‘I was crying': SoftBank's Son says selling firm's Nvidia stake was painful necessity

SoftBank sold its entire $5.83 billion Nvidia stake in October/November to fund a strategic shift into owning AI platforms, reallocating capital toward an estimated $40 billion push that includes an initial $7.5 billion commitment to OpenAI and a planned $22.5 billion follow-up by year-end alongside large infrastructure projects such as the Stargate data-center initiative. Masayoshi Son framed the disposal as a painful but necessary tradeoff—exiting a high-performing chip holding to secure control stakes in what he expects will be the dominant AI models and ecosystems, a high-conviction, high-risk capital allocation that will test investor appetite as OpenAI valuations and AI spending cycles evolve.

Analysis

Market structure is bifurcating: platform owners (OpenAI, MSFT, GOOGL) and data‑centre/hosting vendors (DLR, EQIX, WDC storage suppliers) are the primary winners as capital shifts from “picks-and-shovels” chips to control of models and distribution. Nvidia (NVDA) remains strategic as a supplier — demand for A100/H100 class GPUs should stay tight, supporting pricing power for at least 12–24 months — but liquidity rotation away from public chip equity reduces marginal buying support at extreme valuations.

Tail risks concentrate around platform regulation, export controls and SoftBank’s execution: a hard regulatory action against AI platforms or tightened US export controls on advanced GPUs could trigger >30% downside in correlated equities within weeks. Immediate (days) reaction risk is low; short term (weeks–months) is dominated by news flow (OpenAI funding tranches, NVDA earnings); long term (3–5 years) outcome depends on monetization of models and data‑centre economics.

Trade implications: favor 6–24 month exposure to cloud/platform owners (MSFT, GOOG) and data‑centre REITs (DLR, EQIX); prefer option structures to cap cost — e.g., buy 9–12 month LEAP call spreads on NVDA (2–3% notional) and long-dated calls on MSFT (1–2%). Hedge with cheap put spreads on NVDA or a small (1%) short of SoftBank (SFTBY) to express execution/valuation risk of large private commitments.

Contrarian view: the market underprices illiquidity and execution risk embedded in mega private deals; owning equity in OpenAI through SoftBank-funded rounds is binary and largely non‑liquid. Historical parallel: Vision Fund write‑downs (2019–2020) show that allocation concentration into private platforms can decimate mark‑to‑market returns if monetization or regulation lags; watch OpenAI valuation milestones (next tranche by Q4 2025) as a binary trigger.

Key Decisions for Investors

• Establish a 2–3% long position in NVDA via 9–12 month call spreads (LEAPs) to capture continued GPU tightness while capping premium; target ONGOING holding through next two NVDA earnings (≈6 months) and reassess vs revenue/guide.
• Allocate 2–4% to platform/cloud exposure: buy MSFT and GOOGL equally (1–2% each) with 12–18 month horizon to benefit from OpenAI monetization and platform capture; add 1–2% long in data‑centre REITs (DLR or EQIX) as compounding beneficiaries of Stargate‑style capex.
• Buy protective 3–6% notional 3–6 month put spreads on NVDA (tight width) or purchase a 1% portfolio hedge in VIX call spread around known catalysts (NVDA earnings, US export‑control announcements) to limit a sudden regulatory/compute‑shock downside.
• Consider a tactical 1% short position in SoftBank (OTC SFTBY) or put options to express downside if OpenAI valuations fail to materialize or force further asset sales; monitor SoftBank liquidity metrics and any equity raises over next 90 days.