The Next Phase of AI is an Earnings Story

What investors need to know

• Artificial intelligence’s (AI’s) infrastructure challenge is shifting—not disappearing. Compute was the first bottleneck. Power, cooling, transmission and connectivity are increasingly the next. Higher interest rates may affect valuations, but they don’t solve the physical constraints driving a multiyear infrastructure buildout.
• The next earnings surprise may not come from an “AI company” at all. Industrials, financial services, health care and consumer companies are already using AI to change their own economics. We think productivity gains may be showing up faster than earnings expectations reflect.
• As intelligence becomes abundant, scarcity becomes more valuable. Proprietary data, trusted distribution and embedded workflows are difficult for AI to replicate. As intelligence gets cheaper, we think the value of these scarce assets rises—and a higher cost of capital reinforces the advantage of businesses that already own them.
   

A New Stress Test for the AI Trade

Recent weeks have given the AI trade its first real stress test in years. At its September policy meeting, the Federal Reserve raised interest rates for the first time since 2023, days after the industry's own debate over whether frontier AI development needs to slow down.

Together, those two events raise a fair question for equity investors: Does more expensive capital, arriving alongside a more cautious AI industry, mean the infrastructure investment cycle is over? We don't think so. But the answer runs through earnings, not headlines, and it comes down to three questions.

1. How durable is infrastructure demand?

The market sees a capital expenditure (capex) cycle. We see a capacity problem—and a rate hike doesn't solve it either way.

Compute was AI's first bottleneck. Power, cooling, transmission and connectivity appear to be next. These physical constraints don't scale on a chip's timeline; power generation and transmission take years to permit and build, cooling and connectivity requirements compound with every new data center, and order books across many of these supply chains already extend multiple years.

A higher federal funds rate doesn't solve a transmission bottleneck or shorten a permitting timeline. This scarcity is physical and regulatory, not a function of the cost of capital. It also matters that much of the spending by the largest hyperscalers has been supported by substantial operating cash flow, reducing their sensitivity to higher rates relative to a more highly levered capital cycle.

What a rate hike can do is compress valuations across the supply chain before earnings catch up. That gap, not a slowdown in the underlying buildout, is where we see the opportunity.

2. Where will AI-driven productivity become visible?

The next AI earnings story may come from companies that don’t look like AI companies at all.

Industrials, financial services, health care and consumer companies are already using AI to change pricing, scheduling, inventory and service, often without ever selling an AI product. When a company can grow without adding labor, overhead or working capital at the old rate, its earnings algorithm changes even if revenue growth remains static.

Expensive capital raises the bar on cost discipline, which could accelerate adoption rather than slow it. For many of these businesses, the productivity benefits of AI may not yet be fully reflected in earnings expectations. As those benefits begin to show up in results, we think they could create room for positive earnings surprises in 2027.

3. Which assets become more valuable as AI spreads?

As intelligence becomes cheaper, scarcity becomes more valuable—and higher rates reinforce the distinction.

As AI capabilities become widely available, value may increasingly accrue to the scarce assets that make intelligence useful: proprietary data, trusted distribution and embedded workflows.

AI can analyze information and offer insights, but it can’t manufacture decades of proprietary data. AI can improve a workflow without making deeply embedded systems easy to replace. And AI can generate answers without creating the trust, regulatory standing or distribution required to act on them. As intelligence itself becomes more abundant, these harder-to-replicate assets become more valuable.

Higher rates reinforce the same distinction. Markets become less forgiving of distant cash flows and more focused on businesses with durable competitive advantages and cash generation today.

Why This Keeps Us Constructive on US Equities

Market concentration is real and valuations are elevated in select areas of the market. But neither tells us where the next dollar of earnings growth will come from.

We think that investment opportunity is evolving to a broader set of beneficiaries. AI investment is creating demand well beyond semiconductors and hyperscalers, and AI adoption is beginning to change the economics of companies across industries.

The first phase of the AI trade rewarded the builders—the chips, hyperscalers and physical infrastructure making intelligence possible. We think the next phase runs through the companies putting that intelligence to work: industrials, financial services, health care and consumer businesses already changing their pricing, staffing and cost structures, often without selling a single AI product.

As intelligence becomes cheaper and more widely available, we think the greatest value accrues to scarce assets —proprietary data, trusted distribution and embedded workflows. That’s a much broader opportunity set than semiconductors and hyperscalers alone—and one US companies are particularly well positioned to capture.