The AI Trade Grows Up: Infrastructure Still Leads, but the Proof-of-Profit Era Has Begun

The debate over whether artificial intelligence demand is real has ended. The harder argument, as Globalt Investments¹ portfolio analyst Kimberly Woody frames it in her September 2026 research piece "Positioning for the AI Board Game," is now about durability: which companies can actually convert that demand into lasting earnings and free cash flow.

The answer, Woody argues, still begins with semiconductors. But it does not end there.

Hardware Leads Because It Gets Paid First

The AI buildout's most reliable beneficiaries remain the companies supplying the physical infrastructure. Processors, high-bandwidth memory, networking equipment, advanced packaging, power systems, and cooling have all benefited from a spending cycle that shows no sign of slowing. The connection is straightforward: before an AI system can produce commercial results, the hardware has to exist. Woody puts it plainly: "Semiconductors remain the clearest near-term beneficiaries because they are paid as capacity is built."

Persistent shortages across memory, networking, packaging, and power have kept demand visible and, in several areas, supported pricing. The result is an earnings picture that separates semiconductor companies from the rest of the technology sector. "Earnings estimates for semiconductor companies have generally moved higher while software estimates have been more mixed," Woody notes. Hyperscalers continue investing heavily. Hardware suppliers get paid during the buildout. Software vendors, by contrast, often have to wait.

Software Must Show Its Work

That wait has become a central tension in the market. AI tools are moving out of pilot programs and into commercial deployment, but the benefits are not being distributed evenly. Woody identifies a clear fault line: companies with proprietary data, control over critical workflows, or strong positions in security have a credible path to defending the value they create. Others face a less comfortable set of outcomes. "Some may find that automation reduces seat counts, weakens pricing, or requires heavy investment before new revenue shows up," she writes.

Earnings seasons are now doing the sorting. The market is rewarding specific evidence: larger contracts, growing backlogs, real production deployments, higher usage, better margins. "A good story is no longer enough," Woody observes. The stronger software companies are using AI to make their products more useful and their customer relationships stickier. Weaker businesses remain exposed to competitive pressure on both pricing and market position.

The Profit Distribution Problem

The most provocative argument in Woody's analysis concerns where the gains ultimately land. "The biggest risk may not be that AI fails," she writes. "It may be that AI works, but the profits land in fewer places than investors expect."

Chipmakers, hyperscalers, software vendors, and their enterprise customers all expect returns. But competition and declining computing costs will ultimately determine who keeps the economics. That framing puts a premium on exactly the financial characteristics that tend to get overshadowed in high-growth narratives: free cash flow, capital intensity, utilization, pricing power, and valuation discipline. "A company can post strong results and still disappoint if expectations have run too far ahead," Woody notes. "It is not enough to own a business simply because it has AI exposure."

Where the Opportunity Sits

Portfolio construction, in Woody's view, requires focusing on visible demand, sound economics, and earnings that justify the narrative. Within semiconductors, that means businesses tied to compute, memory, networking, advanced packaging, and power. The trade-off is acknowledged: expectations are already elevated, and supply constraints will not persist indefinitely.

In software, broad sector exposure is a weaker strategy than selective positioning. The qualities that matter are proprietary data, workflow control, security franchises, switching costs, and demonstrated AI revenue.

Woody also points beyond the familiar large-cap technology names. AI data centers require substantially more electricity, transmission capacity, cooling, and specialized equipment. "Companies that enable that expansion could benefit from the same spending cycle without depending on a handful of mega-cap stocks," she argues. The adjacencies, in other words, may carry better risk-adjusted characteristics than the headline names.

The Bottom Line

Woody's conclusion is precise: "The AI trade is entering a more demanding stage." Semiconductor companies retain their advantage because revenue connects directly to the buildout and tight capacity persists. Software can produce meaningful winners, but the burden of proof is higher. The question that now organizes investment thinking is not whether AI will matter. It is who will make money from it, and whether today's share prices already assume too much.

5 Key Takeaways for Advisors and Investors

1. Hardware before software. Semiconductor companies still earn revenue directly from the AI buildout, while software vendors must wait for customers to validate returns. The earnings momentum favors hardware for now.
2. Software selection is everything. Broad software exposure is not a strategy. Proprietary data, workflow control, security positioning, and verifiable AI revenue are the differentiators. Generic AI exposure without these does not constitute a defensible investment thesis.
3. Watch the profit distribution, not just the spending. AI investment is vast, but competition and falling computing costs will compress margins across the value chain. Free cash flow and pricing power matter more than growth narratives.
4. Valuation discipline has returned. Strong AI credentials are no longer sufficient to justify any price. Companies can beat estimates and still disappoint if expectations have outrun fundamentals.
5. The adjacencies deserve attention. Power, cooling, transmission infrastructure, and specialized equipment are integral to the AI buildout and may offer compelling risk-adjusted returns without the concentration risk of mega-cap technology holdings.

Footnote:

Footnote:

Woody, Kimberly. "Positioning for the AI Board Game." Globalt Investments, 2 Sept. 2026, https://static1.squarespace.com/static/6691650f7cf3cc0e81616f61/t/6aa075bab9cd5b76d50acd0d/1794507946625/090226-Positioning+for+the+AI+Board+Game.pdf.

2 Woody, Kimberly. "Positioning for the AI Board Game." Globalt Investments, 2 Sept. 2026, https://www.globalt.com/news-insights.