Hyperscaler AI capex drives semiconductor rally

The race to build AI infrastructure is accelerating capex spending at a pace that has left consensus behind. Hyperscalers are committing $725 billion to AI infrastructure buildout, and semiconductor vendors are reaping the benefits. Broadcom, despite equity markets prioritizing other narratives, has surged with AI revenue up 106% year-over-year to $8.4 billion; Counterpoint Research estimates Broadcom will capture 60% of the custom AI chip design services market by 2027. This secular shift is reshaping the competitive landscape: instead of relying on public chip designs, major cloud operators are investing in bespoke silicon tailored to their specific AI workloads.

AMD's stock has run 47% year-to-date on enterprise-AI cloud deals, with MKTBOX scoring the narrative at 73/100 but flagging valuation divergence at just 38. NVIDIA maintains technical strength, with a 78% probability of rising in the next 60 days and a median target of $248.72. Memory companies like Micron have also rallied, up 103.9% on the back of data center buildout and GPU density requirements. Earnings season is reinforcing the story: companies flagged before the move included Micron, Oscilent, Alphabet, and others tied to the AI capex wave.

However, the story now prices in both the opportunity and the timing risk. Palantir's US revenue doubled year-over-year as hyperscalers deploy AI infrastructure, but the broader sector is grappling with whether capex intensity peaks in 2026 or sustains through 2027. Supply chain dynamics favor first-movers: NVIDIA's dominance in training chips, Broadcom's design services, AMD's custom processors, and ARM's architectural licensing all benefit, but the margin compression risk as competition intensifies cannot be ignored.

Valuation concerns are real. Growth scores are in the 90s while valuation scores languish in the 30s to 40s range, suggesting the market is pricing near-perfect execution. If capex growth slows or data center utilization lags expectations, multiple compression could be sharp.