Big Money Sees Gold in AI - But Isn't Buying the Rush
At Abu Dhabi Finance Week, the biggest names in capital weren't debating if AI matters. They were focused on where the durable returns will come from - and how to avoid overpaying for growth that may take years to hit earnings.
The consensus: AI-equity valuations look stretched, but the infrastructure behind AI - power, data centers, networking - is where disciplined capital can still earn real yield.
Valuation Heat vs. Infrastructure Pull
Franklin Templeton's CEO compared today's AI trade to the early days of a gold rush: the tools are getting expensive, but the real scale hasn't shown up in earnings yet. Translation for allocators: expect a long buildout, intermittent hype cycles, and a lag before productivity gains widen margins at the company level.
That argues for patience on headline AI equities, and a tilt toward "picks and shovels" assets with contracted cash flows.
Electricity Is the Constraint
Blackstone's top brass put it plainly: AI demand implies a step-change in electricity consumption. Doubling grid capacity isn't a spreadsheet exercise - it requires permits, transmission, generation, and time.
For investors, that spotlights opportunities in grid upgrades, flexible generation, long-duration PPAs, and developers with interconnection priority. It also raises risk for AI projects without secured power or cooling.
For context on the power curve, see the IEA's work on data center energy demand here.
Selective on AI Equities, Constructive on Cash-Flow Assets
Sovereign wealth allocations out of the Gulf still like AI and biotech, but with the view that we're midway - not at the finish line. That means staying invested, yet disciplined on entry points.
On the hedge fund side, skepticism is rising. Some managers warned that plenty of "AI plays" have weak unit economics today. Disruption is real; it won't lift every stock.
Data Centers: Attractive, but Don't Overpay
Several leaders called data centers the most investable piece of AI right now - if underwriting isn't built on rosy timelines. The caution: paying high multiples that assume rapid ramp can turn equity into a value trap.
Even major vendors are absorbing heavy capex, with free cash flow set to run negative for years. Others emphasized sticking to principles instead of chasing growth for its own sake.
What This Means for Portfolios
- Favor infrastructure with contracted revenues: data centers with anchor tenants, PPAs, and interconnects in hand.
- Underwrite power first: site selection near cheap, reliable electricity; water availability; realistic timelines for grid connections.
- Be selective on AI equities: prioritize firms with pricing power, distribution moats, or proprietary data - not just model access.
- Consider credit over equity for late-cycle builds: structured capital into data centers and energy assets can offer superior risk-adjusted returns.
- Expect dispersion: AI will boost some margins while compressing others; avoid broad-brush exposure.
Questions to Pressure-Test Any AI-Infra Deal
- Is there a long-term PPA and interconnection approval in place? What's the true queue position?
- How is cooling handled and at what water and energy cost per MW?
- What's the anchor-customer mix, contract tenor, and renewal history? Any capacity pre-sold?
- What multiple implies your base case IRR - and how sensitive is it to a 12-18 month delay?
- Are you underwriting tech risk (chips, racks, thermal limits) or just real estate and power?
Bottom Line
AI is real; the cash flows are in the pipes, power, and racks. Bid discipline matters more than vision statements. If you have to stretch the model to make the deal clear, it isn't.
For finance teams mapping AI exposure across tools and vendors, a curated overview can speed due diligence. Start with this practical roundup of AI tools used in finance here.
Your membership also unlocks:
