The midterm sweep rotated money out of AI tech, and the build-out slips a year

The rotation

Tech stocks plunged two days after the election, on fears of legislation reining in AI data center construction. Robeco reads the move as the start of a rotation rather than a retreat: market rotation and return dispersion are increasing, moving away from the AI tech sector rally, and a Democratic sweep in the US midterm elections has created a stock picker's market. The market is demanding evidence that AI capital expenditure will translate into sustainable cash flows, and the stocks that cannot show it are the ones being sold. m:…0229

The rout did not last the week. NAB reports that US tech stocks, including Nvidia, Broadcom, AMD, and Micron, saw significant gains overnight, though the NASDAQ is currently 7% below its all-time highs. CoreWeave shares jumped by as much as 24% following Q3 earnings results. Morgan Stanley Research calls the post-election selloff in AI-related stocks overdone, and argues that market outperformance is transitioning from AI semiconductor and infrastructure enablers toward software and AI adopters. The suggestion is that the money is not leaving AI. It is moving to the layer where the spending turns into revenue. m:…0451 m:…9808

The build-out slips a year

The physical build-out is what actually moved. Morgan Stanley Research shifted 20% of its projected 2028 data center capacity to 2029, and its podcast team puts the same shift at approximately 5 gigawatts of US data center projects, or 20% of the 2028 pipeline. ARK Invest puts the election effect smaller, at approximately 10% of project timelines, but the direction is the same. US data center capacity additions are projected to be flat in 2028 before growing nearly 40% in 2029, with global data center capacity growth forecast at 9% in 2028 and 20% in 2029. Some data center projects scheduled for 2028 could move to 2029, and the build-out gets longer rather than smaller. m:…9808 m:…3590 m:…4277

The spending behind the build-out is still enormous. Morgan Stanley Research forecasts hyperscaler capital spending of $1.6 trillion in 2028 and $1.8 trillion in 2029, and puts the estimated return on invested capital of AI investment at more than 30%. The constraint is not money but power and permission. The US faces a 44-gigawatt shortfall in data center power through 2028, representing 38% of projected demand. Governor-elect Hinojosa in Texas has proposed pausing new data center approvals until stronger protections are in place, and hyperscalers are shifting some development to states with favorable policies, including Tennessee and Louisiana. The midterm results are more likely to change the location, timing, and cost of data center builds rather than prevent them entirely. m:…9808 m:…3590

Financing is where the delay bites first. Morgan Stanley warns that financing for new construction and specialized AI cloud providers could face greater scrutiny, and that tenant quality, power access, and contractual protections are increasingly important for financing. A slower build-out in 2028 could lead to weaker equipment orders, inventory accumulation, and pressure on supplier margins. ARK adds that the current market environment favors incumbents like Meta, Alphabet, Apple, Microsoft, and Amazon over frontier labs, and that Anthropic and OpenAI are in a worse competitive position post-election due to potential delays and increased compute costs. Existing data centers with stable cash flows may benefit from increased scarcity and stronger lease pricing, which is the quiet upside of a delayed build-out for anyone who already owns capacity. m:…9808 m:…3590 m:…4277

The rates backdrop

The rotation is happening against a rates backdrop that has turned hawkish again. Robeco notes that US Treasury yields are hovering over 5%, and that a 5% 10-year Treasury yield has not significantly damaged the stock market. US budget deficits and the $40 trillion national debt are rising, and bond market volatility may control near-to-intermediate term market volatility. NAB reports that markets are repricing for more than 25 basis points of hikes for the December 9th Fed meeting. The federal government retains an accommodative posture toward AI infrastructure due to its importance as national security infrastructure, so the political risk is local rather than federal, but the financing risk is set by the bond market, and the bond market is charging more. m:…0229 m:…0451 m:…3590

For a VC, the practical read is narrow. Capital expenditure is likely to be lower under the Democrats, according to Robeco, but the AI investment cycle is intact and the returns on it, on Morgan Stanley's estimate, clear 30%. The trade this week is not AI versus everything else. It is enablers versus adopters, new construction versus existing capacity, and frontier labs versus incumbents with their own power and their own balance sheets. The election changed who gets to build, where they build, and what it costs to finance. It did not change whether the build-out happens. m:…0229 m:…9808 m:…4277