Your raise used to go offshore. Then it went to a buyback. Now it’s going to a data center

Goldman Sachs argues the global economy has shifted into a new capex-driven cycle marked by higher real rates, geopolitical fragmentation, and a surge in AI-related capital spending. The top five hyperscalers are expected to spend about $755 billion on capex in 2026, up 84% from 2024, while S&P 500 buyback growth has slowed to +1% versus +38% capex growth in Q1 2026. The report implies a broader market leadership shift away from buybacks and mega-cap tech toward capex beneficiaries, industrials, energy, defense, and infrastructure.

Analysis

The market implication is not simply “AI is bullish.” The more important second-order effect is that capital intensity is reasserting itself as a return driver, which should compress the valuation gap between asset-light software and the picks-and-shovels complex that monetizes the buildout. That favors firms with durable pricing power over scarce inputs—advanced lithography, power management, grid equipment, and high-end interconnect—while making it harder for companies whose growth depends on perpetual multiple expansion rather than incremental cash flow.

The most interesting competitive dynamic is that hyperscalers may become both customers and competitors to the same industrial ecosystem. Their capex binge supports suppliers now, but over 12–24 months it also risks overbuilding if AI monetization lags infrastructure spend, especially in power, cooling, and custom silicon. If utilization disappoints, the market will rotate from “who is spending?” to “who is earning on that spend?”, which usually narrows winners to the few names with mission-critical bottlenecks.

A key contrarian read is that higher rates do not uniformly help the “real economy” basket. Rising real yields can eventually pressure long-duration equity multiples, and the first leg of the trade may already be crowded in the obvious beneficiaries. The cleaner expression may be to own the infrastructure enablers and hedge the broad index, rather than chasing mega-cap AI beta at elevated concentration risk.

The main catalyst to watch is whether capex converts into revenue and free cash flow within the next 2–4 quarters. If AI workloads fail to accelerate enterprise spending, or if policy/geopolitical friction eases faster than expected, the market could unwind the capex-premium trade quickly. Conversely, a further rise in bond yields would likely intensify the relative advantage of firms with contracted demand and regulated returns.

Key Decisions for Investors

• Long ASML / LRCX as the highest-quality semicap bottleneck exposure into 2H26; use a 6-12 month horizon and expect outperformance if hyperscaler capex remains >30% y/y, but trim on any evidence of order normalization.
• Long VST as a power-scarcity proxy against AI infrastructure buildout; pair with a short in a long-duration software basket to isolate capex intensity, targeting 3-6 months with upside if grid constraints keep power pricing firm.
• Initiate a relative-value long NVDA / short GS or broad Nasdaq hedge only if you want pure AI beta; otherwise avoid chasing NVDA at elevated concentration risk and use call spreads rather than outright stock for defined downside over 3-6 months.
• Build a basket long AMZN, MSFT, GOOGL, META, ORCL only on pullbacks, with an explicit stop if capex growth decelerates while revenue guidance stays unchanged; this is a 12-month trade on monetization, not just spend.
• Pair long industrial/energy infrastructure winners (ASML, AVGO, VST) against short-rate-sensitive, capital-light growth names; the risk/reward improves if real yields keep rising and the market rotates away from duration.