Midterms and AI expenditure
A few people have asked me whether the midterms put the brakes on the AI buildout. I went back through the data, and my view is that 'Bluenami' :-) will change the pace and the cost of the AI buildout. They will not change its direction.
Before the elections, Amazon, Alphabet, Meta, Microsoft and SpaceX were on track for somewhere between $803 and $828 billion of capex this calendar year. Analysts expect the group to top $1 trillion next year. I expect this to slow down and go over multiple years but not stall. Capex roughly doubled this year from about $410 billion. Going from $820 billion to a bit over $1 trillion is around 22% growth. Third-quarter growth of 3.4% leant on this. Equipment spending rose 20.9% and added 1.16 points, the biggest contribution since 2011 outside the 2020 rebound. If that growth fades to a fifth, I estimate the contribution falls by 0.5 to 1 point by mid-2027. That figure is gross. Imported chips come back through the import line, so the net hit is smaller.
I did expect the job numbers to be a bigger part of this story, and after having looked through the numbers, I don't think they are. September payrolls were up 29,000, against a 12-month average of 45,000. Unemployment has been stuck between 4.1% and 4.3% since March. The labour market was soft before the elections. Data centres were not expected to create a substantial employment impact. The one visible construction signal in the report was speciality trade contractors, up 12,000. A capex slowdown will show up in GDP more pronouncedly than in payrolls. The only labour risk I am watching for is household purchasing power, with electricity and fuel bills eating into spending.
Costs is where the election results will bite the most. Before the election, Microsoft trimmed its capex forecast to $170 billion from $175 billion because third-party data-centre capacity was late. It expects spending to pick up in the first half of 2027 on higher component costs and delayed deliveries, with most of the next year's capex rise being price, not volume.
The biggest challenges will come from the states and the new Democratic leadership—Democratic governors-elect in Texas, Ohio, and Georgia promised restrictions. Congress is largely irrelevant in this. A federal moratorium or tax needs 60 Senate votes and would be vetoed, so the real hurdles will lie in the states. Pennsylvania is the template. Shapiro’s August order makes developers build, bring or buy their own power and pay for grid upgrades. Ohio residential power prices are up about 26% since early 2023, so the politics persist. Projects with approvals mostly proceed. Before the election, SemiAnalysis counted about 2.3 GW of US capacity actually delayed by moratoriums, against 38 GW of new capacity expected in 2027. The cost arrives through interconnection queues, cost allocation and rates. Private credit is also exposed, but the stress was already there before the elections. As of early October, seven of the ten non-traded private credit funds I could track had prorated redemptions, paying out between 13% and 48% of what holders asked for. This is a liquidity and rates issue, with data centres as one more layer on top.
Pre-energisation construction loans and leveraged neoclouds are the most exposed, because a two-year delay in getting power can break a fund even with a strong tenant. Developer and neocloud equity take the first hit. Private credit follows more slowly, through marks, covenants and redemptions. Its stress predates the election: seven of ten non-traded funds I track prorated redemptions, paying 13% to 48.5% of requests, largely on software-loan fears rather than data centres. Hyperscaler-backed leases are largely insulated.
Warsh now has the room now to do what he has been indicating since taking office. The Fed held at 3.75–4.00% last week on a 10–2 vote after September's hike, and futures put the odds of another 25 basis points on 9 December at roughly 75–85%. Before the elections the market had the policy rate heading to nearly 4.7% by September 2027. After the vote it prices at least 50 basis points, and a full 100 over four quarters is now in the realm of possibility with the headline CPI at 3.7% and Brent at $104. Money supply barely moves. M2 was $23.4 trillion in September, and velocity has sat at 1.42 for three quarters. If data-centre lending slows, bank credit slows with it, so M2 growth cools rather than rises.
The long end is where it bites. The 10-year closed Wednesday at 4.89%, still 0.73 points above a year ago, and the 30-year is around 5.3%. Divided government makes new tax cuts and spending harder, but the 2027 debt ceiling, tariff revenue that can't be replaced if tariffs come down, steady hyperscaler bond supply and Brent at $104 all keep the Fed leaning towards higher funding costs. At the margin that will slow down AI capex, mostly for developers and neoclouds that rely on lenders, less so for the hyperscalers.
The buildout continues, but it gets slower to permit, costlier to power and harder to fund at the edges. The risk I am watching is not a collapse. It is that the cost reaches household electricity bills before it reaches earnings, and voters have just shown what they do with that!
#USEconomy #AIInfrastructure #PrivateCredit #Macro #USMidterms
More in this area
Debt, Inflation, and the Narrowing Room for Policy—The Rising Cost of Running the State
I listened to Andy Burnham’s first Labour conference speech as Prime Minister, as he set out a larger role for the British state: a national care service, more council housing, changes to pensions and greater public involvement…
When the Money Goes Nowhere — and Everywhere at Once
Today, I heard the phrase used by Sunaina Sinha , who described the current state of IPOs and the flood of capital into equity markets as the "Party of Capital". This resonated deeply with me. I may sound like a broken record to…
The Rupee’s Catch‑22: How India’s Weak Currency Became Its Own Worst Enemy
I am watching the current debate raging on social media on the decline of the Indian currency with some intrigue. From 1994 to 2024, I have seen the Indian rupee slip from roughly 31.37 to the dollar to about 83.51, a…