The "AI Bubble" Debate Isn't Scaring Serious AI Investors. Here's the Stock They're Backing.

Applied Digital (NASDAQ: APLD) has rallied ~208% over the past year and ~1,430% over three years as investors pursue AI data-center exposure; the company finalized a second lease with CoreWeave to supply a total of 400 MW at its Polaris Forge 1 campus and cites expected contracted lease revenue of roughly $11 billion, while announcing a loan facility with Macquarie and reporting advanced negotiations with another investment-grade hyperscaler. The stock trades at about 35x trailing sales and the company remains unprofitable, positioning APLD as a high-growth but higher-risk speculative play for allocators seeking AI infrastructure exposure.

Analysis

Market structure: Hyperscalers (NVIDIA-driven AI cloud customers, large cloud providers and specialist lessees like CoreWeave) are clear winners as they internalize AI workloads; Applied Digital (APLD) and a small set of pure-play HPC data‑center developers capture outsized revenue per MW (APLD: 400MW / ~$11bn contracted). Incumbent colos and legacy REITs face pricing pressure for high-density racks but will defend share via scale and power contracts, keeping bidding for hyperscalers competitive and capex‑intensive.

Risk assessment: Key tail risks are a rapid demand repricing (AI compute growth <50% CAGR vs. consensus), a 100–300bps sustained rise in real yields increasing financing costs for developers, or regional power/permits curtailing builds (6–18 months). Short term (days–months) momentum can detach price from fundamentals; medium/long term (12–36 months) profitability depends on utilization rates, contract tenor and tech obsolescence of GPU generations.

Trade implications: Tactical exposure should favor profitable, high‑ROIC beneficiaries (NVDA) and selectively sized, hedged positions in speculative infra (APLD). Options are efficient: defined‑risk debit call spreads on NVDA (12–24 month LEAPs) and put spreads on unprofitable infra names to limit tail loss. Monitor credit markets and 10‑yr yield moves >50bps as a re‑pricing trigger.

Contrarian angles: Consensus underestimates near‑term supply constraints (permitting, power hookups) that could sustain pricing for specialized cages; conversely it also underestimates competition and rapid tech refresh that can shorten lease lives and asset value. Expect episodic volatility: if APLD fails to convert an announced ‘advanced’ hyperscaler within 90 days, re‑rating risk is material and presents shortable inflection points.

Key Decisions for Investors

• Establish a size‑constrained long in APLD equal to 1–1.5% of portfolio now for exposure to contracted lease growth, but hedge with a 6‑month put (cost‑capped) or buy a 6–9 month call spread (defined risk). Trim half of this position on a +30% move and exit if price falls −25%.
• Overweight NVDA by 2–3% (or buy 12–24 month LEAP call spread sized to 2–3% risk) to capture durable AI compute demand; take profits incrementally at +40% and add on any pullbacks >20%.
• Implement a pair trade if seeking relative safety: long NVDA (2%) vs short APLD (1%) sized to neutral beta—target P/L capture from rotation into profitable semiconductors over 3–12 months; close if APLD announces >$3bn new contracted revenue within 90 days.
• Reduce aggregate exposure to unprofitable pure‑play data‑center developers by ~30% across the book and redeploy to semiconductor/cloud names; buy 3–6 month put spreads on any remaining large APLD position if 10‑yr Treasury rises >50bps from current levels (protects vs financing shock).