The AI trade in 2025–2026 was not a monolith — it was three distinct bets that performed very differently. Infrastructure companies (semiconductors, cloud providers, data center REITs) delivered strong returns as AI spending scaled. Foundation model labs remained mostly private. And many AI application-layer SaaS companies that IPO'd in 2023–2024 at peak valuations have spent the last 18 months correcting. Getting the layer right matters more than getting the sector right.
Here is the honest framework for AI investing in 2026.
The three layers of AI investing
Understanding AI investing starts with knowing which part of the stack you're buying:
Layer 1 — Infrastructure. Chips (Nvidia, AMD, Broadcom), hyperscalers (Microsoft Azure, Google Cloud, AWS), data center REITs (Equinix, Iron Mountain), networking (Arista). These companies sell the shovels. Revenue is contracted; the AI buildout is still in early innings.
Layer 2 — Foundation models. OpenAI, Anthropic, Mistral — mostly private. Hard to access as a retail investor outside of Microsoft (OpenAI investor) or Google (Gemini). Indirect exposure only for most investors.
Layer 3 — Applications. AI-native SaaS, vertical AI tools, wrapper products. High growth potential, high valuation risk, high competitive pressure from the model labs moving into their territory.
Why diversification matters across layers
The application layer has two problems in 2026. First, many "AI companies" are building on top of APIs from OpenAI or Anthropic — if those providers change pricing or release a competing feature, the moat evaporates. Second, the valuation hangover from 2023–2024 is real; companies priced at 30× revenue on growth expectations that haven't materialized are still working through that.
Infrastructure doesn't have these problems. Demand for compute, power, and networking is real and contracted. That's where 2025's returns came from.
ETF options for diversified AI exposure
- BOTZ (Global Robotics & AI ETF) — broad exposure including robotics; holds Nvidia, Intuitive Surgical, and international AI companies. Low concentration in pure software.
- AIQ (Global X AI & Technology ETF) — wider AI theme including cloud and data. More diversified than BOTZ.
- QQQ (Invesco Nasdaq-100) — not an "AI ETF" but Nvidia, Microsoft, Alphabet, and Meta make up ~35% of the index. Cheapest way to get concentrated tech + AI exposure.
- ROBO (ROBO Global Robotics & Automation) — equal-weighted across 80+ companies; reduces single-name concentration risk.
Comparison: AI investment approaches in 2026
| Approach |
Concentration risk |
Expense ratio |
Upside |
Volatility |
Best for |
| Direct AI stocks (e.g. Nvidia) |
Very high |
0% |
Very high |
Very high |
Conviction investors, high risk tolerance |
| AI ETFs (BOTZ, AIQ) |
Medium |
0.47–0.68% |
High |
High |
Thematic exposure without stock-picking |
| Broad tech ETF (QQQ) |
Low-medium |
0.20% |
Moderate-high |
Moderate-high |
Simple, cheap, indirect AI exposure |
| S&P 500 index fund |
Low |
0.03% |
Moderate |
Moderate |
Core holding, AI as part of broader economy |
Common mistakes to avoid
Chasing the AI narrative at the application layer. The story is compelling; the valuations often aren't. Check price-to-sales and compare to growth rates.
Treating "AI" as a moat. Most AI features will commoditize. The moat is data, distribution, or switching costs — ask which of those the company actually has.
Over-weighting a single AI name. Nvidia has been a 2024–2025 standout, but a 20%+ allocation to any single stock dramatically increases portfolio volatility without proportional expected return.
Ignoring the indirect plays. Energy infrastructure, cooling systems, and power utilities are legitimate AI beneficiaries that often trade at lower multiples than direct AI companies.
FAQ
Is it too late to invest in AI stocks in 2026?
Infrastructure probably still has a multi-year runway — the buildout is early. Application-layer valuations are more mixed; selectivity matters more than timing.
Should I buy individual AI stocks or ETFs?
Unless you have deep conviction in specific companies and understand their competitive position, ETFs reduce the risk of picking the wrong layer at the wrong time.
What percentage of my portfolio should be in AI?
If you hold a broad index fund, you already have ~10–15% AI exposure through Nvidia, Microsoft, and Alphabet. Dedicated AI allocation on top of that should match your risk tolerance — typically 5–15% of total equity for most investors.
Where to go next
For more investing guidance see best ETFs for beginners in 2026, how to invest during a recession in 2026, and recession-proof portfolio 2026.