Will there be a significant pullback in AI capital expenditures (capex) by December 31, 2027?

probability: 15%

rationale:

As of mid-August 2026, the trajectory of AI capital expenditure is characterized by aggressive acceleration rather than contraction. Every major hyperscaler has recently raised, rather than cut, their near-term capex guidance. Alphabet lifted its 2026 guidance to $195–205B and signaled a significant increase for 2027 2 sources. Amazon raised its 2026 outlook to $220B, citing capacity shortages that will persist into 2027 cnbc.com. Microsoft guided its fiscal Q1 2027 capex to over $50B to meet demand that outstrips supply 2 sources, while Meta and Oracle similarly increased or maintained historic spending levels 2 sources. The baseline environment firmly points away from a near-term pullback.

A critical distinction in this assessment is the difference between slowing growth and an outright contraction. Consensus expectations point toward a sharp deceleration in spending growth, not a decline. For example, UBS models hyperscaler capex growth slowing from 76% in 2026 to 25% in 2027, and eventually to 6% in 2028 reuters.com. Under these projections, quarterly spending throughout 2027 remains substantially higher than 2026 levels. A deceleration from an extraordinary base does not satisfy the requirements for a true year-over-year pullback or a >10% cut in guidance explicitly attributed to moderating AI investment.

Furthermore, structural inertia in infrastructure investments makes a sudden drop in reported capex highly unlikely before the end of 2027. AI infrastructure spending is largely locked in by multi-year supply agreements, land and shell contracts, and massive equipment backlogs—such as AWS's reported $496B backlog and Oracle's $638B in remaining performance obligations 3 sources. Additionally, supply-side constraints like elevated memory prices and component costs mechanically inflate reported dollar capex even if the volume of unit deployments slows 2 sources. While power and permitting bottlenecks may delay specific projects, underspending due to supply constraints is fundamentally different from a demand-driven investment cut.

Despite this momentum, legitimate vulnerabilities keep the probability of a pullback meaningfully above zero. The current investment cycle is unusually leveraged, with hyperscaler capex increasingly funded by debt and projected to exceed incremental operating cash flows 2 sources. There are early warning signs of fragility, including reports of OpenAI missing internal revenue targets and scaling back specific regional projects 2 sources. The Bank for International Settlements has warned that disappointing AI returns could trigger a sudden financing pullback bis.org. If a sharp credit event, monetization failure, or anchor-customer collapse materializes in 2027, it could force rapid order cancellations and industry retrenchment, generating widespread media coverage of an AI infrastructure "slowdown" that fulfills the criteria.