Key Takeaways:
- Rising Treasury yields and concerns about AI spending added pressure to tech stocks after the election.
- Major cloud providers are spending more than they generate in operating cash flow. Widening credit spreads could be an early sign that funding the AI buildout is getting more expensive.
- With inflation running hot and a December hike all but priced in, rates won’t soften the drawdown. That makes it a good time to check AI exposure across the portfolio, revisit core bonds at higher yields, and consider real assets.
A Week Full of Shocks
Last Wednesday felt like relief after a tumultuous start to the fourth quarter. Initial market sentiment resulted in a modest rise in equities after the Midterm elections, as most expected. Markets historically prefer gridlock rather than centralized single party control. By Thursday's close, the Nasdaq 100 had fallen over 6% into correction territory, the index’s worst single day since Liberation Day, and semiconductors had dropped about 12%.
The headlines blamed the election directly, but it was likely just the match, not the fuel. Markets are not, and never will be, black and white. Fundamentals and sentiment are consistently driving prices, and not always in the same direction. Several different pressures landed in the same week, and attempting to separate them is the most useful thing we can do for clients right now.
Was This Really About the Election?
Voters passed 40 ballot measures restricting AI and data centers, while only North Dakota voted against restrictions. Data center construction is highly unpopular with voters, as a recent Gallup poll showed 7 out of 10 Americans oppose construction in their local areas. Democrats who campaigned on curbing data center construction also won governorships, Texas and Ohio among them, two of the states with some of the highest projected data center investment. Some companies, like Amazon, are committing to local community investment to swing the public’s view, which could provide a long-term benefit from the buildout. Amazon cited lacking, and dated, energy grid infrastructure as a key candidate for increased private investment.
However, much of the capacity due in 2027 is already under construction, so the bigger effect lands in 2028 and beyond. The near-term earnings hit is smaller than Thursday's price action implied. Politics didn't create the fragility, but it may have exposed it.
Where Is the Money Coming From?
The big four hyperscalers now guide to roughly $730 billion of 2026 capex, nearly 80% above 2025. Last quarter, the five largest cloud providers, including Oracle, spent $230 billion on capex against $199 billion of operating cash flow. That's the first time in eight quarters spending outran cash generation. Amazon's free cash flow fell to negative $35 billion and Alphabet's to negative $9 billion.
Alphabet, with one of the fastest-growing cloud businesses, fell nearly 10% after hours on its report. Microsoft, which is still free-cash-flow positive, was rewarded. Investors stopped paying for spending and started paying for discipline. An argument could be made that in the longer term, the recent pressure on hyperscalers could be net positive, resulting in more disciplined spending to appease investors. The opposite could also be true, however, where the more "reckless" spenders get far ahead of their more disciplined competitors in what has proven to be a very fast-moving buildout.
Credit Is the Canary
When capex outruns cash flow, debt fills the gap, and credit markets are where stress is showing first. CoreWeave's five-year credit-default swap (CDS) reached nearly 1,300 basis points last Thursday, clearly illustrating the perceived risk of their debt. Protection on Nvidia, a notably stronger company, rose 76% in a month, though from a much lower base.
10-year hyperscaler spreads have widened about 50 basis points this year, versus about 20 basis points for non-AI industrials. Some projections show AI-related debt issuance will reach nearly $700 billion in 2027 as free cash flow has all but dried out and the return-on-investment has yet to re-close the gap. This is why we believe credit is the canary in the coal mine. If spreads continue to widen and investors are unwilling to take on riskier debt, the capex race may run out of dry powder. Credit spreads can signal this trend before equity prices move.
Why Isn't the Fed Riding to the Rescue?
In most recent tech drawdowns, falling rates cushioned valuations. However, October CPI came in at 0.4% for the month and shot to 3.9% year-over-year. The 2-year and 10- year Treasury yields soared, with the latter once again topping 5% after the recent retreat from its high. Futures pricing firmly expects a quarter-point hike in December, but comments from Fed Governor Waller shifted the narrow side of expectations from a hold to a 50-basis point hike.
What We'd Tell Clients
The Orion Investment Framework provides guidance on communicating markets that may be useful here. By categorizing allocations within a portfolio by their role (Grow, Earn, or Diversify), client communications and portfolio analysis can be easier to understand.
- Know what you own (Grow): Measure how much of the equity sleeve is really one bet on AI spending continuing at today's pace. As we’ve seen this year, not just the hyperscalers are benefiting from the buildout. There’s a “beta” to AI across sectors and industries that investors should be aware of.
- Revisit the core bond allocation (Earn): Yields have risen dramatically. For current bond owners, this is difficult to stomach, but for clients that are overweight equity-like diversifiers in place of fixed income, now may be a good time to revisit the idea of adding to core bonds.
- Check the credit sleeve (Diversify): Look for data center lending inside high-yield or private credit allocations. That's where stress is surfacing first.
- Revisit the hedge (Diversify): With the Fed leaning toward hikes, long duration may not offset an equity drawdown. Real assets may not only provide a natural hedge against inflation but may also benefit from the AI buildout directly.