SoftBank just cashed out of Nvidia—again. This time, it’s a full exit worth roughly $5.8 billion, with Masayoshi Son redirecting the proceeds into an “all-in” spree on AI infrastructure and OpenAI.
It’s the latest turn in Son’s high‑conviction, high‑volatility investing playbook. The question isn’t whether SoftBank believes AI will reshape the global economy. It’s whether Son can finance that belief without overleveraging the house—and whether this is a prudent rotation or a flashing sign that the AI trade is topping out.
A Big Bet Begets A Bigger Bet
Let’s start with the facts. SoftBank disclosed it sold 32.1 million Nvidia shares in October, raising $5.83 billion. The company also offloaded a chunk of T‑Mobile for another $9.17 billion as it assembles cash to fund tens of billions in AI commitments, notably an estimated $22.5 billion second tranche into OpenAI this quarter as part of a larger, multi-stage investment program. SoftBank’s fiscal Q2 profit more than doubled to 2.5 trillion yen (about $16.2 billion), aided by Vision Fund gains and soaring paper valuations on its AI holdings. Nvidia dipped on the news; SoftBank ADRs popped. These are not small ripples; they’re the wake of a giant turning its ship.
SoftBank’s CFO Yoshimitsu Goto was explicit: the sales are about liquidity, not a loss of faith in Nvidia. It fits a pattern for Son—selling celebrated winners to feed a thesis that’s even bigger. He did it once before with Nvidia in 2019; that decision aged terribly as the stock rocket‑shipped into the AI epoch. Selling now may be smarter—take profits at the peak of a mega‑cycle to fund the infrastructure and model layer you believe will define the next one. Or it may be another cautionary tale in the making. The truth is, we won’t know for years.
The Market Reads: Rotation Or Red Flag?
Markets are reading two ways:
- Rotation: If you think AI value is migrating up the stack—from chips to foundation models, agents, and services—then selling Nvidia to fund OpenAI and data center buildouts looks like playing the next inning. Son’s alignment with OpenAI and reported infrastructure pushes (including the much‑discussed “Stargate” data centers) suggest he wants to own the demand drivers, not just the suppliers.
- Red flag: If you’re convinced we’re in late‑stage exuberance, SoftBank’s move rings like a bell. The Guardian was blunt: the sale “intensified the debate” about an AI bubble, as tech stocks slipped on the headlines. Investors have hunted for a “top” tell—when a savvy operator decides it’s time to take money off the table. Is this it?
Both readings can be true. Son is rotating aggressively precisely because valuations are rich and capital is plentiful. He’s choosing to pour it not into Nvidia’s margins but into OpenAI’s ambitions and a sprawling chessboard of chips, robots, and hyperscale compute ecosystems.
Financing The Future, With Margin For Error
The math is nontrivial. Beyond the $5.8B from Nvidia and $9.17B from T‑Mobile, SoftBank has leaned on bond issuance and loans against Arm shares to widen its war chest. Analysts estimate tens of billions are earmarked for OpenAI alongside other initiatives such as the planned $6.5B acquisition of Arm‑aligned chipmaker Ampere. The company also announced a 4‑for‑1 stock split effective January 1, 2026—useful for retail access and, often, sentiment.
SoftBank’s thesis is that the risk of underinvesting in AI is greater than the risk of overinvesting. That’s not crazy. U.S. tech behemoths are committed to trillion‑dollar capex trajectories for AI‑ready compute. The bottleneck has been, paradoxically, both chips and credible model‑driven applications that justify all that steel and silicon. SoftBank is trying to sit at the central junction: a top equity holder in OpenAI, a patron of next‑gen chip startups, and a financier (and organizer) of gargantuan AI data centers.
But let’s be clear: leverage stacks quickly. Margin loans against Arm introduce reflexivity; market drawdowns can tighten the very financing that underwrites the strategy. When a trade depends on AI multiples staying buoyant while you sprint to revenue scale, execution risk becomes existential.
Why This Matters Beyond Tokyo And San Jose
There’s a democratic lens here that’s easy to miss. AI infrastructure at nation‑state scale isn’t just a business story. It’s a sovereignty story. Whoever funds and governs the data centers, the model pipelines, and the supply chains will shape not only economic growth but informational power—what gets built, who accesses it, and under what rules.
- Competition and openness: Concentration of model power in a few firms raises antitrust, safety, and access questions. Public interest demands transparency and interop standards so small developers, universities, and startups aren’t locked out of the new computing commons.
- Labor and equity: If AI productivity gains accrue to balance sheets, not workers, we’ll supercharge inequality. Policymakers should tie public incentives—tax credits, land grants, energy subsidies—to worker upskilling, wage growth, and community benefits agreements in data‑center‑heavy regions.
- Resilience: Overreliance on margin financing and a handful of suppliers can turn a tech cycle into a macro shock. Regulators should stress‑test the AI buildout the way they do banks: look through the capital stack, model downside scenarios, and plan backstops for critical infrastructure.
SoftBank’s deals are private, but their effects won’t be. The public will live with the results: in job markets, energy grids, and the information ecosystem. Democratic institutions should get ahead of that curve, not chase it.
The Son Playbook, Revisited
Son’s genius and flaw are the same: he swings for history. Alibaba proved the upside. WeWork proved the downside. Nvidia—twice—adds irony. If you believe OpenAI will capture a disproportionate share of AI’s economics, selling Nvidia to buy more OpenAI looks prescient. If the value accrues elsewhere—specialized chips, open models, vertical apps—this rotation could age poorly.
One practical takeaway for investors: don’t confuse SoftBank with a passive AI index. It’s an actively managed macro‑venture vehicle with real financing risk and timing risk. For the broader economy: we need rules that ensure the AI wealth being engineered by a handful of boardrooms translates into durable, shared prosperity.
In other words: let the visionaries build. But make the contracts democratic.
The Bottom Line
SoftBank’s Nvidia exit isn’t a retreat from AI. It’s a declaration of where, and how, Son wants to win: at the apex of the model stack and the foundation of the compute grid. The move sharpens a debate we should welcome—about capital allocation in the AI era and the public terms under which this future gets financed.
If this is a top, it will be because the story outran the profits. If it isn’t, it will be because the profits finally caught up to the story.