Why Bond Markets Are Pricing Big Tech Very Differently Than Equities

The equity market treats Big Tech as the most durable wealth machine of the decade. The bond market is quietly telling a different story: a 100-year Alphabet bond issued this year is already down more than 10% from where it started, a long-dated Alphabet note is trading near 56 cents on the dollar, and Oracle‘s 2064 bonds have shed more than 20% of their value.

These are not the bonds of distressed companies. They are obligations of some of the most cash-rich enterprises on earth, and their prices are falling anyway.

That divergence is not random noise. It reflects a structural shift in how the largest technology firms finance their artificial intelligence (AI) ambitions, largely through off-balance-sheet arrangements estimated at roughly $1.1 trillion in unrecognised lease obligations, plus another $1.5 trillion in unconditional purchase commitments. The banks funding that build-out are now hedging their exposure using credit default swaps on the very same companies, creating a self-reinforcing loop that some participants compare to dynamics last seen before 2008.

Here is what the data in the bond market is telling you that the equity narrative is not, why the reflexive hedging mechanism matters even if no crisis ever arrives, and what all of it means for a portfolio that looks, on paper, like a simple bet on the best companies in the world.

What $1.1 trillion in hidden obligations actually looks like on a spreadsheet

Start with the aggregate, because the scale is where the story begins.

J.P. Morgan Asset Management estimates disclosed data-centre lease obligations across the major technology firms at around $1.4 trillion, of which roughly $1.1 trillion currently sits off the balance sheet. These figures are only recognised in the accounts once the facilities become operational.

On top of that, the same analysis cites approximately $1.5 trillion in unconditional purchase commitments for chips, computing infrastructure, and power. Economically, much of that behaves like debt, even though it never appears in a standard leverage screen.

The commitments fall into three broad categories:

• Unrecognised leases: data-centre leases signed but not yet commenced, recognised only when the buildings switch on.
• Purchase commitments: take-or-pay contracts for chips, power, and infrastructure that lock in future cash outflows.
• Private-credit joint ventures: off-balance-sheet financing structures with insurers and private lenders, estimated by academic researchers at IESE at roughly $65 billion tied to hyperscalers, including Meta‘s multi-billion-dollar joint venture with Blue Owl and Oracle’s project-finance packages in Texas and Wisconsin.

The accounting mechanism is what keeps most of this invisible. An obligation the company is contractually committed to pay does not hit the recognised balance sheet until the underlying facility is live. The statements most investors read today reflect only a fraction of the total economic exposure already contracted.

The accounting mechanism is what keeps most of this invisible, and AI infrastructure accounting compounds the problem further: hyperscalers assign five-to-six-year depreciation lives to GPU hardware with a realistic two-to-three-year economic life, systematically flattering current-period margins while deferring costs that will eventually land on future income statements.

From aggregate estimates to individual company disclosures

The company-level filings make the gap concrete. Meta disclosed operating and finance leases that had not yet commenced of approximately $278.99 billion as of June 2026, with a further $68 billion of leases entered in July 2026. The same filings reported $349.31 billion in non-cancelable contractual commitments, plus an initial lease commitment of roughly $12.31 billion for a data-centre campus commencing in 2029.

Amazon, by contrast, carried gross lease liabilities of $107.8 billion at year-end 2025, with long-term lease liabilities of $78.3 billion. Microsoft’s disclosures through mid-2026 show substantial not-yet-commenced leases building on a prior $117 billion baseline. On top of all of this, major technology and hyperscaler companies are projected to issue an additional $500-700 billion in new debt over the next 12-24 months.

The gap between what appears on a balance sheet and what is economically owed is not a disclosure technicality. It is the channel through which the financial system is extending credit to Big Tech on terms most investors have never priced into their equity view.

Why the bond market is pricing these companies differently than the equity market is

If the obligations are hidden, the bond prices are not. Start with Alphabet’s 100-year sterling bond, issued in February 2026 and maturing in 2126. It sold just under par. By July 2026, it was indicated at 89.978 pence per pound of face value, a loss of more than 10% in under six months.

That could be dismissed as an oddity of a century-long instrument. Except the pattern repeats. A long-dated US-dollar Alphabet bond, carrying a 2.05% coupon and due August 2050, was quoted at 55.97% of par with a yield to maturity of 6.75%.

A cash-generative, investment-grade issuer with conservative net leverage, and its 2050 paper changes hands at 56 cents on the dollar. A separate Alphabet bond with a 6.125% coupon issued this year slid from par to roughly 88 within about six months.

Then there is Oracle. Its 5.5% notes due September 2064, issued near par in September 2024, have traded at 72-78% of par in 2026, with yields in the 7.3-7.8% range.

Oracle’s 2064 notes represent a loss of more than 20% for buy-and-hold investors, on a bond issued by an investment-grade company roughly 18 months ago.

Oracle’s credit downgrade to BBB-minus by S&P in July 2026, triggered by $55.66 billion in fiscal 2026 capex and negative free cash flow, gives the abstract leverage discussion a concrete single-company case study: a company moving from investment-grade comfort toward the lowest rung of that category within roughly 18 months of issuing debt near par.

One further figure circulates in the original reporting: an Apple bond said to be trading near 50 cents on the dollar. Treat that one with caution. Subsequent research located no independent confirmation, and accessible pricing on various Apple maturities sat near par or in the high 90s across 2024-2026. It is an illustrative claim, not a verified figure.

When investment-grade bonds from the most cash-rich companies on the planet trade at 56 or 75 cents on the dollar, the bond market is embedding a structural rate and duration risk premium that equity multiples have not begun to reflect. Bond prices are not opinions. They are the actual levels at which risk is changing hands, and right now they price a very different future from the one equity screens describe.

The reflexive loop that turns a hedge into an accelerant

Here is where the mechanics get uncomfortable. The banks, insurers, and private-credit funds providing off-balance-sheet financing to hyperscalers hold immense concentrated exposure to a handful of issuers. To protect themselves, they are buying credit default swaps (CDS), contracts that pay out if the borrower defaults, on those same companies’ senior unsecured debt.

Each purchase makes sense for the individual lender. The trouble is what happens when many rational lenders do the same thing at once. The dynamic becomes reflexive, feeding on itself in four steps:

1. Lenders increase CDS hedging, and the added demand pushes CDS spreads wider.
2. Wider spreads inform the pricing of new bond issuance, raising the issuer’s funding costs.
3. Higher issuance costs produce mark-to-market losses on existing bonds and structured deals, which lifts risk-weighted assets at the lending banks.
4. Tighter covenants and risk limits follow, so banks cut new lending, worsening liquidity and reinforcing the perception of risk.

The through-line is that a mechanism designed to reduce risk can, in aggregate, manufacture the very deterioration it was hedging against. That is the channel worth watching, not because it guarantees a crisis, but because it is the pathway through which a shift in sentiment could self-reinforce faster than banks or regulators could respond.

Where the 2006 parallel holds and where it breaks down

The comparison being drawn is explicit. The Bank for International Settlements (BIS) describes hyperscaler lease arrangements as “shadow borrowing.”

The BIS characterises these arrangements as creating opacity, concentration, and interlinkages reminiscent of pre-2008 “shadow banking.”

Commentator Larry McDonald goes further, asserting that off-balance-sheet funding tied to hyperscaler capital expenditure has surged from $500 billion a year ago to $1.8 trillion now, leaving insurers and banks holding opaque exposures. The reflexive CDS loop draws direct comparison to 2006-2007, when banks hedging subprime risk through ABX and CDX indices inadvertently amplified the crisis rather than containing it.

There is a serious counterpoint. J.P. Morgan Asset Management frames this as a market-structure risk, not a weak-link credit risk. Issuer balance sheets are strong, cash reserves are vast, and debt-to-enterprise-value ratios sit comfortably below 5%. The underlying lease contracts are plain-vanilla agreements with investment-grade counterparties, visible in filings, rather than the opaque pools of correlated subprime risk that defined pre-GFC collateralised debt obligations. The debate remains unresolved, and both views are held by credible institutional participants.

What the credit stress signals mean for multi-asset portfolios built around Big Tech equity

This is not hypothetical for anyone holding a diversified fund, because the dynamics already show up in delivered returns. The RPAR risk-parity exchange-traded fund (ETF), built on the traditional 60/40 idea that bonds cushion equities, has produced essentially no return since 2021.

The reason is a shift in correlation. Between 2022 and 2024, the 12-month rolling correlation between stocks and bonds climbed above 0.5, peaking near 0.80 by mid-2024.

A stock-bond correlation near 0.80 by mid-2024 marks the practical death of the 60/40 assumption for that period. Both legs fell together as rates rose.

The equity side offers no easy refuge either. Year-to-date, the MAG 7 ETF (MAGS) has returned roughly 4-5%, underperforming gold miners, copper-related equities, and coal stocks over the same stretch. At the other end of the credit spectrum, effective yields on the ICE BofA CCC and Lower US High Yield Index reached the 12-15% range, with a snapshot of 15.06% on 8 September 2026 and option-adjusted spreads above 800-1,000 basis points.

A portfolio that is long Big Tech equity and long investment-grade bonds is not as diversified as its labels suggest. It is effectively a single macro bet on long duration and AI capex returns, and the credit data suggests that bet carries more unacknowledged risk than most portfolio reviews surface.

How institutional managers are already repositioning

This is not a theoretical adjustment. It is a response to losses that were actually delivered in 2022-2024. Managers have moved in three documented directions:

• Shortening duration by capping bond maturities at around 5 years to limit interest-rate sensitivity.
• Adding explicit inflation hedges to protect against the scenario where bonds and equities fall together again.
• Reassessing volatility-targeting frameworks that assumed a stable negative stock-bond correlation.

The LQD investment-grade bond ETF, which carries significant data-centre financing exposure, illustrates the problem. Its trailing one-year return has been modestly positive, but performance is flat to slightly negative over longer horizons once the 2022 drawdown is included. That is the buffer the 60/40 model assumed would hold.

What would need to change for this to matter, and what to watch

The honest position is that the bearish and bullish cases are both held by serious people. The difference between a contained market-structure quirk and a systemic event hinges on a small number of variables that are observable in real time.

Three are worth monitoring directly:

1. The pace of CDS spread widening on Big Tech senior unsecured debt, the clearest early signal of the reflexive loop gathering momentum.
2. The return-on-invested-capital inflection, the 2029-2033 window when hyperscaler cash consumption is projected to flip to cash generation, cited by J.P. Morgan Asset Management.
3. The rate environment, which determines whether long-duration bonds keep deteriorating or finally stabilise.

Layer on the near-term supply pressure, the $500-700 billion in new Big Tech debt expected over the next 12-24 months, and the picture sharpens.

The $500-700 billion in new Big Tech debt supply expected over the next 12-24 months does not land into a vacuum: Nvidia’s $500 billion AI debt consortium alone represents a multi-year repeat-issuance pipeline that analysts at ING have explicitly identified as a principal near-term risk to financial conditions, adding structural pressure on top of an already stretched Treasury market.

Several market participants characterise the current credit environment as resembling “late 2006” rather than “late 2007,” meaning early-stage rather than acute deterioration.

That framing is what makes the timing useful. If the signals are early, the analytical work still has value, because action remains possible. The reader watching CDS spreads, duration signals, and ROIC timelines is doing something most equity-focused investors are not: reading the credit market as a leading indicator rather than a lagging one.

When equity and credit tell different stories, credit usually finishes first

Pull the threads together and a single argument emerges. Roughly $1.1 trillion in unrecognised lease obligations underpins a financing structure that equity screens do not capture. That structure shows up as bond prices, with Alphabet’s 2050 note at 55.97% of par, Oracle’s 2064 notes down more than 20%, and the LQD ETF flat to negative through the 2022 drawdown. The reflexive CDS mechanism connects those prices to bank lending, and the portfolio data shows the resulting stress already sitting inside standard 60/40 allocations.

Equity markets are pricing these companies on earnings and growth. Credit markets are pricing them on duration, opacity, and hedging dynamics that equity multiples ignore.

J.P. Morgan Asset Management may prove right that this is contained. The point is not to sell. The point is that checking bond prices and CDS spreads on companies you already hold in equity is not advanced analysis. It is basic due diligence, and most retail and semi-professional investors have never done it.

For investors wanting to quantify when the capex-to-cash-flow ratio becomes self-limiting, our deep-dive into hyperscaler cash flow sustainability examines Barclays models showing more than $200 billion in debt issuance required to close the structural funding gap between 2026 and 2028, alongside the Nvidia contingent liability structure that may not surface in balance sheet disclosures until triggered.

This article is for informational purposes only and should not be considered financial advice. Investors should conduct their own research and consult with financial professionals before making investment decisions. Past performance does not guarantee future results. Financial projections are subject to market conditions and various risk factors, and these statements are speculative and subject to change based on market developments.