Will there be a significant pullback in AI capital expenditures (capex) by December 31, 2026?

probability: 13%

rationale:

Current Status: AI Capex is Accelerating

As of mid-August 2026, there is no evidence of a broad AI capital expenditure pullback; rather, spending is accelerating. Major hyperscalers have consistently raised their forward guidance through the first half of the year. Alphabet lifted its 2026 capex guidance to $195B-$205B abc.xyz, Amazon increased its target to roughly $220B reuters.com, and Meta raised its range to $130B-$145B reuters.com. Aggregate calendar-Q2 2026 capex among the top spenders rose sharply to roughly $171.2B from $131.6B in Q1. Consequently, the overarching requirement of a "significant reduction in AI-related capital expenditure plans or spending" is squarely contradicted by current macroeconomic data.

Why Most Resolution Pathways are Closed

Pathways requiring aggregate spending declines or broad consensus of a pullback are mathematically or practically out of reach before December 31, 2026. Two consecutive quarter-over-quarter aggregate declines cannot occur because Q4 results will not be reported until early 2027. Similarly, seeing at least two companies slash forward guidance by 15% or more is highly improbable given that the only remaining major reporting catalyst is Q3 earnings in late October. Finally, the financial press consensus frames the current environment as a period of massive spending under financing strain and eventual decelerating growth—such as UBS projecting hyperscaler capex growth to slow from +76% in 2026 to +25% in 2027 reuters.com—which the resolution criteria explicitly exclude as a qualifying pullback.

The Primary Ambiguity: Project Cancellations

The main risk of a positive resolution rests heavily on a literal interpretation of the project-delay criterion. There is clear, widely reported evidence of severe project-level friction. Amazon, Microsoft, and Google have all abandoned multibillion-dollar data centers recently due to community opposition reuters.com. Furthermore, Microsoft froze significant portions of its U.S. and European data-center pipeline reuters.com, and Oracle and OpenAI reportedly abandoned a planned expansion near the Abilene, Texas Stargate site reuters.com. An overly mechanical assessment could look at these three or more cancelled megaprojects and trigger a positive resolution.

Supply-Side Bottlenecks vs. Demand-Side Pullback

However, these cancellations overwhelmingly represent supply-side bottlenecks—such as power grid constraints, local permitting pushback, and equipment delays—rather than a structural retrenchment in spending plans. Capital and capacity are generally being reallocated to other campuses. For example, evidence suggests that despite headlines claiming massive capacity cancellations, North American hyperscaler self-build forecasts have moved by only about 1% newsletter.semianalysis.com. Because the umbrella clause of the criteria requires these cancellations to serve as evidence of an actual reduction in overall spending plans, interpreting localized permitting failures as a systemic capex pullback defies the core context of the question.

Conclusion and Tail Risks

The 13% probability heavily weights the overwhelming evidence of continued capex expansion against two specific tail risks. Weighing this timeline against macroeconomic expectations for 2027 emphasizes the heavy structural inertia protecting near-term capital expenditure. First, there is meaningful resolution ambiguity: a strict, literal reading of the project-cancellation pathway could result in a positive resolution despite rising aggregate spending. Second, there remains a marginal risk that sharp late-2026 credit market stress or financing repricing forces one or two major firms to unexpectedly guide 2027 capex materially lower during their late October earnings calls. Outside of these edge cases, all current indicators point strictly toward continued, heavily funded infrastructure growth.