The AI Boom Just Drove S&P 500 Profits 51% Higher, but Its Biggest Tailwind Is Fading

S&P 500 Q2 EPS rose 51% year over year and trailing-12-month EPS increased 26%, with Goldman Sachs attributing nearly half of 2026 earnings growth to AI-related investment. Hyperscalers are projected to spend $800B on capex in 2026, rising to $1.2T in 2027 and $1.4T in 2028, but depreciation could subtract 5 percentage points from 2027 S&P 500 earnings growth and fully offset incremental AI-capex benefits by 2028. Goldman still forecasts 11% annual EPS growth to $415 in 2027 and $460 in 2028 and maintains an 8,700 S&P 500 target, about 14% above 7,743, contingent on AI-driven productivity gains replacing fading infrastructure, memory-margin, and unrealized-investment-gain tailwinds.

Analysis

The investable issue is not aggregate AI capex, but the transfer of economics from asset-light platforms to the physical and semiconductor supply chain. As hyperscalers move from a period of rapidly rising spend to one of still-high but decelerating spend, the first earnings revisions should remain concentrated in power-constrained infrastructure—GEV, VRT, CEG, ETN and PWR—rather than in AMZN, GOOG, META or MSFT, where incremental depreciation and financing costs are retained internally. This creates a 6-18 month dispersion trade: suppliers with booked backlog and constrained capacity can sustain margins even if hyperscaler capex growth slows; commodity-like server and memory suppliers are more exposed once incremental orders normalize.

The critical near-term catalyst is quarterly disclosure of AI revenue monetization relative to depreciation and capex, not another capex headline. For the platforms, the market can tolerate elevated investment for the next 1-3 quarters if advertising, cloud, enterprise software or consumer engagement show measurable incremental revenue; absent that, free-cash-flow estimate cuts will likely matter more than EPS because accounting earnings can lag capex through useful-life assumptions. GS benefits tactically from continued AI-related issuance, M&A and equity-market activity, but its read-through is sentiment-sensitive rather than a direct infrastructure exposure.

Consensus is likely too focused on whether spending peaks and insufficiently focused on useful-life risk. A shift toward shorter accelerator refresh cycles, or impairment/accelerated depreciation, would pull costs forward and pressure platform returns even if demand remains robust; conversely, credible productivity savings in non-tech sectors would broaden earnings leadership and reduce concentration risk. The article's aggregate earnings framing is not independently sufficient to justify a broad-market trade: exclude non-cash investment marks and monitor cash EPS, FCF conversion, and AI revenue disclosure before treating reported profit growth as durable.

Key Decisions for Investors

• Maintain a 6-12 month long VRT/GEV basket versus short QQQ or an equal-dollar basket of AMZN, GOOG, META and MSFT. The pair isolates the shift from capex beneficiaries to capex funders; target a 10-15% relative return, with a stop if hyperscaler aggregate capex guidance is cut by more than 10% or VRT/GEV backlog conversion weakens for two consecutive quarters.
• Add CEG and PWR on pullbacks as 12-18 month power-bottleneck exposure, rather than chasing memory suppliers. Data-center load growth is valuable only where interconnection and transmission are scarce; key invalidation signals are delayed utility load forecasts, regulatory disallowance of data-center-linked rate recovery, or a sustained decline in forward power prices.
• Avoid new outright longs in AMZN, GOOG, META and MSFT solely on AI capex announcements over the next 1-3 quarters. Upgrade selectively only where management quantifies AI-linked revenue or operating-cost savings sufficient to preserve FCF conversion; a material upward revision to cloud/ads/software monetization would falsify the underweight platform view.
• Use GS as a tactical long only into capital-markets activity data and earnings, not as an AI proxy. Favor a 3-6 month position if IPO, convert issuance and strategic-advisory pipelines accelerate; reduce if equity issuance slows or market volatility rises enough to impair underwriting windows.