The Missing $1.65 Trillion and Oracle's Rating Cliff: An Ex-Oracle Engineer Looks at AI Debt from Wall Street

Over the past few days, several friends in tech and secondary market investing asked me the same question:

"Henry, Oracle (ORCL) is on an absolute tear with OCI cloud growth crossing 50%. Is it a safe, long-term buy-and-hold? Or are we chasing an over-leveraged bubble?"

Hearing that brought a smile to my face—and a flood of memories.

Those who know me know that I used to write code at Oracle years ago, living the company's fierce engineering and sales DNA; today, I sit in a Wall Street investment bank navigating quantitative risk and structured financial engineering. Over the weekend, while writing Appendix F for my book Mortgage Markets in the Age of AI, I had the chance to systematically dissect the debt filings, offshore SPVs, and rating agency dossiers across the major hyperscalers.

Looking at both the underlying engineering and the corporate balance sheet: where does Oracle's real tactical advantage lie in this AI cycle? And what are the critical debt risks that demand close attention?

Here is a grounded, insider analysis of the mechanics powering Oracle's AI rally—and the trillion-dollar off-balance-sheet debt game unfolding across Silicon Valley.

---

1. Why the Bull Case on Oracle Is Mechanically Strong¶

Anyone who spent time at Oracle understands Larry Ellison's fundamental nature: he never plays for incremental gains. He bets the entire house on the biggest game in town, designed specifically to crush the competition on raw performance. That was true for Exadata database appliances a decade ago; it is even truer for Oracle Cloud Infrastructure (OCI) today.

The bull thesis on Oracle rests on three solid, compounding pillars:

Pillar 1: OCI Bare-Metal Clusters Solve the True LLM Bottleneck¶

Many outside tech assume cloud compute is just rented virtual machines. But distributed LLM training isn't limited by single-chip throughput—it is bottlenecked by inter-GPU network latency and communication overhead.
- While AWS and Azure spent years optimizing traditional virtualized cloud instances, OCI went all-in on bare-metal GPU servers connected via RDMA ultra-low-latency clustering;
- This was a strategic masterstroke. Elon Musk's xAI chose OCI to train Grok across tens of thousands of H100s; OpenAI signed massive compute commitments; even Microsoft Azure turned to Oracle to rent overflow capacity when its own data centers ran dry. OCI revenue growth is comfortably cruising above 50% year-over-year, with a monumental backlog of Remaining Performance Obligations (RPO).

Pillar 2: The Multi-Cloud "Trojan Horse" Strategy¶

For thirty years, the mission-critical transaction and financial data of the Fortune 500 has run on Oracle Database. Migrating away is prohibitively expensive and risky.
Larry's genius move over the past two years was to embed Oracle hardware and database engines directly inside the physical data centers of Microsoft Azure, AWS, and Google Cloud (Oracle Database@Azure/AWS/GCP):
- Enterprise clients don't have to migrate—they spin up Oracle databases inside Azure with two clicks;
- Oracle effectively hijacked the distribution rails of its chief cloud rivals to lock in its high-margin software revenues.

Pillar 3: The Indispensable Financial Backbone: You Can Leave Oracle, but Wall Street Never Leaves Oracle (Laughs)¶

To share a very personal observation: the very database architectures, high-availability RAC setups, and tooling I worked with back at Oracle years ago? When I made the jump to Wall Street and logged into the bank's core trading systems, there they were—still powering the mission-critical ledgers, clearing pipelines, and risk engines across Manhattan.

I often joke with my industry friends: "You can take yourself out of Oracle, but in the financial world, you can never truly leave Oracle (laughs)."

When the operational nervous system of the entire Fortune 500 and the global banking apparatus is deeply hardwired into your relational engine, the switching costs are practically infinite. That sticky, unbreakable enterprise foundation is the deepest moat in tech—and the ultimate cash-generating backstop for Larry's aggressive AI bets.

---

2. A Sober Look at the Balance Sheet: Navigating the Leverage Tightrope¶

However, any grounded analysis must look at both sides. Understanding the potential downside and margin of safety is what keeps a thesis objective and resilient.

From a credit and corporate finance perspective, Oracle took on significant financial leverage to fund this aggressive expansion:

The BBB- Rating Cliff: Half a Step from Junk¶

Unlike Microsoft (AAA rating, fortress cash reserves) or Google (virtually zero net debt), Oracle entered the AI race already carrying significant leverage from its Cerner buyout:
- Annual CapEx surged from $21.2 billion to over $55 billion, outpacing its operational cash generation;
- Consequently, S&P downgraded Oracle's senior unsecured debt rating to BBB- (the lowest rung of Investment Grade).

If ongoing debt pushes leverage past rating thresholds into BB+ (High-Yield Junk), it will trigger the catastrophic "Fallen Angel" forced liquidation—public pension funds and life insurers, legally barred from holding junk bonds, will be forced to dump Oracle paper. Borrowing costs would spike by 200~300 basis points, directly eating into operating profits.

The $260 Billion "Invisible" Forward Rent¶

Beyond on-balance-sheet debt, Oracle holds over $260 billion in uncommenced data center leases spanning 15 to 19 years. Under US GAAP rules, because these facilities are under construction, they do not appear on the balance sheet as liabilities—they are disclosed quietly in the financial statement footnotes. This represents a massive, non-cancellable rigid rent payment schedule for the next two decades.

Whenever I read these filings, I always chuckle with friends: writers could really learn a thing or two from Silicon Valley accountants—from now on, I should start hiding my most important insights in the footnotes too, haha.

---

3. The Grand Wall Street Playbook: How Big Tech Uses Mortgage Tricks to Hide Debt¶

Oracle isn't alone in needing cash. The entire hyperscaler cohort is bleeding liquidity:
- In Q4 2021, big tech spent $48 billion per quarter** buying back their own stock. By early 2026, buybacks collapsed by 90% to just $4.6 billion. Meta logged three consecutive quarters of zero buybacks because every dollar was redirected into concrete and silicon.
- Operating cash flows are growing at ~23% a year, while data center CapEx is exploding at ~70% a year, creating a $1.5 trillion external financing gap by 2028**.

To avoid ratings downgrades, Wall Street's private credit giants stepped in with the playbook from Residential Mortgage-Backed Securities (RMBS):

1. Meta's 80/20 SPV Model (Project Beignet): Blue Owl owns 80%; Meta owns 20% ($2.37B check), achieving full GAAP de-consolidation. Meta provided a **$28B Residual Value Guarantee (RVG), allowing the SPV to issue 24-year notes with an S&P 'A+' investment-grade rating**, anchored by PIMCO ($18B) and BlackRock ($3B);
2. Microsoft's Hidden Leases & Depreciation Magic: Finance lease liabilities doubled to $62.9B under "Other Liabilities," while data center useful lives were extended from 15 to 25 years, reducing reported annual CapEx by $15B;
3. Alphabet's Credit Derivatives: $43.8B in developer guarantees booked as derivatives with just **$815M (<2%)** in fair-value balance-sheet liabilities;
4. NVIDIA's $500B Central Bank: Jensen Huang partnered with Apollo, Blackstone, BlackRock, and Goldman Sachs, providing 25% OEM Residual Value Support to establish a hard liquidation floor on used GPUs;
5. CoreWeave's CCO Revolution: Migrated from 15% junk debt to an $8.5B Blackstone DDTL 4.0 rated Moody's A3 investment grade, pioneering Collateralized Chip Obligations (CCOs).

---

4. Investment Discipline: How to Professionally Evaluate Oracle¶

Beyond the strategic upside and the hidden debt traps, how should an investor approach Oracle?

Oracle has structurally transformed into a high-beta, high-leverage AI infrastructure hyper-growth compounder.

Having spent more than five years writing code at Oracle, watching that underlying architecture we helped build gradually transform a mature, dividend-paying "blue chip" into an AI growth engine is deeply gratifying. It was a chapter of fighting tough engineering battles and seeing the systems truly scale and succeed before moving on.

Still, regardless of personal sentiment, sound investing requires strict risk discipline. Keep a close watch on these three quantitative red-line sentinels:

Quantitative Sentinel	Why It Matters	Risk Management Response
1. S&P Credit Rating (BBB-)	Preserves investment-grade status; prevents "Fallen Angel" institutional dumping	Maintains Stable Outlook: Thesis remains intact; Placed on Negative Watch or downgraded to BB+: Decisively reduce risk exposure immediately.
2. OCI Growth & Cloud Margin	Proves the $260B in forward leases is converting into high-margin cash flow | **Growth $\ge 40\%$ with expanding margins**: Multi-cloud distribution working as intended; **Growth $<30\%$ or depreciation crushes margins**: Indicates the 3-yr chip decay vs. 19-yr lease mismatch is biting; trim position size.
3. Free Cash Flow (FCF) Inflection	Tests when internal cash generation will re-cover capital outlays	CapEx growth stabilizes & FCF turns positive: Valuation re-rating trigger; Prolonged cash drain forcing dilutive debt: Maintain high defensive vigilance.

---

5. Could There Be an "AI Compute Subprime Crisis"?¶

Financial media and short sellers have recently popularized a sensational phrase: "The AI Subprime Crisis."

From a Wall Street credit and risk perspective: is an AI compute subprime collapse actually on the horizon?

My short answer: it will not look like the 2008 Lehman Brothers liquidity freeze, but it could trigger severe, grinding repricing friction across private credit and institutional pension funds.

Here is the structural reality:
1. The Underlying Obligors Are Fortress Cash Engines: The 2008 crisis was built on non-prime NINJA borrowers without verifiable income. Today, the ultimate payers underwriting the $1.65 trillion compute debt stack are Microsoft, Alphabet, Meta, Amazon, and Apple—the most profitable corporate powerhouses in human history. Immediate systemic default by the tech giants is virtually off the table;
2. The True Vulnerability: Private Credit Gating and the RVG Call:
- Private Credit Liquidity Runs: Private debt funds (such as Blue Owl, Blackstone, and Ares) manage hundreds of billions allocated by state pension systems and retail retirement accounts. These semi-liquid funds enforce quarterly redemption caps (typically 5%). When AI monetization lags and compute rents reprice downwards, surging investor redemption requests trigger gating and forced secondary bond liquidations;
- The Residual Value Guarantee Shock: If Moore's law collateral erosion renders older GPU clusters uneconomical after 3 to 4 years and hyperscalers elect not to renew their leases, tens of billions in Residual Value Guarantees (RVG) will suddenly bounce back onto corporate balance sheets as cash liabilities, triggering sharp earnings compressions and rating downgrades.

---

6. Why Understanding AI Compute Demands Understanding Mortgage Finance¶

This brings me back to the core reason I wrote Mortgage Markets in the Age of AI.

Friends in tech frequently ask me: "Henry, I build AI systems and manage GPU clusters—why on earth should I read a book about US mortgages?"

My answer is simple: because the entire financial engineering apparatus powering the multi-trillion-dollar AI infrastructure buildout—from bankruptcy-remote SPVs, Take-or-Pay leases, and Residual Value Guarantees to cash flow waterfalls, negative convexity hedging, early-termination options, and Monte Carlo OAS valuation—was conceived, stress-tested, and perfected in the US residential mortgage-backed securities (MBS) market over the past fifty years.

The $13 trillion mortgage market is the Rosetta Stone of all structured credit and modern fixed-income financial engineering.

There are many books on AI, and plenty on fixed income. But none connect the two from first principles, stochastic calculus, and production Python code all the way to trillion-dollar off-balance-sheet compute debt. This book is the unfiltered distillation of my years writing infrastructure in Silicon Valley and managing quantitative risk on Wall Street:

📘 Mortgage Markets in the Age of AI / 《AI时代的房贷市场》
(From Cash Flows and Stochastic Calculus to AI/ML Modeling — featuring the comprehensive Appendix F quantitative case study on GPU-Backed Securities)
👉 Explore the Book

---

7. Closing: Sailing Between Conviction and Discipline¶

From writing code at Oracle's Redwood Shores headquarters years ago to analyzing complex structured debt in Manhattan today, my core investment philosophy has crystallized:

The essence of rigorous research and great investing is never blind dogmatism. It is understanding the structural mechanics, the leverage points, and the hidden risks completely—and having the objective clarity to evaluate the technology driving our future.

Understanding Oracle's upside means recognizing the sheer engineering brilliance of OCI's bare-metal compute network in the AI revolution;
and watching its balance sheet and BBB- rating threshold means respecting the unforgiving laws of financial engineering—only those who master defense and respect cycles will endure through the market's greatest waves.