甲骨文解雇了 30,000 人，用于建设其无法完成的数据中心

Key Takeaways

• Up to 30,000 fired, $124.7B in debt: Oracle cut an estimated 18% of its workforce via a 6 AM email on March 31, 2026, to fund a $50 billion annual data center buildout, while carrying more debt than any non-financial US tech company in history.
• Free cash flow: negative $24.7 billion: The company’s trailing twelve-month cash burn is the deepest in Oracle’s 49-year history. The last time Free Cash Flow (FCF) went negative was 1992.
• Data centers are late: Oracle has delayed several OpenAI data centers from 2027 to 2028, blaming material and labor shortages. Across the industry, half of planned 2026 US data center builds are delayed or cancelled.
• The Nortel parallel is uncomfortable: In 2000, Nortel peaked at C$398 billion market cap (~US$267B), had “real” demand for bandwidth, and spent aggressively on debt. By 2009, it was bankrupt. Oracle peaked at roughly $800 billion market cap; it has since fallen to around $500 billion.

The 6 AM Email

On the morning of March 31, 2026, Oracle employees across the US, India, Canada, and Mexico opened their inboxes to find a message informing them that their role had been eliminated and that day was their last. Investment bank TD Cowen estimates the cuts hit between 20,000 and 30,000 workers; Oracle has not confirmed an exact figure.

No town hall. No transition period. A 6 AM notification to as much as 18% of a 162,000-person global workforce.

Oracle’s stated reason was straightforward: the company needs to reallocate capital toward Artificial Intelligence (AI) data center infrastructure. The layoffs are expected to free up between $8 billion and $10 billion in annual cash flow, redirected toward a capital expenditure (capex) plan that has ballooned from $21.2 billion in fiscal year 2025 to a guided $50 billion in fiscal year 2026.

The timing was brutal. The same quarter, Oracle reported $17.2 billion in revenue (up 22% year-over-year) and Oracle Cloud Infrastructure (OCI) grew 84%. The company called it “the best quarter in 15 years.” Then it fired up to 30,000 people.

Oracle was already facing internal dissent before the layoffs. Dozens of employees had signed open letters criticizing the company’s political stances and government partnerships, describing an internal culture where public disagreement with leadership’s positions carried professional risk. The mass layoff compounds an employer brand problem that was already building.

The Debt Mountain

The financial picture behind the layoffs is staggering.

As of Q3 FY2026 (ending February 2026), Oracle carries $124.7 billion in non-current debt, up from approximately $85 billion the prior year. The company’s five-year Credit Default Swap (CDS) spread, a measure of how much it costs to insure against Oracle’s default, sits at roughly 198 basis points: the highest on record.

The trailing twelve-month free cash flow stands at negative $24.7 billion. That is not a typo. Oracle’s FCF was “essentially zero” three quarters ago. Now it is the deepest cash deficit in the company’s 49-year history.

Here is what the numbers look like when you stack Oracle against its peers:

Metric	Oracle	Microsoft	Amazon	Meta
AI Capex Funding	Debt	Operating cash flow	Operating cash flow	Off-balance-sheet (Blue Owl)
Debt-to-Equity	~500%	<100%	<100%	<100%
FCF (TTM)	-$24.7B	Positive	Positive	Positive
CDS Spread (bps)	~198	~30-50	~30-50	~30-50

The difference is existential. Microsoft, Amazon, and Meta fund their AI infrastructure from the cash their existing businesses generate. Oracle does not have that luxury. Its legacy database and enterprise software business generates strong margins but not enough raw cash to cover a $50 billion annual capex plan. The gap is filled with debt.

The capex-to-cash gap is roughly $27 billion per year.

That $27 billion gap is the annual amount Oracle must borrow, sell equity, or otherwise source. In March 2026 alone, the company raised $30 billion through bonds and preferred stock.

The $553 Billion Promise

Oracle’s counterargument is a single number: $553 billion.

That is Oracle’s Remaining Performance Obligations (RPO), or committed future revenue not yet recognized, as reported in Q3 FY2026. It is up 325% year-over-year. It represents multi-year AI training contracts, many of them prepaid or using customer-supplied hardware. Management has stated that most AI contract growth does not require Oracle to raise additional funds.

If you take that $553 billion at face value, the math works. The demand is real. The customers, headlined by OpenAI in a joint venture with SoftBank, are committed. The contracts are signed. Guggenheim analysts project an FCF “waterfall” beginning in fiscal 2029-2030 as capacity comes online and revenue recognition catches up with capex.

But backlog is a promise, not cash. The $553 billion converts to revenue only if Oracle physically builds and delivers the data centers those contracts require. Physical delivery is where the thesis breaks down.

The Infrastructure Wall

Half of planned US data center builds in 2026 have been delayed or cancelled.

The bottleneck is not chips. Nvidia is shipping. The bottleneck is everything the chips plug into: transformers, switchgear, cables, and batteries. Lead times for high-power transformers, the massive custom-engineered machines that connect data centers to the electrical grid, have stretched from 24-30 months (pre-2020) to as long as five years.

This site covered the transformer bottleneck in January 2026 in “The $600B Gamble: Why 2026 is AI’s ‘Valley of Death’,” which warned that a 143-week Generator Step-Up (GSU) transformer backlog would strand hyperscaler capital. Six months later, the prediction is materializing.

Oracle has delayed several data centers being built for OpenAI from 2027 delivery to 2028, citing shortages of skilled labor and materials. The overall scope of the OpenAI project remains unchanged, but timeline slippage on a debt-funded buildout is not neutral. Every month of delay is a month of interest expense without revenue.

Oracle’s partial workaround is a 2.8 gigawatt fuel cell offtake agreement with Bloom Energy, which delivered a fully operational fuel cell system to power an OCI site in 55 days rather than the 90-day contract. Fuel cells bypass the grid transformer queue entirely. But 2.8 GW covers a fraction of the capacity Oracle needs to serve its $553 billion backlog.

The Lawsuit

On February 3, 2026, a securities fraud class action was filed against Oracle in the US District Court for the District of Delaware (Barrows v. Oracle Corporation, Case No. 1:26-cv-00127-JLH).

The complaint alleges that Oracle’s executives misled investors by touting AI infrastructure contracts while failing to disclose that: (1) the massive capex increases would not result in near-term revenue growth, (2) the spending created “serious risks” to Oracle’s debt and credit rating, and (3) management’s representations about the business were “materially false and misleading.”

The class period covers investors who purchased Oracle securities between June 12, 2025, and December 16, 2025. During that window, Oracle shares traded as high as $345.72 per share. They closed at $175.08 on April 17, 2026, a 49% decline in share price from the 52-week high.

The lawsuit was triggered in part by Q2 FY2026 earnings on December 10, 2025, when Oracle reported revenue growth below analyst expectations, capex well above expectations, and negative FCF exceeding $10 billion. The stock fell nearly 11% in a single session.

Blue Owl Capital then pulled out of a financing deal, citing concerns about Oracle’s spending and rising debt. The stock dropped another 5.4%.

The Nortel Echo

In September 2000, Nortel Networks hit a market capitalization of C$398 billion (roughly US$267 billion at the time). The Canadian telecom giant employed 94,500 people worldwide. Demand for internet bandwidth was real and growing explosively. The thesis was correct.

But the telecom industry, including Nortel, WorldCom, and Global Crossing, collectively spent more than $500 billion, mostly financed with debt, building fiber optic networks for demand that arrived two years too slowly. By 2002, telecom stocks had lost more than $2 trillion in market value. By 2009, Nortel filed for Chapter 11 bankruptcy.

	Nortel (2000)	Oracle (2026)
Peak Market Cap	C$398B (~US$267B)	~$800B (Oct 2025)
Revenue Growth	Strong (internet boom)	22% YoY ($17.2B/quarter)
Core Thesis	”Bandwidth demand is infinite"	"AI compute demand is infinite”
Funding Model	Debt-financed infrastructure	Debt-financed infrastructure
Was Demand Real?	Yes	Yes ($553B RPO)
Physical Bottleneck	Fiber overcapacity	Transformer shortage
Workforce Cuts	Massive (from 94.5K)	Up to 30,000 (est. 18% of 162K)
Stock Decline	>90% from peak	~37% from peak market cap
Outcome	Bankruptcy (2009)	?

The analogy is not perfect. Nortel’s demand eventually materialized, but for companies that survived (Cisco, AT&T), not for Nortel. The internet DID need massive bandwidth. It just did not need it from Nortel, at Nortel’s cost structure, on Nortel’s timeline.

The question for Oracle is identical: does the AI infrastructure demand arrive fast enough, on Oracle’s servers specifically, to service $124.7 billion in debt before the interest compounds into insolvency?

The Bull Case You Must Consider

This article would be irresponsible without steelmanning Oracle’s position.

Cloud infrastructure revenue grew 84% year-over-year. Multi-cloud database revenue grew 531%. AI infrastructure gross margins exceed 30%, and multi-cloud database margins sit between 60% and 80%. These are not vanity metrics. The product works, customers are buying, and OCI is winning contracts that Amazon Web Services (AWS) and Microsoft Azure cannot match on price.

The $553 billion backlog is mostly pre-funded. Oracle has stated that most AI contract growth is funded through customer prepayments or customer-supplied hardware. If true, Oracle’s actual incremental capital needs for fulfillment are lower than the headline capex suggests.

Bloom Energy fuel cells bypass the transformer queue. The 2.8 GW offtake agreement offers a path to energize data centers without waiting 3-5 years for grid-connected transformers. Oracle is among the first hyperscalers to deploy this workaround at scale.

Amazon was also FCF-negative for years during its growth phase. The “spend now, profit later” playbook is not inherently fatal.

But the difference between Amazon in 2005 and Oracle in 2026 is one word: debt. Amazon funded its expansion from retail cash flow and modest borrowing. Oracle is funding its expansion from $124.7 billion in debt at a roughly 500% debt-to-equity ratio. Amazon had time because it was not paying interest on a mountain of bonds. Oracle’s clock is set by its creditors.

The Talent Death Spiral

Beyond the balance sheet, Oracle faces a quieter crisis: who wants to work there?

The company fired up to 30,000 people with a 6 AM email. Its leadership has drawn criticism for demanding political conformity from employees, contributing to internal dissent well before the layoffs hit.

In a labor market where AI engineering talent can choose between Google, Meta, Amazon, Microsoft, and dozens of well-funded startups, Oracle’s employer brand is damaged. The mass layoff, the internal culture controversies, and the securities fraud lawsuit create a compound reputational problem that no amount of cloud growth can offset.

Oracle needs to recruit thousands of data center engineers, cloud architects, and AI researchers to execute its buildout. It just told up to 30,000 workers they are disposable.

What Comes Next

The next two years are binary.

Scenario A: The Bridge Works. Oracle delivers enough data center capacity by 2028 to begin converting its backlog into recognized revenue. FCF turns positive in fiscal 2029. The stock re-rates upward. The layoffs are remembered as painful but necessary restructuring. Larry Ellison is hailed as a visionary who bet the company and won.

Scenario B: The Bridge Collapses. Transformer delays persist. OpenAI or SoftBank exercises exit clauses. Enterprise customers hit the “AI ROI cliff” — by 2027, they need to show productivity gains that justify multi-billion-dollar contracts, and some cannot. Oracle’s debt becomes unserviceable. The CDS spread blows out. The stock follows Nortel’s trajectory.

The data does not yet tell you which scenario wins. What it tells you is that Oracle has zero margin for error. Every delayed data center, every missed quarter, every talent loss compounds against a debt mountain that does not forgive slippage.

Oracle fired up to 30,000 people to build data centers it cannot yet finish. Whether that bet pays off depends on whether the physical infrastructure supply chain cooperates with a timeline set by Wall Street, not by physics. In the history of infrastructure buildouts, physics usually wins.