OpenAI Is Just $200 Billion Away From Still Losing Money, HSBC Says

OpenAI has publicly committed roughly $1.4 trillion over eight years to expand data-center infrastructure, but HSBC’s analysis projects annual data-center bills of about $620 billion and forecasts OpenAI reaching ~3 billion users by 2030 with a 10% conversion to paid customers, yielding roughly $215 billion in annual revenue (vs OpenAI’s ~$200 billion forecast). Even with aggressive growth assumptions, HSBC estimates a cumulative funding shortfall of about $207 billion by 2030, leaving the company dependent on large fundraising rounds or unpopular alternatives (scaling back commitments or seeking government support), which poses material funding and execution risk for investors and counterparties.

Analysis

Market structure: The HSBC numbers (a $207bn cumulative funding gap by 2030 and an alleged $620bn/year data‑center cost run‑rate) crystallize a winner-takes-most outcome: hyperscalers (MSFT, GOOGL, AMZN), GPU suppliers (NVDA, AMD), data‑centre landlords (DLR, EQIX) and power/commodity suppliers (FCX, utilities) capture majority upside while unprofitable AI pure‑plays and late‑stage VC stakes face dilution or failure. Expect pricing power to shift to GPU suppliers and cloud platforms through 2025–2030 while LLM service margins for independent vendors compress on scale and capital access.

Risk assessment: Tail risks include (1) an OpenAI fundraising shortfall triggering cascading contract defaults with providers, (2) GPU supply shock or capex overbuild creating stranded assets, and (3) regulatory/bailout interventions that reshuffle economics. Immediate (days) risk is sentiment-driven repricing of AI names; short‑term (weeks–months) is funding rounds/convertible issuance news; long‑term (years) is capex realization and customer monetization vs. the HSBC revenue trajectory.

Trade implications: Favor hardware/cloud and infra exposure while trimming speculative SaaS. Specifics: overweight NVDA (hardware demand), MSFT/GOOGL/AMZN (cloud capture), DLR/EQIX (data‑centre rents) and underweight C3.ai (AI) and small AI SaaS names. Use 6–12m call spreads on NVDA to gain convexity and buy 12–24m DLR/EQIX positions sized to 1–3% each; size shorts in speculative names small (<=1–2% notional) with tight stops.

Contrarian angles: The market underestimates hyperscalers’ bargaining power: if OpenAI must cut commitments, GPU vendors see near‑term demand shock but cloud/hardware oligopolists gain share long term. Historical parallel: cloud capex cycles (2010s) produced persistent consolidation — expect similar consolidation here. Hedge with cross‑asset protection (long energy/commodities, buy IG tech credit hedges) ahead of major fundraising windows.

Key Decisions for Investors

• Establish a 2–3% portfolio long in NVDA (ticker: NVDA) via a 12‑month 1:2 call spread (buy 1 near‑ATM LEAP, sell 2 higher strikes) to capture hardware demand while limiting premium; target +30–50% in 6–12 months; stop‑loss if NVDA falls 20% from entry.
• Build 3–4% combined long in cloud leaders: 1.5–2% MSFT (Microsoft) and 1.5–2% GOOGL/GOOGL (Alphabet) or AMZN split, adding on dislocations after cloud capex guides; target ~20% upside by Q4 2025; reduce if combined cloud capex guidance misses by >200bps YoY.
• Initiate a small, tactical short (1% notional) in C3.ai (ticker: AI) — pair trade long NVDA short AI — expecting multiple compression and cash‑flow stress; set stop‑loss at +25% adverse move and profit target at 40% decline or EV/sales reversion to <3x.
• Add 2% exposure split to data‑centre REITs Digital Realty (DLR) and Equinix (EQIX) (1% each) for 12–24 months to play structural demand; hedge with 6–12m put protection if 10‑yr UST yield rises >75bps from current levels (funding cost shock scenario).
• Short volatility/play: sell near‑term implied vol on small AI names with elevated IV (selective covered‑call or put‑sell structures) while buying 9–12m protection on the portfolio ahead of major OpenAI fundraising windows; unwind if OpenAI announces >$20bn new funding tranche.