The AI Funding House of Cards: When Hyperscalers Meet Reality

South Korea's KOSPI index plunged 35% in July 2026. Not gradually. All at once. The circuit breaker fired twice in two days — a historic first — as SK Hynix missed earnings expectations by a whisker and the entire semiconductor narrative collapsed. But this wasn't just about one company or one country. It was the visible part of something much larger. A reckoning.

Here's the setup: SK Hynix reported record operating profit of 60.54 trillion won in Q2, up 557% year-on-year. Sounds great. Except analysts wanted 64 trillion. A 5.6% miss on the headline number. That shouldn't tank a market. Except it did, because everyone suddenly realised the same thing at once. The AI spending spree that's been financing hyperscaler balance sheets can't keep going the way it has been.

The real story lives in the credit markets, where the pain is sharper. Since September, the big six — Microsoft, Meta, Google, Oracle, Amazon, Nvidia — have doubled their collective dollar debt to over $360 billion. In 2026 alone, hyperscalers issued $182 billion in bonds. That's a 1,300% annual jump. Bonds exist for a reason: someone has to pay them back.

Oracle's the canary here. Its five-year credit default swaps widened from 40 basis points at the start of 2025 to 200 basis points by March 2026. Markets are now pricing in a 16% default probability for an investment-grade company. That's not a minor rerating. That's fear.

The structural problem is simple and ugly: these companies don't generate enough free cash flow to fund their AI ambitions in-house. They never have. So they borrow. Combined capex guidance for 2026 is nearly $690 billion — up 81% from 2025 and 207% from 2024. You can't sustain that without external financing or massive equity dilution. Pick your poison.

Meta's $27 billion private-credit deal with Blue Owl for the Hyperion data centre in Louisiana and Oracle's multibillion-dollar project-finance package for AI sites in Texas and Wisconsin are off-balance-sheet leverage structures that equity investors might not fully understand yet. They're not hidden, exactly. Just not obvious.

Seoul also showed us something else: retail leverage in semiconductors. South Korea had only approved single-stock leveraged ETFs months before the crash. Young investors, priced out of housing but bullish on AI, poured $50 billion into these things. They became the marginal buyer at the absolute top. That's a pattern that should keep anyone rational awake at night.

The good news is nobody's saying hyperscalers will default tomorrow. Their balance sheets are still investment-grade. The risk is subtler and worse: forced debt refinancing at higher rates, collapsing capex productivity, or a competitive disappointment that forces a violent valuation reset. Korea's market has already priced that in. The question is whether US equity and credit markets will follow.

Source & further reading:

• Cost to insure AI debt reaches record high amid Asian semiconductor tumble — Cointelegraph
• The traditional 9-to-5 banking day is officially dying, says Morgan Stanley execs — CoinDesk
• Stablecoin firm Brale says new protocol can remove a major hurdle to scaling custom tokens — CoinDesk
• Coinbase hit by spot trading slump: Wall Street trims expectations ahead of earnings — CoinDesk
• Crypto Long & Short: What this year's $972 million crypto hacks actually tell us about security — CoinDesk
• South Korea's stock market drops 35% in July as semiconductor rout triggers circuit breakers — CryptoBriefing
• KOSPI Triggers Historic Back-to-Back Circuit Breakers as AI Memory Rally Faces Structural Reckoning — TechTimes
• SK Hynix Q2 2026 earnings: record profit misses estimates — Quartz
• AI Buildout Fuels Hyperscaler Bond Issuance and Credit Spread Volatility — Sage Advisory
• The Hyperscaler Capex Supercycle: How AI Infrastructure Spending Is Reshaping Small-Cap Opportunity — Penn Capital
• An AI debt wave meets uneven balance-sheet risk — IESE Insight