Hyperscalers commit 725 billion to AI; valuations repriced but spending intact

The AI infrastructure capex supercycle is accelerating in absolute terms, even as equity valuations for semiconductor and AI-optimised companies face headwinds from delayed Fed rate cuts. Hyperscalers are committing 725 billion dollars to AI infrastructure, and this deployment is no longer speculative; it is embedded in capex budgets and order books. Amazon reported negative 18 billion dollars in free cash flowCash generated after maintenance capex; the actual money the business throws off. in Q1 2026 on the back of 44 billion dollars in property and equipment purchases, yet rallied 45 percent over 6 to 12 months as investors recognise that this capex is foundational for long-term competitive positioning. Palantir's US revenue doubled year-over-year, and management notes that hyperscalers are committing these sums, creating a structural tailwind for companies that can service the buildout.

Semiconductor vendors are the primary beneficiaries. Broadcom is positioned to capture 60 percent of the custom AI chip design services market by 2027, per Counterpoint Research, as hyperscalers prioritise in-house silicon to reduce costs and differentiate. Broadcom's AI revenue surged 106 percent year-over-year to 8.4 billion dollars, yet equities analysts have been slow to reprice the stock, creating a divergence between tape momentumThe empirical fact that winners keep winning over the medium term. and analyst sentiment. AMD, up 47 percent year-to-date on enterprise-AI cloud momentum and RXT partnerships, is facing a similar valuation reset. NVIDIA's earnings on May 21 will be a critical inflection; consensus expects sustained AI revenue growth, but forward guidanceCompany-issued forecasts of future financial performance. and capex commentary from their customer base will determine whether the narrative holds.

The macro crosswind is real. Higher rates and energy costs squeeze datacenter margins, and some operators are deferring non-essential buildouts. However, the fundamental drivers (AI model scaling, training on proprietary data, inference serving) require hardware deployment at scale, and the competitive pressure to build is only intensifying. Smaller players like Super Micro Computer face supply-chain and quality headwinds but remain in the capex chain. Private credit and lending to datacenter operators are seeing strong deal flow, but rising rates and potential energy cost pressures are creating credit underwriting challenges.

Skeptics note that AI spending could decelerate if model scaling hits diminishing returns or if capex multiples become unsustainable. However, the commitment from hyperscalers is now contractual and embedded in their strategic roadmaps. The repricing of growth equities from delayed rate cuts is partly a normalization of inflated multiples from late 2023 and early 2024, not a demand destruction signal. Barring a major geopolitical or macro shock that slashes enterprise budgets, the AI capex cycle is locked in through at least 2027.