Microsoft's $25 Billion Bond Ends the Net-Cash Myth: At 65 Basis Points, Its Spread Now Governs AI Capex

On October 29, 2026, Microsoft raised $25 billion for the first time in the US investment-grade market since 2017, in ten tranches running from floating-rate paper due 2028 to a 6.35% coupon due 2066. The 10-year cleared at 65 basis points over Treasuries and the 30-year at about 95 basis points, on a peak order book of $115 billion — roughly 4.6 times the size of the deal. Proceeds are designated for general corporate purposes: they may fund capital expenditure, but they are not ring-fenced for it. That is the whole trick. Microsoft did not sell a data-center bond; it sold a balance-sheet bond and kept the allocation decision inside its own capital plan.

The mix is the tell. A third of the deal — $8 billion of 2046, 2056 and 2066 paper at 6.15% to 6.35% — was sold at coupons that imply about 103 basis points over the 30-year Treasury, which closed at 5.32% on October 29; Bloomberg marked the 30-year spread at 0.95 percentage point. Two-year paper funds a working-capital need. Forty-year paper funds assets that run for decades, and its price locks in a 6%-plus cost of money for a fleet of servers that depreciates. Every incremental data center Microsoft approves from here has to clear that hurdle, and the hurdle is now quoted in public.

The company's own statements say the money was already spent. Cash and short-term investments fell from $102.0 billion at September 30, 2025 to $64.5 billion a year later — a $37.5 billion draw — while property and equipment net of depreciation rose $121.4 billion to $377.0 billion. Against $35.5 billion of funded debt ($4.4 billion current, $31.1 billion long-term), Microsoft still shows $29.0 billion of net cash. That headline is a lease-adjusted illusion: lease obligations of $98.5 billion sit alongside the bonds, and counting them turns $29.0 billion of net cash into $69.4 billion of net debt.

The September quarter shows why the bond was not optional. Operating cash flow was $58.0 billion. Additions to property and equipment took $39.4 billion. Shareholder returns took $10.2 billion. That leaves roughly $8.4 billion of internally generated cash against a $39.4 billion capital program — a hole of about $31 billion in three months. Free cash flow per share fell 27% to $2.50 from $3.44 a year earlier even as revenue per share rose 19%. The $25 billion does not buy new capacity; it replaces cash that has already been spent on servers already sitting on the balance sheet.

The live argument is about which balance sheet gets to finance AI, and at what price. Microsoft is one of only two US public companies still rated AAA by both Moody's and S&P, and it priced 30-year money at about 95 basis points over Treasuries. Alphabet sold 40-year money at roughly 1.33 percentage points over Treasuries — about 38 basis points wider — even though Google Cloud grew 103% to $30.7 billion on $11.1 billion of operating income, a 36% margin, while Azure grew 48% to $33.2 billion. The gap between those two long-end spreads is not a verdict on cloud demand. It is a verdict on the issuer: the market is pricing balance sheets, not returns on data centers.

The economics are lopsided in a way that makes the bond easy to understand and hard to dismiss. Microsoft is running a capital program worth roughly $157 billion a year at current run rates — four times the September quarter's $39.4 billion — on a 20.4% return on invested capital, with capex at 39% of revenue. Its suppliers earn multiples of that: NVIDIA turned $96.2 billion of revenue into $63.7 billion of operating income in the July 2026 quarter, a 66% margin, on 59.5% return on invested capital and capex at 2.4% of revenue; SK hynix reported 104.6 trillion won of revenue and 83.3 trillion won of operating profit in the September quarter, a 79.7% margin, funded from net cash. Microsoft is the only company in this chain borrowing money to buy assets that earn 20% on invested capital, and every dollar of margin the buildout creates is negotiated upstream.

Demand is contracted but concentrated, and the financing is layered. OpenAI has committed to purchase an incremental $250 billion of Azure services, though the April 2026 restructuring of the Microsoft-OpenAI agreement removed Microsoft's right of first refusal over OpenAI's compute. At the far end of the chain, CoreWeave has sold Microsoft about 67% of its 2025 revenue, and its own filings show the top customer at 71% of revenue in the June 2025 quarter before the top three customers fell to 36% in the June 2026 quarter; trailing capex runs at 2.7 times revenue and net debt at 11.5 times EBITDA. The same capital program is financed twice, at 5.55% by the AAA buyer and above 15% by its weakest layer — and the buyer's own capacity decisions set the revenue of the layer beneath it.

The surrounding week is the test. On October 28, Alphabet reported revenue of $127.6 billion, up 25%, Google Cloud revenue of $30.7 billion, up 103%, and diluted earnings of $2.12 a share against roughly $3.03 expected. The stock fell about 13%, erasing roughly $500 billion of market value in the largest one-day decline in the company's history. On October 29, Amazon reported a record 13.9% operating margin, AWS revenue of $46.0 billion, up 39%, and trailing free cash flow of negative $42.4 billion against trailing operating cash flow of $168.3 billion, with 2026 cash capex guided to about $220 billion. Then hours after that equity verdict, Microsoft priced 40-year money at 6.35% on an order book 4.6 times covered.

Here is the mismatch worth trading. The equity market repriced AI capital intensity violently on October 29 — Alphabet lost roughly $500 billion on an earnings miss — while the credit market repriced nothing at all. Microsoft's 40-year coupon went out at 6.35%, and its book was 4.6 times covered. Credit has not yet been asked the question equity is asking: whether AI capex earns its cost of capital. A 4.6x order book is not proof that it does. It is proof that investors believe Microsoft's operating cash flow will still be there in 2066 to pay the coupon. If that belief is wrong, the bond will trade before the equity does.

Two facts settle the argument for now. Microsoft funds the largest capital program in American corporate history at the lowest spread in the cohort — roughly $157 billion of cash capex a year, 39% of revenue, at 65 basis points over Treasuries for ten years — and it can do so because 19% revenue growth and 48% Azure growth have kept operating cash flow arriving at $58.0 billion a quarter. And the same program is consuming everyone else's balance sheet: Alphabet's long-term debt has nearly tripled in nine months, Amazon's trailing free cash flow is negative $42.4 billion, and CoreWeave borrows above 15% for capacity that Microsoft's own decisions help determine.

For investors the conclusion is narrow and testable. The variable that moves fastest in the AI buildout is not the capex number, which is set annually, but the spread, which reprices daily and now carries a published hurdle rate of 5.55% at ten years and 6.35% at forty. Watch the spread, not the spending plan. A software company that could once ignore the bond market now has its capital plan governed by the spread it earns on its own credit, and every hyperscaler, neocloud and memory maker in the chain is priced off the same curve.