KKR’s Raj Agrawal Warns of Froth in AI Investment Stampede

KKR's head of real assets, Raj Agrawal, warned that excess exuberance has emerged around data centers and artificial intelligence investments, and said the firm is being selective to mitigate risk. He noted that with large amounts of capital moving quickly, some of these investments are unlikely to work out, signaling elevated downside risk for overfunded AI infrastructure plays and encouraging a more defensive allocation stance in private markets.

Analysis

Market structure: Rapid private capital inflows will concentrate downside on third‑party colocation and smaller regional operators, while hyperscalers and vertically integrated cloud providers (scale owners) gain bargaining power and optionality to buy distressed assets. Expect pricing pressure in colocation rents and 10–20% effective yield widening in overbuilt metros within 6–12 months as utilization lags new supply.

Risk assessment: Tail risks include an abrupt GPU/compute mismatch (demand drop vs. supply surge), regulatory limits on cross‑border data/capex (6–18 months), and private‑market markdown cascades forcing fire sales that spill into public REIT credit spreads. Short window (next 30–90 days) for sentiment shocks; structural outcomes play out over 12–36 months depending on macro and capex cadence.

Trade implications: Favor long positions in cash‑rich hyperscalers with disciplined capex (MSFT, GOOGL) and selective semiconductors (NVDA) for exposure to compute demand; short/hedge colocation REITs (DLR, EQIX) or buy downside protection if occupancy falls <5ppt or FFO guidance cuts appear in next two quarters. Use put spreads on DLR/EQIX 3–6 month expiries and call spreads on MSFT/GOOGL 6–12 month expiries to monetize skew.

Contrarian angles: Consensus underprices private‑market illiquidity — public REITs can lead private markdowns by 6–12 months, creating a buy‑the‑dip opportunity in 12–24 months for high‑quality assets; conversely, hyperscalers may accelerate own builds to lock cost base, squeezing mid‑tier providers harder than models project. Historical parallels: 2010s fiber/datacenter cycles saw a multi‑quarter repricing before consolidation; expect similar timeline.

Key Decisions for Investors

• Establish a 2–4% long position in MSFT and GOOGL (split) over next 30 days via 6–12 month call spreads (buy ATM, sell +20–30% strike) to capture secular AI compute demand while limiting capital at risk.
• Initiate a 1–2% short or buy 3–6 month put spreads on DLR and EQIX (each) sized to equal 50–75% of your hyperscaler longs; trim if REITs decline 15–20% or report occupancy drop >5ppt.
• Allocate 1% of portfolio to NVDA call spreads (3–9 month) to play continued GPU tightness, but cap exposure given valuation; take profits if IV declines >30% or inventory indicators show 2 consecutive months of easing.
• Reduce private real‑assets/infra exposure by 25–50% of target over the next 3 months; redeploy proceeds into liquid tech leaders and short protection until private valuations realign or cap‑rate signals (REIT spreads) compress by >50bps.