SoftBank’s Son ‘was crying’ about the firm's need to sell its Nvidia stake

SoftBank sold its entire Nvidia stake for $5.83 billion to fund larger AI investments, with founder Masayoshi Son saying the divestment was painful but necessary to bankroll bets including OpenAI and Stargate data centers. The group has stepped up AI activity — acquiring Ampere Computing and signaling potential additional OpenAI funding — while reporting Q2 net profit more than doubled to 2.5 trillion yen ($16.6 billion) largely from OpenAI valuation gains. Son dismissed AI bubble concerns and forecasted long-term AI contributions of at least 10% of global GDP, underscoring a strategic pivot toward aggressive AI deployment and private-market commitments.

Analysis

Market structure: SoftBank’s liquidation of a $5.8B NVDA stake to fund private AI (OpenAI, Ampere, data centers) shifts capital from a public mega-cap to private/infra bets. Near-term this can create incremental supply pressure on NVDA shares and lift demand for data-center real estate and cloud/AI infrastructure suppliers (EQIX, DLR, AMZN, MSFT), concentrating value capture downstream where margins and lock-in can be higher.

Risk assessment: Tail risks include an AI valuation reversal (30-50% drawdown scenario for frothy AI names), regulatory limits on AI exports/usage, or a SoftBank liquidity shock if private investments mark down; these are low probability but high impact over 6–24 months. Short-term (days–weeks) expect elevated implied volatility around NVDA earnings and any OpenAI funding news; long-term (quarters–years) this is a structural bet on AI-driven capex replacing some cloud software spend.

Trade implications: Favor semicap suppliers (ASML, LRCX, AMAT) and data-center owners (EQIX, DLR) for 6–18 month appreciation; use options to buy convexity around NVDA earnings (calendar spreads or 6–12 month LEAP calls to limit premium). For relative-risk, play NVDA vs. peers (long NVDA, short AMD/INTC) to isolate data-center GPU adoption vs. general CPU/GPU cycle risk.

Contrarian angles: Consensus treats SoftBank’s move as capitulation; instead view it as deliberate reallocation into private, illiquid AI assets — potential mispricing if OpenAI private rounds reprice higher and public peers lag. Conversely, if SoftBank overpays for private stakes, expect a multi-quarter write-down that could depress sentiment across AI equities; position sizing and option hedges should assume a 25–40% downside stress test.

Key Decisions for Investors

• Establish a staggered 2–3% portfolio long in NVDA over 4 weeks (buy 25% tranches), target +30% in 6–12 months, hard stop-loss at -12% from entry; alternatively purchase NVDA Jan 2026 LEAP calls for similar exposure if wanting fixed downside.
• Overweight data-center real estate and cloud operators: add 1.5–2% each to EQIX and DLR (or 2% to AMZN/MSFT if preferring capex beneficiaries), hold 6–18 months and trim if occupancy/revenue per kW fail to improve within 9 months or if shares rise >30%.
• Deploy a relative-value pair: go long NVDA (1%) and short AMD (1%) to isolate NVIDIA’s AI GPU premium; rebalance after NVDA earnings or if the spread tightens >20% intraday over 4 weeks.
• Buy downside insurance for AI-bubble risk: purchase 3–6 month 10% OTM puts on SOXX sized to cover 3–4% portfolio downside (or buy a put spread to limit premium); if OpenAI announces >$10B new funding from SoftBank within 90 days, reduce put hedge by 50%.