Will the Blue Wave Wipe Out US Data Centres? What the 2026 Midterms Mean for the AI Build-Out

Last reviewed: November 2026

Data centres lost the 2026 US midterms. Badly. Two days after the 3 November vote, the iShares Semiconductor ETF (SOXX) fell around 4%, Micron dropped about 5% and the neoclouds got belted. The story the market told itself was simple. Democrats won big, Democrats hate data centres, so the AI build-out is in trouble.

We think that's half right. The politics are real and they will slow the next wave of US data centre approvals. But slow is not stopped, and the gap between what politicians promise on election night and what they actually deliver is wide. I've been doing this since 1996, and I've lost count of the election-week sell-offs that looked a bit silly a month later.

So here's how we're reading it. What actually happened, how much capacity is genuinely at risk, why the chips keep shipping for now, and who wins and loses from a slower build-out.

What happened to data centres in the 2026 US midterms?

Short answer: data centres were on the ballot and they lost. Democrats won control of both houses of Congress and flipped six governorships, including three of the biggest data centre states, Texas, Georgia and Ohio. Several of the winners campaigned openly on pausing new projects until power bills, water use and local control are sorted out.

State	Governor-elect	Data centre position (as reported)
Texas	Gina Hinojosa (D), flip	Keep the moratorium until the legislature passes protections for households on power and water costs
Georgia	Keisha Lance Bottoms (D), flip	Statewide pause to assess environmental and utility-bill impacts
Ohio	Amy Acton (D), flip	Conditional moratorium. Projects must cover 100% of their utility costs
Iowa	Rob Sand (D), flip	Opposed a moratorium. His Republican opponent supported one
Florida	Byron Donalds (R), hold	Pro-development, but wants developers rather than residents to pay the costs

Sources: Data Center Dynamics, MultiState, The Guardian, November 2026.

Voters weren't subtle about it either. NBC's exit poll found 76% of voters would oppose a new data centre in their own community, and 56% said data centres were a major factor or bigger in how they voted for Congress. Dozens of local ballot measures restricting new sites passed, mostly in small towns and heavily concentrated in Ohio.

But here's the bit the headlines skipped. Around 35% of the voters who opposed a local data centre were Republicans. And a Republican governor, Greg Abbott, had already halted state environmental permits for Texas data centres in September pending an audit of the ERCOT grid queue. The pause regime was being built before anyone voted. The election hardened it. It didn't invent it.

How much US data centre capacity is actually at risk?

Much less than the scariest number suggests. The figure doing the rounds is from Cleanview, which estimates 41% of planned US capacity, roughly 159 GW, sits in states whose new governors back a pause. Goldman Sachs puts the share of planned expansion that short-term moratoria could touch at around 45%. BMI says 53% of hyperscaler pipeline growth faces slowdown risk.

Those are exposure numbers. Not cancellation numbers. When Jefferies actually sharpened the pencil, it cut its forecast for incremental US capacity through 2030 from 122.4 GW to 112.8 GW. That's a haircut of about 8%. A real cut, but nowhere near a wipeout.

There's another reason to discount the big number. Cleanview's 159 GW implies a total planned pipeline of something like 385 GW. Goldman expects about 64 GW of US data centre capacity to actually be operating by the end of 2026. So the "at risk" pile is a slice of a wish list roughly six times the size of what's switched on today, and a fair chunk of it was never going to be built on schedule anyway. ERCOT's large-load queue in Texas alone held around 470 GW of connection requests. Plenty of that is land-bankers holding options, excuse the cynicism.

Do politicians actually deliver on campaign promises?

Partly. Usually around a third get kept in full, another chunk get watered down, and the big structural ones are the most likely to break. PolitiFact's promise trackers are about the best long-run dataset on this, and they don't support the idea that a campaign pledge becomes law on day one.

President	Promises tracked	Kept	Compromise	Broken
Barack Obama (2009 to 2017)	533	48%	27%	23%
Donald Trump (first term)	102	24%	23%	54%
Joe Biden (2021 to 2025)	99	33%	32%	34%

Source: PolitiFact Obameter, Trump-O-Meter and Biden Promise Tracker. Figures rounded.

And look, the US system is built to slow things down. A few reasons the midterm pledges will land softer than the speeches:

• The President keeps the veto. Any federal law hostile to data centres would need two-thirds of both chambers to override it. That's not happening before 2029.
• Governors need legislatures. In Texas, Democrats took the House but the state Senate stays Republican, which is a handy brake on Hinojosa's agenda.
• Federal regulators are pushing the other way. FERC ordered all six regional grid operators in June to justify or reform how they connect large loads like data centres.
• Executive pauses are easier to reverse than laws. Every one of these can be narrowed with exemptions, cost-sharing deals and community agreements.

The overseas track record backs this up, though it argues for years rather than weeks. Singapore paused new data centres around 2019 and reopened about three years later with selective approvals. Dublin has restricted new grid connections since around 2022 and now wants new sites to bring their own power. Pause, then rules, then slower and more expensive growth. That's the pattern.

Can a moratorium stop data centres already being built?

Mostly no, but it's not a free pass either. Moratoriums target new approvals. They don't unplug energised halls, and they struggle to touch projects that already have their permits, their tenant and their power connection.

The money already committed is enormous. Amazon has kept its 2026 capex guidance at about US$220 billion and Alphabet at US$195 to US$205 billion, with Alphabet flagging a big step up in 2027. Microsoft's latest 10-Q shows US$364.6 billion of leases signed but not yet started, mostly data centres. Meta reports around US$355.7 billion. These are contracts, not press releases.

And the existing market is tight. CBRE's mid-2026 survey of eight primary North American markets had 7,481 MW under construction, 80.4% of it preleased, with vacancy at just 1.4%. When vacancy is that low, every delayed new building makes the powered ones more valuable.

But a couple of honest caveats. Contracted power is not the same as switched-on power, and a later phase of a campus can need a fresh approval even when phase one is running. The town of Wells in Maine just passed a moratorium that reaches back to applications pending from June 2026, so "already in progress" isn't automatic immunity. And Microsoft was already complaining about late third-party capacity in October, before a single vote was cast. Some of the delay was coming anyway.

What does this mean for Nvidia, chip stocks and SOXX?

Near term, not much for reported revenue. Longer term, a slower growth rate rather than a collapse. The chips are still constrained by how fast TSMC can package them and how much high-bandwidth memory exists, not by permits in Ohio.

The latest numbers, all reported before the election:

• Nvidia (NASDAQ: NVDA): data centre revenue of US$89.0 billion in the July quarter, up 117% on a year earlier.
• AMD (NASDAQ: AMD): record data centre sales of US$8.5 billion in Q3, up 97%, reported on 3 November.
• Broadcom (NASDAQ: AVGO): AI semiconductor revenue of US$16.7 billion in its fiscal Q3, guiding to US$21.7 billion for Q4.
• TSMC (NYSE: TSM): high-performance computing hit 68% of Q3 revenue.
• SK hynix (KRX: 000660): long-term agreements already cover more than half of its 2027 and 2028 production capacity.

A delayed site in Texas doesn't destroy a GPU order. It sends the GPU somewhere else, another state, another campus or another country. Meta's president has already said every hyperscaler is asking whether it needs to build outside the US.

Where it does bite is timing. Guidance and order books can wobble this quarter, well before revenue does. Our working view is that the next one or two reporting seasons still mostly reflect the existing pipeline, and any real damage would show up over the second to fourth seasons if near-complete projects miss their power dates. The market already gets this, which is why Nvidia fell around 1% on 5 November and memory stocks fell five times as much.

How should investors think about the data centre slowdown?

We'd treat this as a repricing of who gets paid, not a reason to dump the AI theme. Scarcity favours whoever already owns the power and the buildings. It punishes whoever was relying on the next approval to make the numbers work.

Nvidia and SK hynix remain high-conviction names in our House View, and we like the power infrastructure angle through GE Vernova and Eaton, because a "pay your own way" rulebook means more spending on grid and on-site generation per project, not less.

If you hold the theme, directly or through a semiconductor or Nasdaq-100 ETF, a few things to check:

1. How much of your total portfolio is really riding on US AI capex once you add up ETFs, super and direct shares.
2. Whether you own the memory and neocloud end, which carries the most cyclical and execution risk.
3. Whether you're positioned for a longer, slower build-out with higher power costs rather than a straight line up.
4. What you'll watch next: Nvidia's November result, the ERCOT audit finishing in December and the new governors' first moves in January.

Will the blue wave wipe out data centres? No. It will make the next tranche slower, dearer and more political, and the news cycle will move on long before the policy cycle does. For investors, that's a sorting exercise, and the winners are mostly the ones who already plugged in.

Frequently Asked Questions