A Bet That Won’t Sit Still
SoftBank’s investment portfolio has never been a quiet place. From WeWork’s spectacular collapse to the rollercoaster performance of its Vision Fund holdings, Masayoshi Son’s firm has built a reputation for swinging big and absorbing the consequences publicly. Now, its growing stake in OpenAI is doing something the rest of the portfolio rarely manages: appreciating steadily, at least on paper, while the broader tech investment landscape stays choppy.
SoftBank’s exposure to OpenAI has grown through a combination of secondary share purchases and its involvement in the $500 billion Stargate infrastructure initiative, a joint venture alongside Oracle and OpenAI aimed at building out AI computing capacity across the United States. That positioning gives SoftBank not just equity upside in OpenAI itself, but infrastructure leverage tied directly to how fast AI deployment scales. For a firm that has been trying to stabilize its narrative after years of headline losses, that is not a minor development.
How the OpenAI Position Got This Large
SoftBank’s path into OpenAI was not a single clean entry. It accumulated exposure gradually, first through Vision Fund 2 activity and then more aggressively in late 2024 and into 2025, when it participated in a funding round that valued OpenAI at $157 billion. The firm reportedly committed around $1.5 billion in that round alone, with signals that further investment through Stargate would extend its total exposure well beyond that figure. By early 2025, SoftBank had positioned itself as one of OpenAI’s largest outside investors.
That escalating commitment is notable because it contrasts sharply with how the broader Vision Fund portfolio has behaved. Many of the fund’s earlier bets – in ride-sharing, real estate tech, and logistics – were made on the assumption that growth-at-all-costs models would eventually convert to profit. Most did not convert quickly enough. OpenAI is a different kind of animal: it is generating real revenue, it has enterprise customers paying for API access and products, and its valuation trajectory has moved in one direction while the Vision Fund’s aggregate performance moved in the other.
The Stargate angle adds complexity to the story. Unlike a passive equity stake, SoftBank’s Stargate involvement is an active capital commitment to physical infrastructure – data centers, power supply agreements, and the buildout of compute capacity that OpenAI and other AI companies will need. That means SoftBank’s OpenAI exposure is partly financial and partly operational, which changes how analysts should think about the risk profile. If AI inference demand grows as projected, Stargate’s infrastructure becomes increasingly valuable. If the AI investment cycle cools, SoftBank carries real-world asset exposure that does not simply reprice on a spreadsheet.
The Portfolio Context Nobody Should Ignore
SoftBank’s overall portfolio remains genuinely volatile. Its stake in Arm Holdings has been a bright spot – Arm’s share price surged after its 2023 IPO and held reasonably well through 2024 as semiconductor demand tied to AI training stayed strong. But Vision Fund 2 still carries a long tail of underperforming assets, and the firm has written down billions across its investment history. The OpenAI position, however paper-strong it looks right now, sits inside a structure that has a documented history of dramatic reversals.
That tension is not abstract. SoftBank has used asset appreciation – including the rising value of its Arm stake – as collateral and financial justification for new investment cycles before. If OpenAI’s private valuation continues to climb, it gives SoftBank another lever to pull when raising future capital or justifying its own stock price to investors. The risk is circular: the valuation of an illiquid private stake influences SoftBank’s own market capitalization, which then influences its capacity to invest further. When that loop works, it looks like genius. When it breaks, the losses compound fast.
What OpenAI’s Valuation Actually Rests On
OpenAI’s current valuation is a function of revenue growth, market position, and the sheer volume of capital flowing into AI as a category. The company reportedly crossed $3.4 billion in annualized revenue in early 2024, with projections for continued acceleration. That growth is real. But the valuation multiple applied to that revenue is only sustainable as long as investors believe OpenAI will maintain its position as the default AI infrastructure layer for enterprises, developers, and consumer applications.
That position is being contested. Anthropic’s enterprise push is pulling at corporate buyers who might otherwise default to OpenAI’s GPT suite. Google’s Gemini integration across Workspace gives it a distribution advantage that OpenAI cannot easily replicate. Meta’s open-source Llama models are eroding the argument that cutting-edge AI requires a proprietary API relationship at all. OpenAI is still leading, but the competitive environment is tightening faster than its valuation growth implies.
For SoftBank, the deeper risk is that it has committed to OpenAI not just as a financial bet but as a strategic identity. Masayoshi Son has spoken about artificial general intelligence in near-theological terms, framing SoftBank’s AI investments as a generational mission rather than a fund cycle. That framing is motivating, but it also removes the psychological off-ramps that disciplined portfolio management usually requires. A firm that has decided it is on the right side of history does not quietly trim positions when sentiment shifts.
The Stargate infrastructure commitment sharpens this dynamic. Once data centers are under construction and power agreements are signed, the flexibility to pivot disappears. SoftBank will not be able to mark those assets to market the way it can adjust a public equity position. If OpenAI’s revenue growth stalls, or if a competing model becomes the enterprise default, SoftBank’s infrastructure bet becomes a liability with a fixed cost structure. The upside is real. So is the floor risk – and right now, the portfolio math depends heavily on which one arrives first.
