SoftBank Leads OpenAI's $40 Billion Round at a $300 Billion Valuation

On March 31, 2025, OpenAI closed the largest private fundraise in the history of technology: $40 billion at a $300 billion post-money valuation, with SoftBank Group contributing approximately $30 billion of that total. The round eclipsed the prior record by a factor of roughly three and arrived just five months after OpenAI's $6.6 billion Series E in October 2024 — which itself generated wall-to-wall coverage across the financial press and briefly set a record of its own at a $157 billion post-money valuation.

The capital lands as OpenAI fights simultaneously on three fronts: expanding ChatGPT's product surface against Google Gemini and Anthropic's Claude, building the compute infrastructure required to train the next generation of frontier models, and navigating a governance restructuring that converts it from a nonprofit-controlled hybrid into a public benefit corporation. What SoftBank's billions actually buy — and what they do not — is the question every analyst in the room is working through.

The Round: Numbers in Context

SoftBank committed roughly $30 billion of the $40 billion raise, with a syndicate of existing investors filling the remainder. OpenAI CEO Sam Altman confirmed the close via the company's blog and a post on X, framing the capital as necessary to pursue “the most transformative technology in human history.” The round came five months after the October 2024 Series E, which raised $6.6 billion at $157 billion post-money — meaning the company's paper worth nearly doubled in under six months.

At $300 billion, OpenAI briefly became the most valuable private company in the world by most measures. For scale: that valuation sits at rough parity with Netflix's or Nike's public market capitalization at the time of close — two companies that together employ over 60,000 people and generate more than $40 billion in combined annual revenue. OpenAI, by contrast, had roughly 3,500 employees and had never published an audited income statement.

SoftBank's Bet: Why Son, Why Now

Masayoshi Son's involvement is not incidental. SoftBank and OpenAI were already co-anchors of Project Stargate, the $500 billion AI infrastructure initiative announced January 21, 2025, at the White House alongside Oracle. Under that structure, SoftBank committed to providing the financing backbone, OpenAI would supply the model technology, and Oracle would handle data-center construction — starting with a facility in Abilene, Texas, with $100 billion earmarked for the first tranche.

Son has been unusually explicit about his conviction. In presentations circulated in late 2024, he reportedly predicted that artificial superintelligence — which he called “ASI” — would exceed human intelligence by a factor of one thousand within the decade. Whether that framing is visionary or theatrical, it explains SoftBank's willingness to write a $30 billion check to a company whose annual revenue was roughly $3.4 billion at the time of close — roughly an 88× revenue multiple on current ARR.

What the Money Is Actually For

OpenAI's stated uses for the capital break into three categories. The first is compute: frontier model training at the scale OpenAI is contemplating requires clusters of tens of thousands of high-bandwidth GPUs running continuously. The company was reportedly spending hundreds of millions of dollars per month on inference alone for ChatGPT by early 2025, before accounting for training runs for the generation of models succeeding o3.

The second bucket is research headcount. At $300 billion, recruiting has become counterintuitively complicated: top candidates at Google DeepMind and Anthropic hold equity in companies they can independently benchmark against OpenAI on public leaderboards. Fresh capital gives OpenAI room to compete on compensation and to fund speculative long-horizon programs — interpretability, alignment, and the agentic infrastructure behind products like Operator and ChatGPT's deep-research modes.

Third is distribution. OpenAI has moved aggressively into enterprise, voice, and API pricing that undercuts rivals on a per-token basis. The capital provides runway to sustain below-cost pricing while capturing market share in the API and enterprise tier — the same playbook cloud providers ran in the 2010s, funded by patient capital and justified in retrospect only by the scale they eventually achieved.

Revenue Reality Check

OpenAI's ARR was widely reported at approximately $3.4 billion in late 2024, growing at roughly 2× year-over-year from the prior period. At that trajectory, the company would reach roughly $7 billion ARR in 2025 and somewhere between $12 billion and $15 billion by 2026. At $300 billion, that implies a forward revenue multiple of 20–25× on projected 2026 ARR.

That multiple is not absurd by hypergrowth software standards — Snowflake and Datadog have traded above 20× forward revenue at peak. But it demands sustained 2× annual growth, and it assumes gross margins improve materially from where they stood at close: the company was reportedly operating at a negative gross margin on a fully loaded basis in late 2024, because inference costs were outrunning subscription and API revenue at scale.

The Skeptical Case

SoftBank's Vision Fund is perhaps the most famous cautionary tale in growth-stage investing. WeWork cost the fund an estimated $14–16 billion in write-downs across its investment and subsequent $9.5 billion bailout. Katerra, Greensill, and OneWeb are additional high-profile implosions in the same portfolio. Son's conviction track record makes the OpenAI bet a pattern worth flagging, even if OpenAI's fundamentals — actual revenue, a growing user base, broad API adoption — are measurably stronger than WeWork's were at equivalent valuation milestones.

The AGI timeline argument is also structurally unfalsifiable. OpenAI's o3 model scored 87.5 percent on the ARC-AGI benchmark in December 2024 — a genuine leap from prior-generation models — and 25.2 percent on FrontierMath, an independent set of competition-grade mathematics problems evaluated by Epoch AI. These numbers are real. They also measure narrow, specific capabilities. ARC-AGI tests abstract visual pattern completion; FrontierMath is one benchmark constructed by one research group. Neither says anything definitive about general intelligence, and the definition of AGI has reliably shifted to remain just beyond the frontier every time a leading model approaches it.

The competitive landscape had also intensified sharply by the time the round closed. Google DeepMind's Gemini 2.5 Pro was posting top scores on MMLU, HumanEval, and MATH benchmarks by Q1 2025. Anthropic's Claude 3.7 Sonnet introduced extended thinking modes that made it a credible reasoning rival to OpenAI's o-series. xAI's Grok 3, announced February 17, 2025, claimed state-of-the-art results on AIME 2025 and GPQA Diamond. OpenAI's benchmark moat — once near-absolute — was measurably contested on every axis by the time the round was signed.

“The question isn't whether AI will be transformative. It's whether OpenAI specifically captures the value — or whether the returns accrue to whoever has the cheapest compute.” — a framing that circulated widely among AI investors and analysts in early 2025, reflecting a broader uncertainty about whether frontier-model leadership converts to durable margin

The Governance Wrinkle

OpenAI was mid-conversion at close: transitioning its for-profit limited partnership subsidiary into a standalone Delaware public benefit corporation. The change was contentious. Former board members and early investors — including Elon Musk, via ongoing litigation filed in 2024 — argued the conversion undervalued the nonprofit parent's stake and effectively monetized an organization that was structured for public benefit rather than shareholder return.

Altman had historically held no reported equity in OpenAI, which is unusual for a founder-CEO at this valuation. Reports in late 2024 indicated he was negotiating for approximately 7 percent equity as part of the restructuring. At $300 billion, 7 percent would represent roughly $21 billion in paper wealth — a figure that, in critics' view, creates incentive structures at odds with the nonprofit mission the organization was founded to pursue in 2015.

What to Watch

Several signal events will determine whether the SoftBank bet pays off over the coming years: whether OpenAI's next frontier models maintain benchmark leadership against Google and Anthropic; whether Project Stargate data centers come online at the committed pace and produce unit economics that justify the infrastructure spend; whether the nonprofit-to-PBC conversion survives litigation and produces clean governance that institutional investors will accept at IPO; and whether ChatGPT's subscriber base sustains under free-tier pressure from Gemini, Microsoft Copilot, and Claude.

The $40 billion raise buys OpenAI at minimum two to three years of aggressive runway at current burn rates. That is enough time for the competitive landscape to clarify considerably. Whether SoftBank's conviction was visionary or — as with WeWork — a spectacular misread of a narrative-driven market is a verdict that belongs definitively to the far side of this decade.