The smart money isn't bailing on AI, it's just done betting on the people selling shovels.
The Summary
- JPMorgan Private Bank says the AI trade is in a "digestion phase" but remains constructive long-term, shifting focus to integration and application layers over infrastructure
- M&G Investments notes investors are "more wary" about whether the capex boom will actually translate into delivered earnings
- The shift signals a maturation from speculation on picks-and-shovels infrastructure to demanding proof that AI applications actually generate revenue
The Signal
The institutional money managing trillions is saying the quiet part out loud: the AI infrastructure trade is overheated, and the smart bet is moving downstream. Sitara Sundar at JPMorgan Private Bank isn't calling AI dead. She's calling the current trade tired. Her firm is pivoting to "integration and application layer" plays, the companies actually using AI to build products people pay for.
This isn't a bear case. It's a rotation. Maria Municchi at M&G Investments sees prices holding but cautions that markets are questioning whether the massive capital expenditure on GPUs, data centers, and model training will convert into actual earnings. Translation: Nvidia had its run. Now show me the companies turning inference costs into profit margins.
"Markets are becoming a little bit more wary about whether the capex boom will translate into delivered earnings."
The timing matters. This comes after AI industry leaders floated slowing development, a move that spooked investors who'd priced in exponential scaling forever. The digestion phase is what happens when hype meets the hard work of monetization. The companies that survive this aren't the ones with the biggest training clusters. They're the ones with distribution, integration into actual workflows, and a path to positive unit economics.
JPMorgan's call to lean into application and integration layers is a tell. They're looking at:
- Vertical SaaS companies embedding AI into existing workflows
- Agent platforms that automate revenue-generating work
- Tools that replace headcount, not just speed up research
This is the transition from Web4 infrastructure to Web4 business models. The infrastructure got built. Now the question is who captures value when agents start doing the work humans used to bill for. The money is following the companies that can answer that question with a number, not a vision deck.
The Implication
If you're holding pure AI infrastructure plays, JPMorgan just told you to take profits and rotate. The new trade is companies that can prove AI drives margin expansion or revenue growth, not just R&D spend. Watch for earnings reports that break out AI-driven efficiency gains in measurable terms: customer acquisition cost, support ticket resolution time, sales cycle compression.
For builders, this is validation. The market is done rewarding potential and starting to pay for performance. If your AI product doesn't have clear ROI within 90 days of deployment, you're competing for a shrinking pool of VC dollars. Build for the application layer. Build for integration. Build for the CFO, not the CTO.