SoftBank Just Wired the Final $22B, Closing Its $40B OpenAI Mega-Bet — Son Pivots From “Vision Fund” to “Vision Wager,” Betting the House on Sam Altman Reaching a Trillion-Dollar IPO Before Anyone Else
TOKYO / SAN FRANCISCO — The largest single cheque ever written to a private company has finally cleared. CNBC confirmed Friday that SoftBank Group transferred the last tranche of more than $22 billion to OpenAI’s Cayman holding vehicle, ending a nine-month capital-raising odyssey that saw Masayoshi Son liquidate iconic positions, freeze new Vision Fund deployments, and remake himself from a portfolio sprawl-artist into an all-in AI maximalist. The aggregate $40 billion commitment — SoftBank’s entire market cap just fifteen years ago — values OpenAI at $260 billion in the February round’s terms, but street chatter ahead of a widely expected 2026 IPO now floats a jaw-dropping $1 trillion pre-money, a four-fold re-rating that would make this the most lucrative moon-shot in venture history.
How do you come up with $40B? Sell the crown jewels, slowly but publicly
- $5.8B: SoftBank’s entire Nvidia stake (accumulated at an average cost < $120/sh) sold through Goldman block-trades in March.
- $4.8B: T-Mobile US holdings unloaded via Morgan Stanley accelerated book-build in May.
- $3.1B: partial exit from Coupang secondaries plus ARM dividend recap (Son still controls ARM, but ARM itself issued debt to upstream cash).
- $15B: four-bank syndicated bridge loan secured against Sprint spectrum licenses and Alibaba call-spread collateral.
- $11.3B: SB Northstar — SoftBank’s internal hedge-fund sleeve — sold short-dated tech index puts, capturing premium as volatility collapsed post-Fed cuts.
The final $22B wire last week retired the bridge and leaves SoftBank with net debt of roughly $25B — high, but still inside the company’s self-imposed 25% gearing covenant. Vision Fund 1 & 2 are officially in “harvest-only” mode; no new buyouts, no follow-ons, no sidecars. Son told investors on the November call: “I am done spreading seeds. We now water one redwood.”
OpenAI’s war-chest and the IPO horserace
The cash lands as OpenAI is simultaneously:
The cash lands as OpenAI is simultaneously:
- negotiating a $4B convertible from Amazon (cloud credits + AWS exclusivity carve-outs),
- finalizing a $1B IP-license-plus-equity deal with Disney (theme-park conversational characters + Marvel IP bots),
- sounding out banks for a 2026 public float that would leap-frog rival Anthropic, also rumored to be eyeing a listing the same year.
SoftBank’s term-sheet gives it a board observer seat, liquidation preference at 1.25×, and most importantly a ratchet: if the IPO price < $300B valuation, SoftBank receives additional shares equal to the shortfall — effectively underwriting a 15% minimum IRR. Analysts at Bernstein estimate OpenAI needs ~$65B in total capital before cash-flow breakeven (projected 2029). With the SoftBank close, cumulative funding rises to $58B; the Amazon round would push the cushion north of $60B, placing the company within striking distance of self-funding its own compute build-out.
Market ripples: who gets squeezed
- Anthropic: still raising at a $60B private valuation, it now faces an opponent with 10× the cash and a partner that owns ARM chips + data-center fiber.
- Google: had floated a $20B convertible in September; term-sheet talks cooled after SoftBank’s exclusivity window with OpenAI widened.
- Nvidia: loses SoftBank as a top-10 shareholder but gains a customer that will likely buy > 3 million H100-equivalents through 2027. Jensen Huang reportedly toasted the deal on an internal call: “More demand, less supply on the float — bullish.”
- Venture capital: Sequoia, Thrive, a16z all have OpenAI stakes, but their pro-rata capacity is dwarfed; the era of VC-led mega-rounds may be over for frontier models.
Three scenarios for valuation D-Day
Bull case ($1T IPO, 2026): GPT-5 launches, consumer subs hit 500M, Azure/AWS bidding war drives effective compute cost to zero, and OpenAI trades at 25× 2028E FCF — think Apple meets Google in 2004. SoftBank’s $40B stake compounds to $160B, eclipsing the entire current SoftBank market cap.
Neutral case ($400B IPO, 2027): Regulatory delays cap upside, revenue growth slows to 40% y/y, public markets assign a Tesla-like 12× sales multiple. SoftBank exits at 3-4× money — still the best trade in Son’s career, but not empire-redefining.
Bear case (<$300B IPO, 2028): A breakthrough open-source model (think China + EU coalition) commoditizes frontier capabilities, ad-spend recedes, and the ratchet triggers, diluting common shareholders. SoftBank gets its money back plus teens IRR; public bag-holders learn what “frontier risk” means.
Bull case ($1T IPO, 2026): GPT-5 launches, consumer subs hit 500M, Azure/AWS bidding war drives effective compute cost to zero, and OpenAI trades at 25× 2028E FCF — think Apple meets Google in 2004. SoftBank’s $40B stake compounds to $160B, eclipsing the entire current SoftBank market cap.
Neutral case ($400B IPO, 2027): Regulatory delays cap upside, revenue growth slows to 40% y/y, public markets assign a Tesla-like 12× sales multiple. SoftBank exits at 3-4× money — still the best trade in Son’s career, but not empire-redefining.
Bear case (<$300B IPO, 2028): A breakthrough open-source model (think China + EU coalition) commoditizes frontier capabilities, ad-spend recedes, and the ratchet triggers, diluting common shareholders. SoftBank gets its money back plus teens IRR; public bag-holders learn what “frontier risk” means.
What it means for the rest of tech
SoftBank’s transformation is complete: from a diversified bettor on the information revolution to a single-asset, single-sector, single-management-team wager. Analysts now value SoftBank on an NAV basis almost entirely by:
SoftBank’s transformation is complete: from a diversified bettor on the information revolution to a single-asset, single-sector, single-management-team wager. Analysts now value SoftBank on an NAV basis almost entirely by:
- 90% implied OpenAI stake
- 8% ARM Holdings
- 2% Alibaba stub and Sprint debt pieces
In practice, Son has built a 1.5× leveraged ETF ticket on Sam Altman’s next five-year plan. When he unveiled the strategy to SoftBank’s board, directors asked why not keep the Nvidia shares instead. Son’s slide was blunt: “Nvidia sells the picks; OpenAI owns the gold mine. I want the mine.”
The clock starts now
OpenAI must prove it can convert capital into compute, compute into intelligence, and intelligence into cash faster than SoftBank’s interest expense compounds. If it does, the $40 billion bet will look like the bargain of the century. If not, Masayoshi Son will have swapped a diversified tech portfolio for the right to say he once bank-rolled the most expensive science experiment in human history. Markets close Friday with SoftBank stock up 11 % on the confirmation—roughly the value of the entire Nvidia stake he just sold. The market’s message is clear: in 2025, AI is the only vision that matters.
OpenAI must prove it can convert capital into compute, compute into intelligence, and intelligence into cash faster than SoftBank’s interest expense compounds. If it does, the $40 billion bet will look like the bargain of the century. If not, Masayoshi Son will have swapped a diversified tech portfolio for the right to say he once bank-rolled the most expensive science experiment in human history. Markets close Friday with SoftBank stock up 11 % on the confirmation—roughly the value of the entire Nvidia stake he just sold. The market’s message is clear: in 2025, AI is the only vision that matters.