease commitments only just before they hand over the keys to a paying customer is that a risk or an opportunity?

Idle Time, Idle Minds

Oracle’s management frames capital efficiency as a timing problem.

In the fiscal first-quarter earnings call, then-CEO Safra Catz described the model directly:

“We do not own the property. We do not own the buildings. What we do own and what we engineer is the equipment. What we do is we put in that equipment only when it’s time and usually very quickly assuming that our customer accept it we’re already generating revenue right away.”

She avoided calling the model asset-light, but acknowledged the distinction:

“I don’t want to call it asset light from the finance world, but it’s asset pretty light.”

Founder and chief technology officer Larry Ellison added a specific example:

“We just turned over a giant data hall to one of our customers… The acceptance time could have been as long as a couple of months. It was one week.”

In the second quarter, Oracle sharpened the mechanics. EVP of Global Infrastructure and Cloud Operations Doug Kehring said:

“The vast majority of our CapEx investments are for revenue-generating equipment that is going into our data centers and not for land, buildings, or power that collectively are covered via leases… Oracle does not pay for these leases until the completed data centers and accompanying utilities are delivered to us.”

Equipment purchases are intentionally delayed:

“The equipment CapEx is purchased very late in the data center production cycle, allowing us to quickly convert cash spent into revenues earned.”

The Q2 call also introduced financing flexibility. Kehring said customers may bring their own chips and suppliers may lease accelerators rather than sell them, allowing Oracle “to synchronize our payments with our receipts and borrow substantially less than most people are modeling.”

Translation: even less cap ex on the Oracle cash flow statement. The chips are the customer’s problem. 

The accounting treatment did not change. Oracle’s 10-Q states that deferred cloud revenues represent customer payments made in advance, with revenues “generally being recognized ratably or based upon customer usage.”

As of November 30, 2025, Oracle reported $523.3 billion in RPO, up from $97.3 billion a year earlier. During the quarter, RPO increased by $68 billion, driven by contracts signed with Meta, NVIDIA and others. This should be seen as nothing but good news. Surely every other big AI provider, including Amazon Web Services (AMZN: NASDAQ), Google Cloud (GOOG: NASDAQ) and Microsoft Azure (MSFT: NASDAQ) wer all after those gigs. But they failed to land them. 

Betting against the Oracle sales team has historically been a mistake.

Remaining performance obligations provide the contractual context for what’s about to happen next on the income statement. 

Kehring said:

“The vast majority of these bookings relate to opportunities where we have near-term capacity available, which means we can convert the added backlog to revenue sooner.”

Oracle’s financing costs are not trivial, but they are not new. In fiscal 2025, Oracle reported $3.6 billion in interest expense, reflecting the cash cost of servicing its long-term borrowings. At recent balance-sheet levels, that implies an effective cost of debt in the mid-3% range. The company has carried substantial leverage for years. 

What has historically made that leverage tolerable is return. Over the past decade, Oracle has consistently generated returns on invested capital in the mid-teens, well above its cost of debt and always above its estimated cost of capital. That spread has underwritten buybacks, acquisitions and a capital structure that has long been more aggressive than most large software peers.

The current build-out tests that history. Capital is rising faster than revenue, free cash flow has turned negative, and returns will compress before they recover. Whether Oracle’s cloud investments preserve its historical ROIC profile depends not on financing costs, which remain manageable, but on how quickly the new capacity moves from installed to utilized.

Yes: the income statement still lags the strategy. Capital expenditures are immediate. Lease commitments are disclosed but not expensed until delivery. Revenue is recognized over time. Negative free cash flow during the build phase is structural – and makes perfect sense. 

Oracle’s approach is more capital-efficient than the worst historical precedents. Unlike the crooks of the Dot Com era, it is not capital-light.

Whether it works depends on utilization and duration, not on the quarter in which capital expenditures peak. Perhaps the market will figure that out.

Epistrophy In The News

On Schwab Network, I walked through why solid results from Broadcom, Ciena, and Oracle were met with selling—and why the market’s reaction says more about expectations than execution.

On NewsNation, the conversation widened. We examined the structural problems in President Trump’s executive order attempting to block state AI laws—federalism, enforcement, and unintended consequences. And, who knew, I got to tell my 45-year old story about hacking into the Miracle On Ice Olympic hockey team. It’s a funny one, check it out at 4:45 into this clip.

📆 of Epistrophy Events

Availability This Week

I’m in San Francisco all week. Feel free to text for time-sensitive items; email works too, replies later in the day.

Written reports are available to clients, with video summaries on YouTube, and of course our popular summaries of the summaries on Instagram, TikTok, and YouTube Shorts.