Ohio Voted On Data Centres And Seoul Hit A Circuit Breaker

Technology stocks recorded their steepest one-day fall since the onset of the pandemic on 5 November, as investors concluded that the US midterm results would slow data centre construction.

The Nasdaq 100 fell 6.36 per cent, its worst session since March 2020, taking the Nasdaq more than 10 per cent below its record. The S&P 500 dropped 4.36 per cent and the Dow Jones Industrial Average 3.44 per cent, closing 1,833 points lower and erasing the previous day's post-midterm rally. Estimates of market value erased range from roughly $2.5 trillion to $3 trillion depending on the index measured.

The Philadelphia Semiconductor Index fell as much as 11 per cent intraday for its steepest daily loss since March 2020. Micron and AMD each lost around 15 per cent, Nvidia fell around 9 per cent, erasing its gains for the year, and Arm and Dell dropped about 10 and 9 per cent, wiping roughly $20 billion from each. Astera Labs and Applied Digital fell about 11 per cent, Vistra 14 per cent and Vertiv 13 per cent. CoreWeave and Nebius dropped roughly 10 and 15 per cent, leaving CoreWeave down 76 per cent from its June 2025 peak and Nebius 58 per cent from its June 2026 high.

The selling followed Democratic gains in the midterms and voter approval of data centre restrictions in 40 of 41 local ballot measures across 10 states. UBS estimates nearly a third of planned new US data centre capacity is at risk. Wedbush's Dan Ives wrote that hyperscaler capital spending should still exceed $1 trillion in 2027, but that growth into 2028 will moderate significantly given the political map, particularly in Texas.

The selloff deepened across Asia on 6 November. The MSCI Asia Pacific Index fell 3.9 per cent, Japan's Nikkei 225 dropped 6.2 per cent for its largest one-day loss since April 2025, and Taiwan's benchmark fell 4.9 per cent. South Korea's Kospi dropped as much as 8.8 per cent, triggering a circuit breaker, with Samsung Electronics and SK Hynix each losing more than 6 per cent.

The memory companies took among the heaviest losses, and they are the ones whose own guidance says this should not reach them for years.

SK Hynix said in late October that the memory shortage would run to 2030 and that more than half its production was already booked for the next two years. Micron said in October it had sold most of next year's output and expected shortages into 2028. Both fell more than 6 per cent and around 15 per cent respectively on news that American data centre construction may slow from 2028. If those order books are genuinely contracted, a construction slowdown cannot touch the revenue they have already described as spoken for.

So one of three things is being priced. The market may not believe the backlogs, which would be a serious charge against two companies that stated them in public filings weeks ago. It may be looking past them to 2029 and 2030, when the contracted period ends and the replacement demand depends on a buildout that is now politically constrained. Or it may not be repricing earnings at all, and is simply compressing the multiple it will pay for the same earnings, which is what usually happens when a crowded trade finds a reason to unwind.

The third is the most likely and the least dramatic. Memory is cyclical and has always traded on the market's view of where the cycle turns rather than on this year's contracts. What changed on Thursday was not the order book. It was the date at which investors expect the cycle to turn, which moved closer.

The uncomfortable part is how little information produced the move. No company revised guidance. No contract was cancelled. Forty-one towns voted, and roughly $3 trillion changed hands on an inference about what that implies for capital spending three years out.

The thesis that held on Wednesday did not survive Thursday, which is worth recording.

On Wednesday the market split cleanly. Data centre builders were sold while Microsoft rose 6.7 per cent, Amazon 6.1 and Alphabet 3.2, on the reasoning that scarcity created by regulation accrues to whoever already owns capacity. It was a coherent argument and it lasted one session. By Thursday the selling had spread to the whole complex, Nvidia included, and the Dow gave back all of its post-election rally.

That tells you the first day was a rotation and the second was a de-rating. A rotation moves money between winners and losers within a theme. A de-rating questions the theme. When Nvidia loses a year of gains on news about municipal zoning in Ohio, the market is not reallocating within AI; it is marking down what AI infrastructure is worth.

Dan Ives' framing is the useful one, because it separates the two horizons. Capital expenditure above $1 trillion in 2027 is already contracted and will happen. What is being repriced is 2028 and beyond, where the growth rate depends on breaking ground in places that have just voted against it.

For Indian technology services the read is mixed and mostly unhelpful. Slower data centre construction does not reduce demand for the application layer, and the firms building AI services on top of rented compute are not the ones exposed. But Indian IT trades on US technology sentiment regardless of exposure, and a week like this moves the whole sector whether or not the connection is real.

For daily, sharp analysis of the biggest moves in the Indian business and startup ecosystem, follow StartupFox.