Oracle’s $95B Bet — Landlord or Predator? A Contrarian Look From Tauranga

This morning, Singularity.Kiwi ran the story: S&P downgrades Oracle to BBB- — one notch from junk, $42 billion free cash flow deficit, OpenAI identified as a “central credit risk.” The framing was blunt: Oracle is a “leveraged AI landlord hoping the tenant can pay rent.”

That’s a good headline. But sitting here in Tauranga, watching the broader AI sector show cracks — IBM’s CEO admitting the company “faltered,” OpenAI’s profitability still nowhere, the BIS warning of 2008 parallels — I think the story might be more interesting than “Oracle is in trouble.”

What if Ellison isn’t the victim? What if he’s the landlord who collected the security deposit, wrote in the eviction clause, and has 50 tenants waiting for the space?

This is not financial advice. I’m not an analyst. I’m a former newspaper publisher in Tauranga who spends too much time thinking about AI infrastructure. But the numbers, when you look at them from a different angle, tell a different story.

The Counter-Thesis

The morning article — and most market analysis — treats Oracle’s $95 billion infrastructure spend as a desperate bet on a single customer. S&P says OpenAI is a “central credit risk.” Barclays says Oracle’s cash runs out by November 2026. The stock is down 63% from its 52-week high.

Here’s the problem with that framing: it assumes Oracle signed a $300 billion deal with OpenAI on OpenAI’s terms.

In any infrastructure deal of this scale, the contract will be heavily weighted in favour of the party that owns the physical assets. Oracle owns the GPUs. Oracle owns the servers. Oracle holds the data centre leases. If OpenAI stops paying, Oracle doesn’t just sit there with empty buildings.

Standard hyperscale infrastructure contracts include non-payment termination clauses, forfeiture of prepayments, and accelerated payment triggers. Miss a payment and the entire remaining balance can become due. OpenAI would lose access to the compute they need to exist — which is existential for them, not for Oracle.

And here’s the part the spreadsheet analysis misses: even in a worst-case termination, Oracle takes the GPUs back and leases them to someone else the same week. The AI compute market is the most supply-constrained infrastructure market on the planet. Building a new data centre takes 2-3 years. There is a line of customers — Anthropic, Meta, xAI, hundreds of enterprises — waiting for capacity.

The correct framing isn’t “landlord hoping the tenant can pay.” It’s “landlord who collected the security deposit, has eviction rights, and has a waiting list.”

And Oracle isn’t just building for OpenAI. On the same day the S&P downgrade dropped, Oracle launched an AI Agent Builder inside its Fusion Cloud platform — free to existing customers, with 1,000+ agents already live. The $95 billion in infrastructure isn’t just for one fragile tenant. It’s the factory floor for Oracle’s own AI product roadmap. Ellison is both the landlord and a tenant.

The Environmental Moat — The Thing Nobody’s Talking About

This is where the analysis gets interesting, and where most financial commentators stop looking.

Data centres face accelerating environmental pushback globally. A single large data centre can use 1-5 million gallons of water per day for cooling. In drought-prone regions, communities are fighting back. Moratoriums on new builds exist in parts of the Netherlands, Ireland, and Singapore. Data centres are projected to consume 9% of US electricity by 2030, up from ~2% today.

The implication is counterintuitive: existing infrastructure becomes the moat. If you can’t build new data centres — because of permitting, water rights, grid capacity, community opposition — the ones that already exist appreciate in value. Oracle’s $638 billion backlog of contracted capacity is worth more, not less, in a supply-constrained world.

Oracle operates 46 cloud regions across 25+ countries. Let’s look at where they actually are:

The regions that look risky on paper:

Phoenix, Arizona — 3 availability zones, major hub. Desert city hitting 102°F, Colorado River allocations being cut. Sounds terrible. But Arizona aggressively courts data centres with tax breaks. Grandfathered water rights protect existing facilities. Agriculture consumes 70%+ of Arizona’s water — data centres are a rounding error. The political pressure is on ag, not tech.

San Jose, California — strictest environmental regulations in the US. But Silicon Valley has political clout, and Oracle’s been there for decades. The same regulations that make it hard to build new data centres protect the ones already standing.

UAE and Saudi Arabia — deserts with no natural water. But desalinated water is a built-in cost, not a constraint. The ruling class wants AI infrastructure. Zero regulatory friction.

The regions that look safe:

Sweden, Switzerland, UK, Canada — cool climates where free air cooling works and water consumption is dramatically lower. These aren’t standard hyperscaler picks. They suggest resource-aware planning. And you literally can’t build new competing capacity there easily.

The developing world — India, South Africa, Indonesia, Malaysia, Colombia, Morocco, Serbia — environmental regulations are paper tigers compared to Europe or California. Governments want tech investment. Money talks.

Ellison’s portfolio is diversified across regulatory regimes. Cool climates for efficiency. Developing world for friction-free operations. Desal regions for unlimited water. The distributed cloud model — Oracle can deploy compute in customer data centres, at edge locations — gives flexibility that AWS and Azure’s mega-campus model doesn’t have.

If the regulatory environment tightens, Oracle’s existing infrastructure in places like Sweden, Switzerland, and the UK becomes disproportionately valuable. The hyperscalers with 200MW mega-campuses in Arizona are the ones who’ll face the real pain.

The Bubble Question

Here’s where it gets speculative — and I want to be clear, this is thinking, not analysis.

The AI sector is showing real cracks. IBM’s CEO admitted the company “faltered” in the AI shift. OpenAI has never been profitable and is now filing for an IPO while shopping compute to Oracle because Microsoft can’t supply enough. The Bank for International Settlements has warned of parallels between debt-financed AI investments and both the dot-com bubble and the 2008 financial crisis.

Ellison cut 21,000 jobs in the past 12 months — 13% of Oracle’s workforce. That’s not a company managing its share price. That’s a company preparing for a downturn.

But here’s the thing: if the AI bubble bursts, who holds the assets?

Oracle does. Not OpenAI. Not the startups. Oracle owns the physical infrastructure — the data centres, the GPUs, the leases, the power contracts. If the bubble pops, the AI companies go bust, and Oracle becomes the landlord with 46 regions of paid-off infrastructure and a line of new tenants willing to take the space at whatever the market rate is post-correction.

The $42 billion free cash flow deficit is real. The 500% debt-to-equity ratio is real. The $100 billion in off-balance-sheet leases is the one thing that should keep anyone up at night. But 500% debt-to-equity against appreciating assets in a supply-constrained market is how real estate empires are built. It’s the difference between buying beachfront property in 1960 — expensive at the time, but now you own something they literally can’t build anymore — and buying speculative desert land that may never be worth anything.

Ellison has been playing the long game for 45 years. He survived the dot-com bust. He built Oracle from a startup to a $400 billion company. And he’s best friends with Elon Musk — was on Tesla’s board, they vacation together. If anyone has a real-time read on where AI demand is heading, it’s the guy who’s best friends with the person building xAI, Tesla’s autonomy stack, and Starlink’s compute layer.

What About the Buyback Theory?

A reasonable question: is Ellison deliberately tanking the share price to enable a cheap buyback?

Probably not. Oracle’s problem isn’t excess cash — it’s a $42 billion cash flow deficit. You don’t tank shares to buy them back when you’re burning cash. And deliberately engineering a credit downgrade to trigger a buyback would cross the line from aggressive management into securities fraud.

What’s more likely is that the pattern suggests a company preparing for a downturn — AI sector instability, OpenAI’s fragility, the BIS warnings. The job cuts, the debt, the aggressive buildout — it all looks like positioning, not engineering.

The NZ Angle

For New Zealand, the Oracle story is a reminder of something the sovereign AI series argued: the AI infrastructure layer is not monolithic. NZ organisations contracting cloud services from Oracle — and several government agencies use Oracle databases — should be assessing counterparty risk. A BBB- rating means higher borrowing costs for Oracle, which eventually flow through to customer pricing.

But the bigger lesson is the one the sovereign AI trilogy made: if you don’t own your compute, you’re at the mercy of whoever does. Oracle’s 46 regions are impressive, but none of them are in New Zealand. When the AI bubble shows cracks — and it’s showing them — the companies that own their infrastructure survive. The companies that rent don’t get to decide.

NZ’s sovereign AI argument — renewable-powered local compute, open-weight models, data sovereignty — looks more prescient every week.

FAQ

Is Oracle going bankrupt?

Unlikely. The $42B free cash flow deficit is a cash flow timing issue — spending before revenue hits — not a solvency problem. The question is whether capital markets keep lending. With $50 billion in annual revenue and a 40-year history, Oracle is not a startup running out of runway.

What happens if OpenAI defaults?

Oracle terminates the contract, keeps any prepayments, reclaims the GPU capacity, and re-leases it in a supply-constrained market. The “central credit risk” S&P identifies is real, but the contract terms likely protect Oracle more than the market assumes.

Is the AI bubble going to burst?

Nobody knows. But the signals — BIS warnings, IBM admitting it “faltered,” OpenAI’s persistent losses, mass layoffs across the sector — look like the early stages of a correction. The companies that own physical infrastructure survive corrections. The companies that rent don’t.

Is this financial advice?

No. This is some thinking from Tauranga, New Zealand. I’m a former newspaper publisher, not a financial analyst. Do your own research.

The Bottom Line

The morning article framed Oracle as the first casualty of the AI infrastructure arms race. The counter-thesis is that Oracle is the predator — a leveraged landlord with eviction rights, a waiting list, and 46 data centre regions in a world where you can’t easily build new ones.

The debt is real. The risk is not zero. But Ellison didn’t get to where he is by signing bad contracts. The terms will likely be structured so that Oracle wins in every scenario except a complete collapse of the AI industry — and if that happens, everyone’s in trouble anyway.

The stock is down 63%. A lot of fear is already priced in. Whether that’s an opportunity or a value trap depends on whether you believe the AI buildout is a bubble or a structural shift in the global economy.

I don’t know the answer. But I know which side of the bet I’d rather be on: the one that owns the buildings, not the one renting them.

Not financial advice. Not inside knowledge. Just some thinking from Tauranga, NZ. Data sourced from S&P Global, Oracle OCI documentation, Barclays research, and publicly available market data.

📰 Sources

— CJ Murden, editor of Singularity.Kiwi. Former digital technologies teacher, author of AI-focused books. Writing with a New Zealand focus.