Rethinking AI’s Investment Beneficiaries — and Why Quality Software Is Positioned to Lead the Next Chapter
Why the Portfolio Underperformed
Between October 2025 and March 2026, the software complex returned −26.2% while the Russell 2500 Index gained 2.9% — a nearly 30-percentage-point relative shortfall in five months, with the index rising throughout. The capital rotation was not ambiguous: capital moved from high-quality, recurring-revenue software businesses into direct beneficiaries of hyperscaler capital expenditure — semiconductors, hardware, data centers, and power infrastructure. Investors sold what they knew to buy what the AI headlines demanded.
Unfortunately, the SMID portfolio’s largest active exposure, information technology at +19.1 percentage points above the benchmark, sat squarely at the center of that rotation. The de-rating was a compression of multiples, not a deterioration of underlying business fundamentals.
Major technology transitions have followed a two-chapter structure. The first chapter rewarded the builders of infrastructure. The second rewarded those who deployed the technology to compound competitive advantage within their own businesses. Heron Bay’s technology stocks have characteristics that will define the maturation of AI adoption. These stocks are cheap, have strong fundamentals, and adoption supporting durable expansion with AI as a tailwind.
Wave 1
The Infrastructure Build-Out — and Its Limitations
The first chapter of the AI investment narrative has been about building. Memory suppliers, chipmakers, semiconductor equipment companies, and data center operators have seen margins and stock prices rise sharply. Semis and energy alone accounted for 86% of the increase in S&P 500 forward profit margins over the past six months. Combined cloud revenues from the major hyperscalers are approaching $100 billion per quarter, growing at nearly 50% year-over-year. The bull case has been numerically supported — for now.
The critical question is durability. History offers a consistent template. During COVID, vaccine manufacturers built booster capacity the market never needed. Amazon overbuilt fulfillment infrastructure on the assumption of permanently elevated e-commerce penetration.
rates, net revenue retention, and seat expansion dynamics. These are not one-time purchases.
- AI as capability multiplier: Contrary to the market’s de-rating thesis, AI tooling embedded in corporate workflows is more likely to expand the volume of work these platforms perform than to substitute for them. Incumbents holding the data, distribution, and switching cost moats are positioned to monetize AI, not be displaced by it.
| Characteristic | Portfolio | Russell 2500 | Spread |
|---|---|---|---|
| Return on Invested Capital | 18.3% | 5.0% | +13.3 pp |
| EBIT Margin | 23.3% | 13.0% | +10.3 pp |
| Free Cash Flow Yield | 8.8% | 5.4% | +3.4 pp |
| Price / Earnings | 15.7x | 19.9x | −4.2x |
| Active Share vs. Russell 2500 | 99.05% | — | — |
Price / Earnings 15.7x 19.9x −4.2x Active Share vs. Russell 2500 99.05% — —
The portfolio earns 3.6 times the index’s return on capital while trading at a 21% discount on earnings and generating a free cash flow yield 63% higher. This quality-valuation gap is not a narrative; it is measured.
The most direct evidence of the market’s capacity to reprice these businesses: the portfolio’s largest holding recovered 61.2% in the second quarter of 2026 off its March lows — same business, largely unchanged fundamentals, dramatically different multiple. The de-rating reflected sentiment, not structural impairment.
The businesses that compound value through the AI cycle will be those whose advantages AI augments rather than threatens. We believe those businesses are already in the portfolio — at a significant discount to the market.
The telecom build-out of the late 1990s, early railroad expansion, and dot-com hardware and fiber spending all followed the same arc: genuine transformative technology, genuine near-term demand, then the recognition that capital deployed exceeded the durable demand that would absorb it.
“Expected long-term earnings growth for the S&P 500 rose to 20.2% — above the 18.6% peak reached during the 2000 technology bubble.” — Ed Yardeni, Yardeni Research
Much of the revenue accruing to the most celebrated AI-adjacent names is transactional — tied to discrete capital projects rather than recurring customer commitments. Rising IPO activity and equity issuance by hyperscalers are consistent with a late-cycle dynamic. The momentum trade driving these valuations has been among the most powerful since World War II; controversy scores for AI plays are approaching historical levels that have previously signaled elevated risk.
Wave 2
The Durable Beneficiaries — Software as a User of AI
The railroads created enormous wealth — but the durable beneficiaries were the retailers, manufacturers, and logistics businesses that used them to build defensible advantages. The Internet’s first chapter enriched hardware and connectivity providers; the enduring winners were businesses that embedded connectivity into operations and created customer relationships that compounded over time. We believe AI follows the same pattern.
The software businesses the market has treated as AI’s primary victims hold, in our view, the attributes most likely to define the second wave:
- Entrenched workflows: Displacing enterprise software requires more than a better model — it requires a complete migration of process, data, and institutional practice built over years.
- Proprietary data advantages: AI is only as powerful as the data it operates on. Domain-specific datasets accumulated over decades of customer relationships cannot be replicated by open-sourced models.
- Recurring, contractual revenue: Unlike the transactional revenue driving Wave 1 earnings growth, enterprise software operates on multi-year contracts with measurable renewal
For more information on Heron Bay’s SMID portfolio, please visit www.heronbaycap.com or email info@heronbaycap.com.