Reality check for artificial intelligence: correction or end of cycle?

The AI investment cycle is still in an early phase and the June 2026 sell-off is better viewed as a correction than as evidence that the cycle has ended.

• Since ChatGPT launched in November 2022, the MSCI Emerging Markets Information Technology Index has risen more than fivefold, while cloud revenue growth has reaccelerated towards 50%; OpenAI and Anthropic reportedly quadrupled revenue in 2025.
• US hyperscalers are expected to spend roughly USD 745 billion on data centres this year, with capex growth above 80%; their debt issuance reached USD 194 billion year to date by July 9, making financing a risk to monitor but not yet a red flag.
• Cloud businesses are growing at almost 50% with operating margins of 35-40%, while a well-utilized current-generation data centre can generate around 20-30% ROIC; returns on the broader capital base remain around or slightly below the cost of capital because much capacity is still under construction or uses older chips.
• Supply remains constrained: TSMC announced an additional USD 100 billion investment in US advanced-chip facilities, Samsung Electronics and SK Hynix announced expansion plans exceeding USD 1.7 trillion and USD 750 billion, and memory prices have risen roughly tenfold since early 2025.
• Valuations after the correction stand at 49% of the dot-com peak in US technology, 28% in emerging markets and 39% in Taiwan, measured on an economic P/E basis; enterprise adoption and agentic AI workflows are still at an early stage.

Takeaway: Investors can continue to gain exposure to the long-term AI growth theme, but should monitor hyperscaler leverage, circular financing, data-centre bottlenecks, supply-demand balance and the path from training costs to profitable inference.