Everything About AI
NextFin News - The worst one-day rout in US technology stocks since April 2025 has reignited the AI-bubble debate, but the forces now slowing the data-center buildout may be doing the real economy a favor. After the Dow Jones Industrial Average fell 1,833 points and the Nasdaq Composite slid 5.5% into correction territory on Thursday, the question is no longer whether the AI investment boom has been excessive. It is whether a politically enforced slowdown in data-center construction could be the soft landing that a market-driven bust would not be.
The sell-off was broad and brutal. The S&P 500 dropped 4.36% to 7,353.73, the Dow shed 3.44% to 51,523.06, and the Philadelphia Semiconductor Index sank 12.25%. Nvidia fell 9.3%, Micron 17.5%, AMD 17.3% and Oracle 7%. Retail traders unloaded $265 million of equities, their biggest single-day exit since 2020. "They're shooting the generals," wrote Nomura cross-asset strategist Charlie McElligott, as investors dumped the winners of 2026 to lock in gains.
Yet the same week, voters across the United States approved dozens of ballot measures restricting data-center development, and Democrats won governorships in Nevada, Georgia, Ohio and Texas on pledges to tighten rules on the facilities. The combination is the story: the AI trade is unwinding at the same moment that the physical buildout behind it is running into political, power and financing constraints. That collision does not guarantee a soft landing. But it changes the shape of the risk - and it may spare the broader economy the kind of abrupt investment reversal that has ended every comparable technology boom in modern history.
The Selloff Is a Valuation Event. The Buildout Slowdown Is a Structural One.
The first distinction matters because the two are being conflated. Thursday's decline was a classic de-rating: a crowded positioning unwind in stocks that had led the market higher for two years. The Nasdaq fell more than 10% from the record it set last month, entering correction territory, while the Magnificent Seven lost close to $900 billion in market value heading into the final hour of trading. Nothing in the day's price action reflected a sudden deterioration in earnings. It reflected a repricing of how much investors are willing to pay for AI exposure when the path from capital expenditure to profit looks longer than it did six months ago.
The buildout slowdown is different. It is physical, contractual and political. Hyperscaler capital commitments have risen to roughly $725 billion in 2026, up about 77% from around $410 billion in 2025. Alphabet has guided to $195 billion to $205 billion, Amazon to roughly $220 billion, Meta to $130 billion to $145 billion, and Microsoft to roughly $175 billion. Global AI investment for the year is forecast to exceed $1 trillion. But money committed is not capacity delivered. Shell and core data-center construction now averages about $11.3 million per megawatt, with tenant fit-out adding up to $25 million per megawatt on top. Substation transformer lead times exceed 160 weeks in 2026, up from around 140 weeks in 2023. Data centers already face $200 billion in project delays, according to Mike Mathews, global digital infrastructure practice leader at risk adviser Marsh.
Into that bottleneck steps the voter. Ballot measures restricting data centers passed in 33 of 45 contests across 10 states, according to the Data Centers on the Ballot tracker; a separate analysis counted 38 of 41 measures approved. Only one municipality, Harwood, North Dakota, voted against a restriction. New York imposed the first statewide moratorium in July, pausing environmental permits for facilities above 50 megawatts. One tracker of state and local restrictions counted measures in force across 46 states; only Alaska, Hawaii, West Virginia and Wyoming had none at any level of government.
The geographic concentration is what makes this economically meaningful. Goldman Sachs Research estimates that come January, states controlled by Democrats who support extending AI construction moratoriums will account for 45% of America's planned data-center capacity. A quarter of the entire US AI infrastructure buildout is planned for Texas alone - and Texas just elected a Democratic governor, Gina Hinojosa, who campaigned on a temporary construction ban. Wells Fargo calculates that for every 10% reduction in planned Texas capacity, Big Tech companies will cut AI infrastructure spending by 1.6%. That is not a stock-market multiple. It is a direct subtraction from the investment line that has been carrying US growth.
Why the Real Economy Has Been Riding the AI Buildout
To understand why a slower buildout could be protective rather than destructive, you have to measure how large the buildout has become. Goldman Sachs Research estimates that AI investment in the US will total almost $600 billion in 2026, equivalent to nearly 2% of US GDP and more than 10% of business fixed investment in recent quarters. ING economists put the contribution even higher: on the broadest measure, information-processing, software and data-center construction investment accounted for 50.2% of year-over-year GDP growth in the second quarter of 2026. After adjusting for imported equipment, ING's figure falls to 36% for the quarter and 37% on a four-quarter average - still roughly one-third of growth.
That concentration is historically unusual. Columbia Business School economist Stijn Van Nieuwerburgh estimates AI infrastructure spending is approaching $1 trillion a year, every year - a share of GDP that would outpace every other investment boom in modern US history, including the construction of the railroads. The Bank for International Settlements, in its Annual Economic Report 2026 released in June, estimated that the five largest hyperscalers are set to spend more than $1 trillion on AI-related capital expenditure across 2025 and 2026 combined. The report noted that these commitments are outpacing earnings and free cash flow, forcing many firms to issue debt. It also found that the sector's net economic surplus - total payoff minus total costs - is declining and could eventually turn negative.
Here is the mechanism that turns a slowdown into a shield. When an investment boom is financed out of operating cash flow and grows in line with demand, a deceleration is absorbed gradually: companies cut capex guidance, suppliers adjust, employment shifts rather than collapses. When a boom outruns cash flow, depends on continuous external financing, and is priced by markets as if the growth rate is permanent, the reversal is abrupt. The Bank for International Settlements put it plainly, comparing the current AI boom to canal mania in 1835, railway mania in 1843, the roaring 1920s and the dot-com boom: these episodes "ended with an eventual reversal in investment, inducing economy-wide recessions."
A politically and physically constrained slowdown stretches the same total investment over more years. That is the difference between a controlled descent and a stall-turn. The demand for compute has not disappeared - Microsoft's Azure still holds an $80 billion backlog of orders it cannot fill for lack of power - but the pace at which capital can be deployed has been capped by three independent constraints: grid interconnection, local permitting, and the cost of external finance.
The Second-Order Effect: What Happens When Capex Growth Decelerates
The first-order effect of a data-center pullback is obvious: less spending on chips, servers, construction and power equipment. The market has already priced much of that. Nvidia, Micron and AMD fell between 9% and 17% on Thursday because investors understand that hyperscaler capex growth cannot stay near 100% forever. Goldman Sachs Research itself projects deceleration in capex growth from nearly 100% in 2026 to 54% in 2027 and 12% in 2028, even under its own higher estimates of $1.2 trillion in 2027 and $1.4 trillion in 2028.
"We actually don't think supply and demand on AI finds a balance until the first half of 2028," said Eric Sheridan, who leads the technology, media and telecommunications research group at Goldman Sachs. "That means more capex, more revenue growth, and the supply chain remains under a lot of constraint."
The second-order effect is where the real-economy protection comes in. A slower, constraint-limited buildout reduces three systemic risks at once.
First, it lowers the financing cliff. When capital expenditure exceeds free cash flow, the gap is filled with debt. The Bank for International Settlements warned that the current tension between exuberant risk appetite and elevated macroeconomic risks "could unwind abruptly." If the buildout is forced to slow because permits and power arrive gradually rather than all at once, the debt issuance needed to fund it also spreads out. The rollover risk that worries credit markets - the moment when a cohort of AI-linked borrowers must refinance into a higher-rate, lower-appetite environment - becomes a sequence of smaller refinancings rather than a single wall.
Second, it reduces the inflation impulse. The AI buildout has not just added to GDP; it has added to demand for the scarcest inputs in the economy - electricity, transformers, copper, skilled construction labor and land. A Goldman Sachs "dashboard" of leading indicators, including semiconductor manufacturing equipment imports in Taiwan and South Korea, memory prices and graphics-processing-unit rental rates, is designed precisely to detect when that demand pressure is peaking. A politically enforced pause lets those markets clear. That matters for the Federal Reserve: an investment boom that pulls one-third of GDP growth from a single source also pulls a disproportionate share of inflation risk from the same source.
Third, it buys time for monetization. The central bear case against the AI trade is not that AI is unimportant. It is that the revenue required to justify the capital stack will not arrive before the capital stack needs refinancing. A slower buildout does not solve that problem, but it pushes the reckoning further into the future - and it does so while the underlying adoption curve keeps rising. That asymmetry is what a soft landing looks like in a capex-led boom: not an end to investment, but a glide path that lets earnings catch up to assets.
There is precedent for this kind of politically mediated slowdown producing a better long-run outcome. In the previous decade, hyperscalers building cloud data centers committed to renewable energy before it was cost-effective, partly in response to community pressure.
"If you want to go fast, you go it alone; if you want to go far you go together," said Harry Krejsa, director of studies at the Carnegie Mellon Institute for Strategy and Technology. "If you squander political license to operate at home, that'll be a bigger impediment to remaining competitive with China."
The Counter-Thesis: A Slowdown Could Tip a Fragile Economy Into Recession
The strongest case against the soft-landing view is straightforward and deserves to be stated without softening. The US economy in 2026 is not running on many engines. Consumer spending has been restrained by tariffs and elevated prices; manufacturing has been hit by trade disruption; fiscal support is fading. Into that environment, the AI buildout has been the one genuinely new source of demand. If investment in AI infrastructure accounts for roughly one-third of GDP growth - and, on the broadest measure, more than half of quarterly growth - then a sharp deceleration is not a stabilizer. It is the removal of the only pillar holding the expansion up.
This is the Bank for International Settlements' core warning, and it is not a fringe view. The bank's comparison to canal mania, railway mania, the 1920s and the dot-com boom is not an analogy about valuation. It is a statement about what happens when investment reverses: economy-wide recessions. The historical record supports the concern. In each of those episodes, the boom was followed not by a gentle normalization but by a contraction in which the capital stock already built became stranded, the workers who built it lost their jobs, and the lenders who financed it took losses that spread through the financial system.
The political constraints make this risk worse, not better, in the bear case. A market-driven slowdown would be led by prices: as expected returns fall, companies delay marginal projects. A politically driven slowdown is led by permits, moratoriums and election outcomes, which can move faster than fundamentals. Texas alone accounts for 25% to 30% of the data-center capacity expected to come online by 2030. If a new governor and a wave of local bans freeze projects that are already under construction or awaiting grid interconnection, the stop is not gradual. Capital already committed - equipment orders, construction contracts, financing agreements - does not unwind cleanly. The $200 billion in project delays already in the system could multiply, and the companies sitting on those commitments would face a choice between writing off assets or continuing to pay interest on capacity they cannot operate.
There is also a global dimension. The US AI buildout has been a growth engine not just for America but for its suppliers: semiconductor equipment makers in Taiwan and South Korea, memory producers in Korea and Japan, and commodity exporters. A US-led pullback would transmit through those channels. The Bank for International Settlements noted that the AI boom was largely responsible for the fact that the global economy proved more resilient than expected in 2025, through investment and through exports for Asian economies. A reversal would run the same channel in reverse.
The falsifying signal for the soft-landing thesis is specific and observable. If US business fixed investment excluding AI-related categories fails to pick up as AI capex growth decelerates - specifically, if non-AI business fixed investment grows below 2% year-over-year for two consecutive quarters while AI capex growth falls below 25% - then the buildout was carrying the economy, not complementing it, and the recession risk moves from theoretical to high. A second signal: if the unemployment rate rises by more than 0.5 percentage point within six months of the capex deceleration beginning, the labor market was more dependent on the buildout than current data suggest.
What to Watch: The Three Constraints and the One Number
The near-term path turns on three constraints and one number.
The constraints are grid interconnection, permitting and financing. Interconnection queues in Texas and Virginia are the first bottleneck to watch: if queue times lengthen rather than shorten through the first half of 2027, the physical slowdown is binding regardless of what companies say about capex. Permitting is the political bottleneck: the governors-elect in Nevada, Georgia, Ohio and Texas each campaigned on tighter rules, and their first 100 days will show whether campaign rhetoric becomes administrative action. Financing is the market bottleneck: hyperscaler issuance has already moved spending beyond operating cash flow, and the cost of that debt will determine how much of the announced $725 billion actually gets built.
The one number is AI capex as a share of US GDP. Goldman Sachs Research puts it at 1.8% in 2026, rising to 2.5% in 2027 and 2.8% in 2028. Those levels sit inside the 2% to 5% peak investment impulse observed in prior general-purpose technology buildouts. The soft-landing case requires that the share stabilize rather than accelerate from here. If it keeps climbing toward the top of that historical range while financing conditions tighten, the boom is extending, not cooling - and the eventual reversal becomes larger, not smaller.
Outlook: Three Scenarios for 2027
Base case - a managed glide path. Data-center deployment slows through 2027 as permits and power arrive gradually, but does not stop. AI capex growth decelerates toward the mid-50% range, in line with Goldman Sachs Research's projection, while non-AI business investment holds up. The stock market absorbs the de-rating through 2026 and early 2027; the real economy grows more slowly but does not contract. Beneficiaries: utilities and grid-equipment makers with visible backlogs, companies with pricing power in transformers and cooling, and hyperscalers that already secured power. Exposed: pure-play AI infrastructure names trading on perpetual triple-digit capex growth, and memory suppliers whose pricing depends on uninterrupted hyperscaler orders.
Upside case - monetization catches up. AI applications reach enterprise adoption faster than expected, revenue per watt rises, and the return on invested capital stabilizes before financing conditions tighten. In this scenario the political constraints look prescient rather than obstructive: they slowed the buildout just enough to let demand arrive. The market re-rates the hyperscalers, and the semiconductor supply chain resumes growth on a healthier foundation.
Downside case - the pillar gives way. Permits and power constraints bite harder than expected, capex growth falls below 25%, and non-AI investment fails to fill the gap. The unemployment signal fires, credit spreads widen on AI-linked debt, and the historical analogy from the Bank for International Settlements plays out. In this scenario the stock-market rout of November 2026 was not the end of the AI trade but the first leg of a broader investment reversal.
Short term, sentiment and positioning dominate: the unwinding of 2026's thematic winners can overshoot, as it did on Thursday. Medium term, fundamentals dominate: the gap between committed capex and delivered capacity will decide which companies keep their guidance. Long term, the structural question is whether AI is a general-purpose technology on the scale of electricity - in which case the buildout will span decades and today's pullback is a pause - or a narrower automation wave, in which case today's capital stock will be the stranded asset of the next cycle.
The market spent Thursday pricing the first possibility. The economy will spend 2027 finding out which one is true. The investors who treat a data-center slowdown as unambiguously bad news are making the same mistake as those who treated the buildout as costless: they are confusing the pace of investment with its value. A boom that cannot be permitted, powered or financed is not a boom - it is a claim on a future that may not arrive. The pullback does not eliminate that risk. But by stretching the buildout over time, it gives the real economy a chance to grow into it rather than break beneath it.
Data as of the November 5, 2026 market close; nf-data snapshot taken during this run. Data cutoff: 2026-11-06.
Explore more exclusive insights at nextfin.ai.
