April Was the Best Month for the Market Since 2020. Here's What's Driving It.

AI spending is expected to reach about $670 billion in 2025, with UBS seeing roughly $770 billion by 2026 and Goldman Sachs projecting nearly $800 billion in 2027. The article argues that this AI build-out is driving a meaningful share of S&P 500 EPS growth and supporting a broad rally across hyperscalers, semiconductors, construction equipment, power technology, and utilities. Notable moves cited include Nvidia up 20% in April, Micron up 61%, GE Vernova up 33%, and the S&P 500 gaining about 10.5% in April to a new all-time high.

Analysis

The market is increasingly pricing AI as a multi-year industrial buildout rather than a transient software rerating, and that matters because capex has a longer earnings tail than narrative-driven multiple expansion. The second-order effect is that the winners are shifting from “model owners” to the companies that monetize the bottlenecks: power delivery, memory, networking, and data-center construction. That widens the trade from a narrow mega-cap AI basket into a broader capital-expenditure complex that can keep compounding even if end-demand growth normalizes.

The most underappreciated dynamic is constraint inflation. Memory scarcity, grid interconnection delays, turbine lead times, and transformer shortages all create pricing power for suppliers with scarce manufacturing capacity, while pressuring hyperscalers to finance growth more aggressively. If the capex cycle extends into 2027, the market will likely reward suppliers with near-term earnings revision momentum much more than it rewards the hyperscalers themselves, because the latter face eventual ROI scrutiny and depreciation drag.

The main risk is that the trade becomes self-defeating: if AI spending accelerates faster than monetization, investors may start discounting lower free-cash-flow conversion for the hyperscalers and re-rate the basket lower even while revenues grow. A slower-than-expected enterprise adoption curve, or any pullback in bond issuance / capex guidance, would hit the “AI supercycle” narrative within 1-2 quarters. Conversely, if power or memory constraints persist, the supercycle can extend longer than consensus expects, but with returns concentrated in picks-and-shovels, not platform names.

The contrarian view is that the crowd is still too focused on Nvidia-style exposure and not enough on utilities and industrial enablers where estimates are less stretched and revisions are just beginning. The best risk/reward likely sits in names with operational bottlenecks and pricing power, not the most obvious AI leaders. That argues for a barbell: own the infrastructure beneficiaries and selectively fade the most crowded expressions of the story if multiples outrun forward EPS revisions.

Key Decisions for Investors

• Overweight MU on a 3-6 month horizon: memory shortage plus pricing power should keep earnings revisions positive; risk/reward favors continued upside if supply remains tight into 2027.
• Buy GEV / CAT as a basket against a short in the most expensive hyperscaler exposure over 6-12 months: infrastructure beneficiaries have cleaner near-term EPS leverage, while hyperscalers face rising depreciation and financing costs.
• Initiate a long D position or call spread for 12 months: data-center power demand is a multi-quarter catalyst, but size smaller due to regulatory and rate-base execution risk.
• Use NVDA strength to sell upside via call overwrites or reduced delta into rallies: the stock still benefits, but the market may be overpaying for already-visible growth while supply-chain beneficiaries have better incremental upside.
• Pair long XSD or SOXX versus short a basket of higher-multiple software names over 6 months: the article’s setup favors capex suppliers over downstream software beneficiaries if AI spend remains the dominant earnings driver.