- Published on
Red v Blue, Two Years On: The 2026 Midterms
- Authors
In November 2024 I wrote about an election that returned Donald Trump to the White House. This is the two-year mark, and the direction has reversed hard.
On 3 November, Democrats took control of both the House and the Senate. The House was expected to be competitive. The Senate was not, and that is where the result becomes interesting.
Table of contents
- What happened
- The number that explains it
- The map broke in an unusual place
- Why this landed on data centres within three days
- What a midterm does and does not mean
- What I would watch
What happened
Democrats won the House with victories across battleground districts, and took the Senate by flipping seats in Iowa, Kansas, Maine, Ohio, North Carolina, South Carolina and Texas. Susan Collins, who had represented Maine for nearly three decades, lost to Democratic challenger Troy Jackson. In Georgia, Keisha Lance Bottoms won the governorship against Republican Rick Jackson, replacing the term-limited Brian Kemp.
As of writing, the Senate races in Alaska and Mississippi remain too close to call, so the final margin is not settled. The control question is.
President Trump's approval sat at 38% going into the vote - historically low for a president at a midterm, and low enough that the result is less surprising in retrospect than the pre-election coverage implied.
The number that explains it
If you only keep one figure from this election, keep this one: 89% of voters said cost of living was a major or crucial factor in their vote.
That is close to unanimity, in a country where almost nothing commands 89% agreement. It is the kind of number that makes most other explanations redundant. Not immigration, not foreign policy, not any of the cultural axes that absorb most of the commentary. Prices.
There is a lesson here that outlasts this particular election, and the 2024 post arguably under-weighted it. Electorates respond to the level of prices, not the rate of change. An administration can preside over decelerating inflation and still be punished, because voters are comparing today's prices to a remembered baseline from several years ago, not to last quarter. Disinflation is a statistic. Expensive is a lived condition. The gap between those two things is where a lot of political risk hides, and it is systematically mispriced by people who read the CPI release.
The map broke in an unusual place
Senate flips in Iowa, Kansas, Ohio, South Carolina and Texas are not a normal midterm swing. Those are not marginal states that oscillate; several are structurally difficult, and a party winning them in the same cycle suggests something closer to a uniform national shift than a set of local contests.
Collins losing Maine after nearly thirty years is the clearest single data point. Long-tenured senators with personal brands independent of their party are precisely the incumbents who survive national waves. When they do not, the wave was not about them.
I would resist the temptation to read this as realignment. Midterm electorates are smaller, older and more motivated than presidential ones, and a cost-of-living referendum is a poor basis for predicting 2028. What it does establish is that the coalition which won in 2024 was contingent on economic conditions, not structural.
Why this landed on data centres within three days
This is the connection I did not anticipate, and it is the most interesting second-order effect.
On 6 November, three days after the vote, AI stocks sold off hard, with the midterm result cited as part of the trigger. The buildout that I wrote about in June and returned to this week got repriced partly on a political event.
The mechanism is local and physical. Data centres consume enormous amounts of power and water, employ comparatively few people once operational, and raise residential electricity prices in the regions that host them. They have become genuinely unpopular in the places they are built, which was a manageable problem while federal and state policy was uniformly accommodating - permitting, power allocation, tax abatement.
A cost-of-living election changes that calculus directly, because residential electricity bills are part of cost of living. A newly elected official who won on prices has an obvious incentive to scrutinise an industry that visibly raises local energy prices while creating few local jobs. The market appears to have marked that risk within seventy-two hours.
So the chain is: prices decide an election, the election changes who governs permitting and power, and the capital expenditure plans of four of the largest companies in the world get repriced. That is a cleaner transmission from retail politics to capital allocation than I would have predicted, and it suggests the political economy of where AI infrastructure physically sits is a more serious variable than the industry's planning has treated it as.
What a midterm does and does not mean
Worth being disciplined about the limits.
A midterm sweep constrains legislation and strengthens oversight. It does not change the executive, the courts, or the regulatory apparatus already staffed. Divided government in the American system produces less legislating and more investigating, which means the practical effect over the next two years is more likely to show up in hearings, appropriations fights and blocked nominations than in new law.
For anyone modelling policy risk - energy, trade, immigration, AI regulation - the correct update is narrow: the probability of significant new federal legislation in either direction falls, and the probability of friction on implementation rises. That is not nothing, but it is less than the headline suggests.
What I would watch
Alaska and Mississippi, for the final Senate margin, which determines how much friction is available in practice.
Whether cost of living remains the organising issue. If prices stay the dominant frame into 2028, both parties will converge on supply-side responses in energy and housing, which has real implications for the permitting environment the AI buildout depends on.
State-level energy policy. This is where the data centre story will actually be decided, and it will be decided in utility commissions and state legislatures rather than in Washington. That is a less visible venue than Congress, and probably a more consequential one for the capex trajectory.
The 2024 post closed on the observation that American elections now function as volatility events for the rest of the world. Two years on, I would add a correction: they are also volatility events for things that look nothing like politics. A sentence about electricity bills in a swing state turned out to be a sentence about a trillion dollars of data centres.
Sources and further reading:
- CBS News - 2026 House midterm election results as Democrats capture majority
- CNBC - 2026 midterm election results: House, Senate
- Yahoo News - midterms 2026 results and latest updates
- Wikipedia - 2026 United States elections
- CNBC - AI stocks hammered by midterm shock as data centre buildout becomes a "war of attrition"