AI Infrastructure Spending Could Nearly Triple by 2029. Here Are 2 Stocks to Buy.

Statista projects AI infrastructure investment to rise to $902 billion by 2029 from $334 billion in 2025, supporting continued data-center buildouts. Nvidia's data-center business accounted for >90% of revenue and grew 75% YoY; the company reported about $120 billion in net income on $215 billion revenue last year, trades at ~16x next-year EPS with analysts forecasting ~38% annual earnings growth, though historical spending waves have produced >50% drawdowns. Iren has secured >4.5 GW of power, a $9.7 billion contract with Microsoft, and targets $3.4 billion in annualized revenue by end-2026 (≈10% of secured capacity); financing needs and potential dilution are the primary risks against its roughly $13 billion market cap.

Analysis

The market is pricing a concentrated technology stack (accelerators + power + interconnect) as a near‑perfect growth trio; the second‑order consequence is that bargaining power will bifurcate — hyperscalers will pay premiums for guaranteed, turnkey capacity while suppliers that can deliver contractual certainty (capex, interconnect, grid dispatch) will capture outsized spreads. That dynamic favors vertically integrated site developers and turnkey server suppliers able to convert scarcity into multi‑year take‑or‑pay contracts, but it simultaneously creates single‑vendor and single‑route‑to‑market concentration risk that will incentivize customers to invest in optionality (custom ASICs, alternate interconnects).

On the energy and supply‑chain side, a wave of large blocks of demand will shift the bottle­neck from silicon to kilowatts and interconnection timelines; this elevates the value of near‑term interconnection rights, PPAs with short‑delivery windows, and fast‑deployable dispatchable assets (battery+genset). That means monetary value will accrue not just to chips and racks but to firms that can move from permit-to-service in 9–18 months, creating a multi‑year premium for executed sites versus paper capacity.

Key tail risks are cyclical hyperscaler capex pullbacks, rapid customer in‑sourcing of accelerators (reducing third‑party GPU spend), financing/dilution for site builders, and geopolitical/export controls that can compress supply and spike option volatility. Monitor high‑impact, short‑lead catalysts: signed multi‑year purchase orders, PPA pricing moves, utility interconnection clarifications, and quarterly guidance changes — these turn valuation multiples materially in 1–6 months.

Consensus blind spot: investors treat GPU demand and site power as separate bets rather than a bundled contract negotiation where the hyperscalers can arbitrage supplier concentration against financing and delivery risk. My favored posture is asymmetric exposure — buy optional upside to the GPU winner while keeping downside capped on site‑developer exposure until counterparties and financing terms are visible.

Key Decisions for Investors

• Long NVDA via defined‑risk LEAP call spread (buy 2028 Jan calls, sell higher‑strike 2028 Jan calls) sized 1–2% NAV. Rationale: capture multi‑year secular upside while capping premium spend; max loss = premium, upside = 2–4x if NVDA keeps pricing power and hyperscaler orders flow. Timeframe: 18–36 months.
• Relative pair: Long NVDA / Short INTC, equal dollar notional, hold 6–24 months. Rationale: expresses continued accelerator‑led architecture adoption vs legacy CPU share erosion. Risk management: tighten or reduce if NVDA underperforms by 20% absolute or if INTC announces a materially credible alternative ecosystem.
• Selective long IREN equity (2–4% NAV) only after one of two triggers: (a) announced funded take‑or‑pay hyperscaler amendment, or (b) secured non‑dilutive financing close. Hedge: buy 12‑month puts ~15–25% OTM for protection. Target: 2–3x upside over 2–4 years if conversion of secured power accelerates; downside concentrated in financing/dilution (potential −40–60%).
• Tactical long DELL exposure via 6–12 month call spread or 2–3% equity position on a server‑cycle pullback. Rationale: capture OEM ASP and services upside as GPU density lifts server revenue without taking outright single‑chip risk. Timeframe: 6–18 months; expected upside 20–40%, max loss = premium or equity drawdown.