OpenAI is digging a deeper money pit than anyone guessed

HSBC updated its OpenAI model to include new cloud deals — a $250bn arrangement with Microsoft and $38bn with Amazon — bringing contracted compute to 36 GW and cumulative cloud agreement value to as much as $1.8tn (annual data‑centre rental ~ $620bn). HSBC projects 3 billion users by 2030 (10% paid conversion by 2030 vs ~5% today), forecasts consumer AI revenue of $129bn (search $87bn, advertising $24bn) and enterprise AI revenue of $386bn by 2030, but also estimates rental costs of $792bn to 2030 ($1.4tn to 2033). After modeled cumulative free cash flow of ~$282bn to 2030 and potential injections (~$26bn from Nvidia/AMD moves, $24bn facilities, $17.5bn liquidity), HSBC finds a ~$207bn funding shortfall (plus a $10bn buffer), warning OpenAI may need partner concessions or to renegotiate data‑centre commitments absent additional financing.

Analysis

Market structure: The immediate winners are chipmakers (NVDA) and hyperscalers (MSFT, AMZN) that capture datacentre lease revenue and GPU demand; landlords, utilities and industrial metals suppliers also stand to gain from multi-GW capacity growth. Losers include smaller cloud players, legacy enterprise infra vendors (ORCL) and any AI lab that cannot finance long-term power contracts — HSBC’s $207bn shortfall through 2030 implies material counterparty renegotiation risk. The pricing power tilt is toward GPU suppliers and cloud operators, but margin capture will be contested as OpenAI subsidises users and searches monetize slowly (HSBC: consumer conversion 5%→10% by 2030).

Risk assessment: Tail risks include an AGI breakthrough (upside) or a liquidity shock where Microsoft/SoftBank pull back (downside) — either could move markets 20–50% for incumbents in weeks. Immediate (days) risks: sentiment-driven moves around NVDA/MSFT prints and any OpenAI funding announcements; short-term (months) risks: contract renegotiations and pricing pressure on compute; long-term (years) risks: structural overcapacity or regulatory clampdowns on monetization. Hidden dependencies: power grid bottlenecks, regional energy price spikes, and cross-holdings (AMD stock sales) that could cascade into equity volatility. Key catalysts: Nvidia supply cadence (next 3–9 months), MSFT funding commits, and any public disclosure of OpenAI capex re-pricing.

Trade implications: Favor concentrated, hedged exposure to NVDA and selective cloud exposure (MSFT > AMZN) while using options to cap downside; buy industrial metals/power exposure for 6–24 month cyclic upside. Use relative trades to express conviction (long GOOGL vs short ORCL) because HSBC’s model underweights Google’s ad/search franchise and Oracle’s cloud relevance is fading. Avoid unhedged long positions in pure-play AI labs and keep 3–5% liquidity to exploit forced asset sales.

Contrarian angles: Consensus underestimates Google’s default-monetization moat and overestimates OpenAI’s unassisted ability to fund 36GW — a modest contraction (25% of capacity) would materially re-rate datacentre suppliers and relieve power-driven commodity inflation. Historical parallel: 2010s cloud hyperscale buildouts initially tightened GPU/energy markets then normalized over 24–36 months; expect mean reversion in compute pricing, not perpetual margin expansion for AI labs. Watch for renegotiation events as tactical entry points.

Key Decisions for Investors

• Establish a 2–3% portfolio long in NVDA using a 12-month call spread (buy ATM call, sell ~50% OTM call) to capture GPU scarcity; target +40% upside in 12 months, implement a -20% stop-loss on delta-equivalent stock exposure.
• Initiate a relative-value pair: long GOOGL 1.5% vs short ORCL 1.5% (equal notional). Rationale: GOOGL’s ad/search moat is under-forecast; ORCL faces secular cloud share erosion. Target 15% relative outperformance in 6–12 months, exit if spread widens adverse by 10%.
• Add 1% exposure to copper via FCX or short-dated copper futures (6–24 month horizon) to play higher datacentre power/infrastructure demand; increase to 3% if contracted AI capacity grows by another +4GW or regional power forwards rise >10%.
• Protect tech/AI exposure against funding shocks: allocate 1% to MSFT equity and sell 9–12 month covered calls at ~15% OTM to reduce cost; if MSFT publicly reduces AI funding or AA credit spreads widen >50bps within 30 days, reduce net tech beta by 50% within one trading week.