Are AI stocks building on thin ice?

Over the past year, broad AI equity baskets returned roughly 10–25%, while many AI infrastructure names and leaders posted far higher single‑name gains and drawdowns, so averages hide sharp dispersion and volatility. Over two years, AI‑thematic indices show mid‑teens annualised returns, reflecting a strong 2023 and a flatter 2024–2025 path. This gap between selective outperformance and modest basket averages is one reason some investors worry about a bubble in parts of AI rather than the whole market.

Core AI suppliers have delivered exceptional earnings growth, with Nvidia leading the way as its latest quarterly revenue rose 56% year over year and margins remained very high, demonstrating genuine profit expansion rather than just narrative-driven gains. Hyperscalers cite robust AI‑related cloud growth and capacity constraints, with Azure in the high‑30s to ~40% and Copilot/AI services adding meaningful points to growth, signalling monetisation progress even as capex surges. A few platforms and chip suppliers drive this profit and revenue momentum, so applying it to all AI-labelled companies can overstate the sector’s overall earnings power.

Order books are swelling across AI infrastructure and cloud, giving multi‑year visibility if projects convert on schedule. Oracle signalled that its cloud backlog, worth hundreds of billions and driven by AI demand, requires careful management due to concentration and timing risks. Analysts also note that the backlog becomes cash only when the services are delivered. Meanwhile, AI is solving practical problems today, copilots that lift workplace productivity, healthcare triage and imaging support, faster drug discovery and engineering simulation, stronger cyber detection, and infrastructure optimisation across chips and data centres, all pointing to real utility beneath headline‑grabbing valuations.