OpenAI Closes $10B Series F at $300B Valuation for Compute Buildout
OpenAI has closed a $10 billion Series F round led by SoftBank and sovereign wealth funds, valuing the company at $300 billion. Proceeds are earmarked for datacentre expansion and training the next generation of models.
Original sourceOpenAI has officially closed its Series F funding round at $10 billion, pushing its valuation to $300 billion — a figure that makes it one of the most valuable private companies in history. The round was led by SoftBank, with participation from multiple sovereign wealth funds, continuing a pattern of capital concentration in frontier AI infrastructure that has defined the past two years.
The stated use of proceeds is compute: datacentre construction and the GPU clusters required to train models at the next scale threshold. This is not product development funding in any traditional sense — it is a direct bet that raw training compute remains the primary lever for capability improvements, and that whoever controls the most of it wins the next round of the model race.
The $300 billion valuation is notable context. OpenAI's last publicly reported annualized revenue figure was roughly $3.4 billion, implying a revenue multiple in the high double digits. That multiple only makes sense if you believe the company will capture an outsized share of an AI services market that doesn't fully exist yet, or if the compute infrastructure itself becomes a strategic asset with value independent of the consumer products built on top of it.
The round also signals that the capital requirements for frontier AI have moved well beyond what traditional venture funding can support. Sovereign wealth funds and strategic conglomerates like SoftBank are now the marginal buyers of equity in companies at this scale, which has implications for governance, incentive alignment, and the long-term independence of the organization from geopolitical capital flows.
Panel Takes
The Skeptic
Reality Check
“A $300 billion valuation on roughly $3.4 billion in annualized revenue is not a valuation — it's a prayer. The specific scenario where this math works requires OpenAI to become the default inference layer for a significant fraction of global software, at margins that don't currently exist, before a better-capitalized or better-aligned competitor gets there first. What kills this in 12 months isn't competition — it's the possibility that scale stops being the primary driver of capability gains, which would make this $10 billion of compute spend a sunk cost with no corresponding moat.”
The Futurist
Big Picture
“The falsifiable thesis here is: compute scarcity at the frontier persists long enough for whoever holds the most of it to lock in durable advantage before algorithmic efficiency equalizes the field. The dependency that has to hold is that training compute, not inference optimization or data quality, remains the binding constraint on capability for the next 18-24 months. The second-order effect nobody is talking about is governance: sovereign wealth fund capital at this scale creates alignment pressure that is categorically different from VC pressure, and it will eventually show up in what products get built and for whom.”
The Founder
Business & Market
“The moat being purchased here is not product — it's capital exclusion. By burning $10 billion on compute before competitors can raise equivalent rounds, OpenAI is trying to price out the next tier of challengers before they reach parity. That's a real strategy, but it only works if the spend translates into a model capability gap that customers pay a premium for, and right now the gap between frontier models on most production tasks is narrowing faster than the fundraising cadence. The business question I'd want answered: what is the gross margin on API revenue at current compute costs, and how does this round change that number?”
The PM
Product Strategy
“This is infrastructure investment dressed up as a funding announcement, and the product strategy question it raises is real: OpenAI is simultaneously a model lab, an API platform, a consumer app, and now a datacentre operator, and none of those jobs-to-be-done have the same buyer or the same success metric. The compute buildout makes sense if the thesis is that internal capacity gives them a training and inference cost advantage that flows through to product margins — but that only matters if the product layer is cohesive enough to capture the value, which right now it isn't obviously is.”