The economist who coined the phrase ‘Magnificent 7’ says the very thing that made them extraordinary is under threat

BofA strategist Michael Hartnett warns that AI hyperscalers are expected to spend more than $1 trillion annually on AI capex and are collectively negative free cash flow by roughly $200 billion, making the Magnificent 7 increasingly dependent on corporate credit markets. With 30-year Treasury yields above 5.5%—their highest since 2002—and 10-year yields near two-decade highs, higher borrowing costs could constrain AI investment and undermine a key driver of U.S. equity-market wealth effects. Hartnett argues policymakers have strong incentives to prevent a disorderly bond-market move, but identifies voter backlash against AI as an additional risk.

Analysis

The relevant regime shift is from an asset-light, net-cash duration trade to a capital-intensity trade whose valuation now depends on both AI revenue realization and credit-market tolerance. MSFT, AMZN, GOOG and META should not be treated as a single basket: AMZN and GOOG carry the clearest risk that infrastructure spend outruns near-term cloud/advertising monetization, while META's ad-funded cash generation gives it more flexibility to self-finance. AAPL's relative exposure is through buybacks and ecosystem demand rather than direct data-center capex, making it a cleaner defensive leg if long-end yields rise further.

NVDA is the key second-order vulnerability: hyperscaler capex cuts would hit accelerator orders with a lag, likely 2-4 quarters after financing conditions tighten, while its multiple can react immediately to any reduction in customer capex guidance. Consensus is likely underweighting the feedback loop in which higher yields pressure equity multiples, weakening the collateral/value of hyperscaler equity precisely when debt issuance and capex commitments rise. Conversely, a credible Treasury-market stabilization or declining real yields could rapidly re-rate the entire AI complex before any change in reported AI revenues.

Banks are not a clean long-duration-yield hedge. BAC, JPM and GS may benefit from issuance, underwriting and trading activity, but a disorderly long-end selloff would revive unrealized-securities-loss, deposit-cost and credit-quality concerns; GS is the cleaner tactical beneficiary of capital-markets volume, while BAC has greater balance-sheet sensitivity. The article's aggregate cash-flow claims require company-level verification: quarterly capex, incremental debt issuance, lease obligations, and management commentary on committed versus discretionary AI spending are the decisive data points over the next 1-3 months.

Key Decisions for Investors

• Initiate a 1-3 month relative-value trade: long META / short AMZN in equal dollar amounts. META has greater capacity to absorb investment through operating cash flow, while AMZN has the highest combined retail-margin, AWS-monetization and capex-execution sensitivity; reassess if AWS growth reaccelerates materially or META raises 2026 capex without upward EPS revisions.
• Buy downside protection on NVDA through 3-6 month put spreads rather than an outright short. The catalyst is any hyperscaler earnings call that trims capex or extends depreciation assumptions; invalidate the hedge if aggregate MSFT, AMZN, GOOG and META capex guidance rises while cloud/AI revenue guidance is also raised enough to preserve FCF expectations.
• Use AAPL as a relative defensive AI-equity substitute: long AAPL / short an equal-weight basket of AMZN, GOOG and MSFT over the next 1-3 months if the 10-year real yield continues higher. Exit if long-end yields retrace decisively or Apple-specific demand data deteriorate, since the thesis is duration/capex dispersion rather than a fundamental Apple growth call.
• Do not add a broad long BAC/JPM/GS exposure solely on rising yields. Instead, monitor investment-grade tech credit spreads and bank deposit betas; favor GS tactically only if issuance pipelines improve without a parallel widening in financial credit spreads, which would signal disorderly rather than constructive rate normalization.