What it is
Building the data centers, chips, and power infrastructure that AI models run on costs far more than the companies involved can generate in free cash flow. So a growing share of the buildout is being financed with debt rather than earnings. In practice this takes several forms: hyperscalers issuing large corporate bonds to public markets, private-equity and asset-management firms extending private credit to data-center projects, and investors borrowing against the value of their AI stakes to raise cash without selling shares.
The scale has grown quickly. Hyperscalers issued about $108 billion in investment-grade debt in 2025 — roughly 26% of that year’s capital spending — and had already issued $194 billion by mid-2026, on pace for roughly $250 billion (about a third of 2026 capex), according to Goldman Sachs research cited in market commentary. Individual deals have become some of the largest corporate bond sales on record: Oracle sold $18 billion in bonds in September 2025, Meta followed in October with a $30 billion offering — reportedly the largest non-acquisition-related high-grade bond sale ever — and Alphabet and Amazon each raised billions more that November. On August 6, 2026, Alphabet returned to the market seeking up to $25 billion across bonds maturing from 2 to 40 years, after lifting its 2026 capex forecast to as much as $205 billion; the offering drew roughly $115 billion in investor orders, more than four times the amount sought.
A related but distinct mechanism appeared the same week: SoftBank Group, the Japanese investment conglomerate that holds a large equity stake in OpenAI, secured a $10 billion two-year margin loan from a group of banks (Goldman Sachs, JPMorgan, Mizuho, Apollo, and Sumitomo Mitsui) using that OpenAI stake as collateral. This is unusual — lenders are typically wary of large loans backed by shares in private, loss-making startups, since there is no public market price to mark the collateral against, and the loan includes clauses requiring SoftBank to post more cash or repay early if OpenAI’s valuation falls substantially.
Why it matters for AI governance and narratives
This financing pattern is a load-bearing part of the observatory’s opacity and leverage threads, alongside circular vendor financing and off-balance-sheet SPV structures. Where those mechanisms obscure who ultimately bears AI investment risk, debt financing changes how exposed the system is to a demand shortfall. Equity-funded buildouts can absorb disappointing returns by diluting shareholders; debt-funded buildouts carry fixed obligations — interest and principal — that must be paid regardless of whether AI revenue materializes on schedule. Moody’s flagged this directly in July 2026, warning that the scale of AI-related spending threatens the credit quality of Amazon, Meta, Alphabet, and other hyperscalers, and separately noted that circular investment relationships between hyperscalers and AI labs compound the risk by concentrating it among a small number of interdependent counterparties.
For the framing contest the observatory tracks, the volume and terms of this debt function as a real-time credibility signal that competes with company rhetoric. Alphabet’s bond sale drawing four times its target in orders is being read by some coverage as evidence that fixed-income investors still trust the AI growth story; Oracle’s July 2026 downgrade by S&P (from BBB stable to BBB- stable, one notch above speculative grade) is being read as the opposite. SoftBank’s use of a private, hard-to-value stake as loan collateral raises a related question the observatory should track: as more capital is raised against paper valuations of AI companies that have no public market price, how much of the reported financial confidence behind the AI boom is being tested by markets at all, versus assumed.
Key facts and dates
- 2025: Hyperscalers issue roughly $108 billion in investment-grade debt (~26% of capex).
- September 2025: Oracle sells $18 billion in bonds.
- October 2025: Meta sells $30 billion in bonds — reportedly the largest-ever non-M&A high-grade corporate bond deal.
- November 2025: Alphabet raises $17.5 billion and Amazon $15 billion in bond sales.
- July 24, 2026: Moody’s warns that “unprecedented” AI spending threatens the credit quality of Amazon, Meta, Alphabet, and other hyperscalers, citing circular investment exposure.
- Mid-2026: Hyperscaler bond issuance reaches $194 billion year-to-date, on pace for roughly $250 billion (~33% of 2026 capex); Goldman Sachs projects $400 billion (~35% of capex) for 2027.
- July 2026: S&P downgrades Oracle from BBB stable to BBB- stable, citing rising capex, business risk, and weaker cash flow.
- August 6, 2026: Alphabet launches a bond sale seeking up to $25 billion (10 tranches, 2–40 year maturities, Aa2/AA+ rated), drawing about $115 billion in orders; the same day, SoftBank secures a $10 billion two-year margin loan collateralized by its OpenAI stake, arranged by Goldman Sachs, JPMorgan, Mizuho, Apollo, and Sumitomo Mitsui.
What remains genuinely unsettled — and should not be overstated — is whether this debt load represents prudent, well-subscribed financing of durable infrastructure (the bullish reading strong bond demand supports) or an early-stage version of the fixed-obligation stress that turned prior tech buildouts into balance-sheet crises. Coverage on both sides of that question is active and contradictory as of this writing.
Where to learn more
- Bloomberg (via TFN/Investing.com syndication): SoftBank Secures $10 Billion Margin Loan Backed by OpenAI Stake
- Invezz: Alphabet plans up to $25B US bond sale to fund AI spending; attracts $115B in orders
- CNBC: Moody’s says ‘unprecedented’ AI spending threatens credit quality of Amazon, Meta, Alphabet and others
- FactSet Insight: Hyperscalers Tap External Financing as AI Capex Outruns Cash Flow