In this month’s memo, we break down the growing debate around an “AI Bubble” and examine why the unprecedented surge in AI infrastructure spending, despite rising investor concerns, may ultimately create a powerful tailwind for expansion-stage enterprise AI companies.
We look at where risks are forming in the ecosystem, how physical and economic constraints are shaping industry sentiment, and why enterprise AI demand, capabilities, and company performance continue to strengthen even amid macro uncertainty.
Here’s a preview:
- AI Infrastructure CapEx is reaching historic levels: Hyperscalers are projected to spend $591B in 2026 and $700B in 2027 on data center CapEx, with forecasts reaching $1.4T by 2030.
- OpenAI’s spend-to-revenue gap is fueling skepticism: The company expects ~$20B ARR this year while committing to $1.4T+ in infrastructure, raising questions about long-term economics.
- Capacity remains tight, even as buildouts scale: Microsoft, Google, Amazon, CoreWeave, and Nebius continue to report sold-out capacity and strong demand that exceeds supply.
- Power (not chips) is becoming the constraint: U.S. data center power needs are set to rise to 325–380 TWh by 2028, requiring 100 GW in new generation.
- An “industrial AI bubble” would benefit enterprise AI startups: An eventual oversupply of compute could significantly reduce GPU pricing, lowering COGS and improving margins for AI-native companies.
- Enterprise AI demand is accelerating: AI remains the top C-suite priority, with cloud providers reporting rapid growth and AI-native companies reaching $500M–$1B+ ARR far earlier than prior software waves.
- This market looks nothing like the DotCom era at the application layer: Expansion-stage AI companies today have real revenue, strong unit economics, and validated demand, not speculative valuations on minimal revenue.