AI is probably the most important technology transition since the internet. That much seems defensible. What's less defensible is the idea that every company with "AI" in its press releases deserves a premium multiple, or that the infrastructure buildout is guaranteed to translate into returns for investors who bought at 2025–26 prices.
The distinction worth making: real versus priced.
What's genuinely real in the AI trade
The infrastructure spending is concrete and measurable. Microsoft, Google, Amazon, and Meta have committed to hundreds of billions in capital expenditure on data centers, chips, and networking. Nvidia's data center revenue grew from roughly $15 billion in fiscal 2023 to over $90 billion in fiscal 2025. Those are real revenues from real customers paying real money — the earnings reports validate the buildout quarter after quarter.
Cloud providers are seeing accelerating demand for AI inference compute. Enterprise software companies are embedding AI features that are genuinely improving productivity metrics. The demand signal from hyperscalers has been large and consistent.
What's hype
Dozens of smaller companies appending AI to their marketing without meaningful AI revenue. Software companies claiming AI-driven growth when the actual numbers don't show it in revenue. Startups raising massive rounds on AI narratives without product-market fit. And the general assumption that the current infrastructure capex cycle translates cleanly into durable earnings for every company in the supply chain.
Cyclicality is the risk nobody in the AI bull narrative addresses directly. Every major technology buildout in history — telecom in the 1990s, data centers in the early 2000s — involved genuine underlying growth and genuine overcapacity simultaneously. The infrastructure gets built. Demand catches up eventually. But companies that built too aggressively or too early get hurt badly in the interim.
The valuation problem
Nvidia's forward P/E at various points in 2025 was above 30x on earnings that themselves assumed continued hyperscaler capex growth. Microsoft and Google trade at above-market multiples on AI growth expectations. At these prices, you're not paying for what the business earned last year. You're paying for a specific future that must materialize on schedule.
That's not necessarily wrong. High-quality compounders deserve premium valuations. But it means the margin of safety is thin — any disappointment in the AI capex cycle, any delay in enterprise adoption, any competitive development that commoditizes AI chips faster than expected, hits these valuations harder than it would hit a cheaper stock. Diversification matters more, not less, when a single sector represents this much implicit portfolio concentration.
The parts of the AI trade that look most defensible
Direct infrastructure picks — companies selling the physical components required regardless of which AI models win: chips, networking, power infrastructure, cooling. These benefit from the buildout itself with less exposure to which software layer ends up dominating. Companies with demonstrated AI revenue growth in their actual earnings, not just commentary on calls about AI potential. Enterprise software companies where AI features are already reducing churn and improving net revenue retention in measurable ways.
What to watch going forward
Hyperscaler capex guidance in quarterly earnings. If Microsoft, Google, Amazon, and Meta start pulling back on data center commitments, the upstream supply chain feels it quickly. Enterprise AI adoption metrics — actual usage, not pilots. Nvidia's competitive moat as AMD, Intel, and custom silicon from hyperscalers try to reduce dependence on a single vendor. Track these via our analyst notes — they tend to be leading indicators before they show up in stock prices.
Frequently asked questions
Are AI stocks a good investment in 2026?
Depends heavily on which ones and at what price. Companies with actual AI revenue growth, strong competitive positions, and reasonable valuations relative to that growth are defensible. Companies riding the narrative without demonstrable revenue uplift are harder to justify. The sector broadly trades at premium valuations that require continued execution.
Is the AI boom a bubble like the dot-com era?
The parallel is worth understanding but also imprecise. The dot-com era involved companies with minimal revenue and speculative business models. The current AI leaders — Nvidia, Microsoft, Google — have massive real revenues and genuine earnings. The bubble risk is more about whether current valuations price in too much of a specific future, not whether the underlying business is real. Different and more subtle risk than 2000.
Should I own AI stocks in my portfolio?
If you own a broad market S&P 500 ETF, you're already significantly exposed to AI — the top AI-related names represent a substantial percentage of major index weights. Additional concentrated AI exposure adds sector risk on top of what you already have. The question is whether that concentration is intentional and sized appropriately.