The U.S. economy is running hot and stuck on a hamster wheel as GDP growth must outpace borrowing costs—or else get sucked into a debt spiral

U.S. nominal GDP growth has remained above 6%, exceeding the 5.16% 10-year Treasury yield despite the yield rising more than 100bps since the Iran war began, but this favorable debt-growth arithmetic is at risk. AI hyperscaler capital expenditures are projected to reach $870B this year, up from $470B in 2025, supporting broad economic activity while potentially adding inflation pressure. With the federal deficit at roughly $2T annually and new Treasury borrowing costs near 5% versus expected medium-term nominal growth near 4%, the CRFB warns the U.S. could enter a debt spiral if yields stay elevated or growth slows.

Analysis

The key transmission is not simply higher discount rates; it is a capital-allocation collision between Treasury issuance and hyperscaler debt-funded infrastructure. MSFT, AMZN, GOOG, META and ORCL can absorb higher nominal funding costs near term because of cash generation, but their marginal AI-project IRRs are increasingly sensitive to long-end yields. The first equity casualties of a sustained 5%+ 10-year are likely second-tier data-center developers, leveraged power/compute lessors, and unprofitable AI software—not the cash-rich platforms.

CAT and GE have a more favorable 6-18 month setup than the hyperscalers if AI capex remains intact: their exposure is to physical bottlenecks—generation, turbines, grid equipment, construction and backup power—where supply remains constrained and contracts can include escalation provisions. However, these stocks are vulnerable to a sharper 1-3 month duration selloff because industrial multiples have rerated on the same capex narrative; order backlog, pricing and free-cash-flow conversion must offset multiple compression.

Consensus treats AI spending as a homogeneous tech factor. The more important distinction is whether spend is financed from operating cash flow or external capital, and whether vendors sell scarce physical capacity versus commoditizing compute. A yield rise accompanied by widening investment-grade spreads would be materially more damaging than a Treasury-only move: it would signal crowding-out rather than growth optimism and force hyperscalers to prioritize capex, buybacks and acquisitions.

The near-term catalyst path is Treasury auction quality, inflation data and any downward revision to hyperscaler capex commentary. A weak auction or 10-year yield holding above 5.25% for several weeks should pressure long-duration AI beneficiaries; conversely, stable yields with continued order acceleration at CAT/GE would validate the picks-and-shovels rotation. Falsify the industrial thesis if backlog growth decelerates materially or margins fail to improve despite elevated orders.

Key Decisions for Investors

• Favor a 6-12 month pair: long CAT and GE / short a basket of high-duration, subscale AI infrastructure and software exposures via IGV or ARKQ. Target 10-15% relative upside if physical-capex bottlenecks persist; exit if CAT/GE report order or backlog deceleration, or if the 10-year falls below 4.50% and duration leadership reasserts.
• Maintain core exposure to MSFT and GOOG rather than ORCL for AI capex participation over 3-12 months: stronger internally funded investment capacity lowers refinancing and execution risk. Avoid adding ORCL until debt issuance, lease obligations and incremental cloud margins demonstrate that AI revenue conversion is keeping pace with capex.
• Use a 1-3 month hedge against the crowding-out regime: buy TLT puts or establish a modest TLT short only on a confirmed 10-year yield break above 5.25%. Risk is an inflation downside surprise or flight-to-quality geopolitical event; cover if yields retreat below 4.90%.
• Treat Treasury auction tails and investment-grade spread widening as risk triggers, not headline GDP strength. If IG spreads widen more than 25-30bp while long yields rise, reduce AI-capex beta promptly: that combination threatens funding economics and is more bearish than a growth-driven yield increase alone.