Oracle's $20 Billion Stock and Bond Raise and the $3.2 Trillion AI Obligations Question

On November 2, 2026, Oracle filed with the US Securities and Exchange Commission to sell $10 billion of common stock and $10 billion of senior notes, a $20 billion package to help fund its AI data-center buildout. Oracle's shares fell 8.8% that Monday, according to CNBC. Within the same 48 hours, a Nikkei Asia report, as relayed by secondary outlets, put the off-balance-sheet AI obligations of five US technology giants at roughly $3.2 trillion. The two stories are different in size but connected: both are about how the cost of AI infrastructure is being financed, and where that financing is visible to investors.

This post separates three things that are easy to blur together: the new Oracle raise itself, the different ways Oracle's debt can be measured, and the broader gap between debt on the balance sheet and contractual commitments disclosed elsewhere. It does not take a view on whether the buildout will pay back.

What Oracle filed, and what it said beforehand

The filing was a preliminary prospectus supplement (Form 424B5) dated November 2, 2026. It covers $10 billion of common stock and, concurrently, a $10 billion public offering of senior notes. Neither offering is conditioned on the other, and the common stock ranks junior to the notes. BofA Securities is the global coordinator and joint book-running manager. Net proceeds, the underwriting discount and the note terms were not yet filled in on the version we read.

The filing was not a surprise to the market. On October 28, at its investor day, Oracle said it was committed to limiting additional capital raises through fiscal 2030 to $20 billion, split between $10 billion of equity and $10 billion of gross debt, excluding refinancing of existing debt. CNBC reported that the shares initially rose after that event, and Baird analyst Rob Oliver said the filing "seems to be exactly what they previewed." UBS analyst Karl Keirstead called the plan "a major relief given concerns about the state of debt markets and the cost of capital," per CNBC.

Oracle's investor-day release also said a separate $20 billion at-the-market equity raise had already been completed in the first quarter of fiscal 2027, in addition to the $20 billion cap. That sequencing matters for the dilution maths below.

The use of proceeds is deliberately broad

The prospectus says proceeds are for general corporate purposes, which may include capital expenditures, debt repayment, investments or acquisitions, and dividends or buybacks. Oracle's investor-day release adds that vendor financing, customer-provided hardware and customer prepayments are expected to reduce upfront cash requirements until free cash flow turns positive, which Oracle expects in the second half of fiscal 2028. It targets free cash flow of $30 billion or more in fiscal 2029, and net debt below 3.5 times adjusted EBITDA that year. Capital expenditure is expected to peak in fiscal 2027 and 2028.

For scale, Oracle's earlier fiscal 2027 first-quarter results showed capital expenditure of $28.5 billion in one quarter and negative free cash flow of roughly $5.4 billion, as covered in our Oracle Q1 FY2027 earnings analysis. Measured against that run rate, $20 billion is not a full funding of the buildout. It is a bridge alongside other sources, which is how Oracle itself frames it.

Dilution: the sourced arithmetic

The prospectus gives enough to calculate the stock side directly. It lists 3,023,726,695 common shares outstanding as of August 31, 2026, and a last reported price of $130.67 on October 30. At that price, $10 billion buys about 76.5 million shares, which the prospectus uses to show 3,100,255,354 shares outstanding after the offering.

• Dilution at the reference price (our calculation): 76,528,659 new shares divided by 3,023,726,695 existing shares is about 2.53%.
• Implied market value at the reference price (our calculation): 3.0237 billion shares times $130.67 is about $395 billion, so $10 billion is roughly 2.5% of that.
• If the stock prices lower: the number of shares scales inversely. At a hypothetical $119.17, which is the $130.67 reference price reduced by CNBC's 8.8% fall (our assumption that the fall applied to that base), the offering would need about 83.9 million shares, or roughly 2.8% of the August share count. The actual pricing will differ.

Other potential share-count additions are disclosed in the same filing: 19,474,564 options at a weighted average exercise price of $224.13 (well above the $130.67 reference price), 76,789,897 restricted stock units, and up to 31,238,285 shares from the 6.50% Series D Mandatory Convertible Preferred Stock at its maximum conversion rate, with mandatory conversion expected on January 15, 2029. The earlier $20 billion at-the-market raise is a separate transaction, and we did not find its share count in the sources, so it is not added here.

The takeaway is that, on the filing's own numbers, the new stock is a low-single-digit percentage of the share count. Whether the market reaction reflects dilution, the signal about funding needs, or the broader debt picture is a question the share-price move alone cannot answer.

Which debt number? $125.34 billion versus $236.1 billion

Two figures for Oracle's debt are in circulation, and they are not the same measure. CNBC cited $125.34 billion in total debt as of August 31, mostly senior notes, which matches the notes payable and other borrowings line. The new prospectus supplement states that, excluding intercompany indebtedness, Oracle had approximately $236.1 billion in aggregate principal amount of consolidated indebtedness outstanding at the same date. The difference is $110.76 billion (our calculation).

The prospectus text we read does not say what makes up the gap, and we have not seen a reconciliation, so we do not attribute it to any specific item. What the comparison shows is practical: any ratio built on Oracle's debt depends heavily on which definition is used. The prospectus also says Oracle could borrow up to a further $19.7 billion under its revolving credit facility and commercial paper program, and that it expects additional borrowings or liabilities from vendor financing, leases and other arrangements.

On either base, the $10 billion notes are modest in proportion: about 8.0% of $125.34 billion and about 4.2% of $236.1 billion (our calculations). The full $20 billion is about 8.5% of the $236.1 billion figure.

The cost of carrying the notes

Pricing of the new notes was not available in the filing we read. CNBC reported that Oracle's 4.95% bond due 2031 yielded 7.276%, and that S&P downgraded Oracle to BBB-, one notch above high yield, in July. As an illustration only, $10 billion at a 7.276% yield would be about $728 million a year in interest (our calculation). The actual coupon, maturities and spread will be set at pricing.

CNBC also reported Oracle's five-year credit default swap spread at 292, up 265% in a year, and quoted S&P Global Market Intelligence's Gavan Nolan saying that level is more in line with a high-end junk rating. A CDS spread is the annual cost of insuring against default, so a rising number means the market is charging more for that protection. It is a market signal rather than a rating, and CDS markets can be thin, but it is the same kind of indicator we examined in our post on CoreWeave's debt and credit warning.

The Nikkei report: commitments versus debt

The second story concerns a different category. Nikkei Asia reported that off-balance-sheet AI obligations at Alphabet, Microsoft, Amazon, Meta and Oracle have almost doubled in six months to about $3.2 trillion. Nikkei's earlier estimate for the same five companies, published in July, was about $1.65 trillion. Dividing the two gives an increase of about 94% (our calculation), in line with the "roughly 90%" in the report's headline. We could not retrieve Nikkei's original article, which is paywalled, so these figures come from outlets relaying it, and the underlying methodology is not published in the pieces we read.

The July study, as reported by The Next Web, compared the $1.65 trillion with about $1.35 trillion of debt the five companies report on their balance sheets. It described the off-balance-sheet category as costs for chips, servers and power packaged into separate legal entities, often joint ventures, and the article compared the structures to those Enron used. Relays differ on how the category is defined: some describe lease and purchase commitments disclosed in SEC filing footnotes, others frame it as structures kept off the company's own accounts. Those are not identical things, and the sources do not settle which the $3.2 trillion measures.

Two cautions apply. First, the brief version of this story in some circulation includes a figure for obligations as a share of total assets, but we found no source for it and have left it out. Second, a competing November tally reported by Kalkine put the total at about $3.3 trillion with a very different split (for example, $866 billion for Meta against Nikkei's roughly $420 billion), which points to different scoping rather than a settled number. The Nikkei relay also states that Meta's $420 billion is "nearly triple" its reported long-term debt of about $58.7 billion, but $420 billion divided by $58.7 billion is about 7.2 times (our calculation), so that wording and the figures do not agree.

Google

AWS

Microsoft

Meta

Why commitments and debt are different instruments

A commitment is not a loan. A lease that has been signed but has not yet commenced, or a multi-year purchase contract for chips, is disclosed in the footnotes of a filing, and it becomes a recorded liability only when accounting rules say so. Some commitments can be reduced or exited, others cannot, and many are backed by customer contracts that produce revenue. This is why our earlier piece on the Wall Street Journal's $3 trillion count of off-balance-sheet commitments and our look at Meta's $825 billion of future commitments stress that the headline totals mix categories of very different firmness.

Oracle sits at the intersection of the two stories. It is simultaneously raising balance-sheet capital through stock and notes, and, according to the relays of Nikkei's July work, carrying large future lease commitments that are not yet on its books. Those relays give figures of $260 billion in one outlet and $273.3 billion in another, with differing as-of dates, so we do not rely on a single number. The direction is clearer than the amount: part of Oracle's capacity build is financed through channels (leases, vendor financing, customer prepayments, customer-provided hardware) that never pass through a bond prospectus.

Scenarios and what to watch

• Pricing and demand for the notes. The coupon, tenor mix and order book will show what investors charge for Oracle credit relative to the 7.276% yield on the 2031 bond. A tight book would suggest the market separates the funding plan from earlier concerns.
• Stock pricing versus the reference price. The share count issued depends on the pricing level. A lower price means more shares for the same $10 billion.
• Whether $20 billion stays the cap. Oracle committed to the limit through fiscal 2030, excluding refinancing. Any deviation, or a rise in vendor financing and leases, would shift the picture even if the cap holds.
• Cash conversion. Oracle expects positive free cash flow in the second half of fiscal 2028. Quarterly capex and free cash flow figures are the checkpoints.
• Disclosure convergence. If Nikkei, Kalkine and the Wall Street Journal totals narrow as filings update, the commitments question becomes easier to measure. If they diverge further, the range itself is the finding.
• Rating path. A BBB- rating sits one notch above high yield, so rating-agency commentary after the raise will matter for borrowing costs.

What it means for teams building on AI

For organizations choosing AI infrastructure, the financing structure of a cloud provider is a supplier-risk input alongside price and performance. Capacity that is contracted today depends on the provider completing a long, capital-intensive buildout. Keeping workloads portable across providers, which is the approach platforms like Metir take with models, is one way to keep that dependency flexible, though it does not remove it.

Sources:

• CNBC, "Oracle files to sell $20 billion in new stock, bonds to finance AI buildout" (Nov 2, 2026): https://www.cnbc.com/2026/11/02/oracle-files-to-sell-20-billion-in-new-stock-bonds-to-finance-ai-buildout.html
• Oracle Corporation, preliminary prospectus supplement (Form 424B5), Nov 2, 2026: https://www.sec.gov/Archives/edgar/data/0001341439/000119312526442729/d181425d424b5.htm
• Oracle Investor Relations, "Oracle Investor Day 2026: Accelerating Value Creation": https://investor.oracle.com/investor-news/news-details/2026/Oracle-Investor-Day-2026-Accelerating-Value-Creation/default.aspx
• Nikkei Asia, "US tech giants' hidden AI debt swells 90% to $3.2tn in 6 months" (original paywalled; not retrieved): https://asia.nikkei.com/business/technology/artificial-intelligence/us-tech-giants-hidden-ai-debt-swells-90-to-3.2tn-in-6-months
• AI Weekly, "Nikkei: Big Tech Off-Balance-Sheet AI Debt Doubles to $3.2T" (Nov 3, 2026): https://aiweekly.co/alerts/nikkei-big-tech-off-balance-sheet-ai-debt-doubles-to-32t
• DEV Community, "$3.2T hidden AI debt grew 90% in 6 months" (comparison with Kalkine tally): https://dev.to/frankchu/32t-hidden-ai-debt-grew-90-in-6-months-gj6
• The Next Web, "Five tech giants are hiding $1.65tn in AI debt" (July 2026): https://thenextweb.com/news/tech-giants-hidden-off-balance-sheet-debt-ai

Image credits

• Hero: Larry Ellison on stage, Oracle PR, CC BY 2.0, via Wikimedia Commons: https://commons.wikimedia.org/wiki/File:Larry_Ellison_on_stage_(2).jpg
• In-body: Larry Ellison, Ilan Costica, CC BY-SA 4.0, via Wikimedia Commons: https://commons.wikimedia.org/wiki/File:Larry_Ellison_CEO_of_Oracle_Corporation.JPG