AI's Real Bottleneck Is Capital, Not Power or Chips, Goldman Warns
CoreWeave shares fell more than 15% on Oct 14, with its $2.75 billion unsecured bond trading below 80 cents on the dollar and five-year credit default swaps at 881 basis points, according to a 36Kr repost of Wall Street CN Max; Oracle, Dell and several data-center names fell sharply the same day. Goldman Sachs estimates hyperscaler capital spending will reach $1.2 trillion next year against roughly $1.1 trillion in operating cash flow, forcing at least 35% of buildout spending — over $400 billion — to be debt-funded, while the 10-year Treasury yield briefly topped 5.3% and AI issuance competes with a near-$2 trillion US fiscal deficit for long-dated money. Broadcom CEO Hock Tan and Intel CEO Lip-Bu Tan said capital and its cost are the biggest constraint, and Goldman warned of a possible semiconductor supply-chain "air pocket" in 2027–2028 if credit tightening delays projects.
If financing costs stay elevated, compute buildouts could slip, which would in turn shape how fast models scale and whether AI supply-chain orders are actually placed.