Hyperscalers commit $725B to AI, capex cycle peaks

Hyperscale data-center capex is becoming the dominant capital deployment story in tech. Palantir highlighted that US hyperscalers have collectively committed $725 billion to AI infrastructure; AWS, Google Cloud, and Azure are racing to secure chips, power, and real estate. NVIDIA's Q1 guidanceCompany-issued forecasts of future financial performance. and May 21 earnings report will be the key catalyst, with markets pricing in sustained margin strength but watching for any signs of demand normalization or inventory build.

Supply-side pressures are mounting. Innio Holding (gas engine manufacturer) and Fervo Energy (geothermal) both filed IPOs or raised targets, betting on a decade of elevated energy demand from data centers. AMD, trading at $465 in early May, saw a 47% YTD rally on the back of RXT enterprise-AI cloud momentumThe empirical fact that winners keep winning over the medium term. and strong tape. Broadcom (AVGO) is similarly positioned as a chokepoint for AI infrastructure. However, the Hormuz closure introduces a new wildcard: semiconductor logistics depend on oil-powered shipping, and any sustained disruption could squeeze supply chains.

The narrative is bullish on capex acceleration but faces a valuation risk: if the Fed holds rates higher for longer due to the oil shock, the present value of hyperscaler cash flows declines sharply. Additionally, Amazon reported Q1 free cash flowCash generated after maintenance capex; the actual money the business throws off. of negative $18 billion following $44 billion in property-equipment capex; the stock dipped 12% short-term on FCF concerns, though it rallied 45% over six to twelve months as growth materialized. This pattern (near-term pain, long-term gain) is repeating across big tech.

Sceptics warn that capex-to-revenue growth is not infinite and that AI model training may hit diminishing returns, forcing a pivot toward inference and edge compute. Additionally, if energy costs spike from the Hormuz closure, hyperscalers could face pressure to reduce data-center footprint or relocate to cheaper-power jurisdictions, potentially delaying capex deployment and deflating the near-term euphoria.