Will AI ever make big profits? Experts weigh in as bubble fears loom

AI-related investment has become a material driver of the U.S. economy — JPMorgan Asset Management estimates AI spending accounted for roughly two-thirds of GDP growth in H1 2025 — yet questions about sustainable profits persist. Notable datapoints include Nvidia’s outsized semiconductor profits, OpenAI’s reported pace toward roughly $13 billion in 2025 revenue (about $3.25B/quarter) and ~800 million weekly active users for ChatGPT, while an MIT study found ~95% of businesses investing in AI have not monetized those investments (combined spend ≈ $40B). Analysts warn that heavy infrastructure costs and trillions in data‑center investment create pressure for rapid returns and raise bubble/recession risks if spending reverses.

Analysis

Market structure: The immediate winners are AI-infrastructure suppliers (NVIDIA, GPU foundries, data-center owners) who enjoy pricing power from constrained H100-class capacity; end-user monetizers (consumer apps, many enterprise AI vendors) are losers for now as CPU/GPU-driven OPEX blunts unit economics. Expect elevated GPU rents and spot premiums for 6–18 months, then potential margin pressure as fabs and cloud scale raise supply by 2026–2027.

Risks: Tail risks include a sharp regulatory clampdown (export controls, EU/US AI safety rules) or a demand collapse that forces large write-downs in data-center capex — both could shave 30–50% off exposed equities in 3–12 months. Near term (days–weeks) sentiment/volatility plays dominate; medium (3–12 months) earnings/guidance cycles matter; long term (2–5 years) depends on productivity adoption and energy/capacity constraints.

Trade implications: Favor concentrated, convex exposure to GPU scarcity (defined-risk long NVDA options) and select Meta (META) ad/AI monetization optionality, while underweight/short high-multiple unprofitable AI software names and margin-exposed cloud names (GOOGL). Cross-asset: higher data-center demand supports copper and power prices; equity drawdowns would push Treasury yields lower — keep cash/credit hedges.

Contrarian view: Consensus underestimates two things — (1) near-term pricing power for top GPU suppliers can sustain outsized margins into 2026, and (2) monetization of ubiquitous assistants (ads/subscriptions) can scale faster than current revenue traction implies. The market may be over-discounting long-term value while under-discounting short-term scarcity, creating tactical dispersion opportunities.

Key Decisions for Investors

• Establish a 2–3% portfolio long in NVDA via a 9–15 month call spread (buy delta ~0.45, sell higher strike to fund), target 30–60% upside over 12 months; trim if NVDA falls >20% from entry or rally exceeds 80% from entry.
• Enter a relative-value pair: long META 2% vs short GOOGL 2% for 6–9 months — thesis: META has faster consumer AI monetization (ads/subs) while GOOGL faces margin pressure from cloud capex; exit if META misses revenue by >4% or GOOGL posts >5% upside guidance.
• Reduce exposure to small/mid-cap unprofitable AI/SaaS names by 50% and allocate 1–2% to put spreads on a basket of these names (3–6 month expiries) to capture downside if AI spending retrenches; size to limit max loss to premium paid.
• Allocate 1.5–2% to interest-rate tail hedge (long 7–10 year Treasury ETF/futures) to protect against an AI-capex reversal causing a growth shock; unwind if 10y yield rises >50 bps from entry or equity drawdown does not materialize within 6 months.