DeepSeek closed $7.4 billion. Its founder still won't write down the vision.

DeepSeek is the Hangzhou-based AI lab that rattled US AI pricing in early 2025 with a public release of a competitive reasoning model. The lab just closed its first outside round: roughly RMB 50 billion, about $7.4 billion, at a pre-money valuation above $50 billion, according to a translated transcript of a four-hour investor meeting published by Recode China AI and sourced from Tencent Technology. The meeting produced 118 numbered remarks covering organizational culture, AGI roadmap, compute posture, and the rationale for staying open source.

Now the lab is reportedly in talks for an additional $1.5 billion at a $71–74 billion valuation, TechCrunch reported, with a 2027 onshore IPO already on the calendar. The Manila Times, citing unnamed sources, has corroborated the cap-table numbers. None of this has been confirmed in a DeepSeek filing or official statement.

Against that cap table, founder Liang Wenfeng told the same meeting, according to the Recode China AI translation, that DeepSeek is vision-driven rather than KPI-driven. There is no formal performance review. The vision is not written down. The original intent, he said, was explicitly not commercial. A separate translation by Geopolitechs, the third English-language version of the same meeting material, gives an independent cross-check on the wording.

The natural read is either that Liang was performing humility for investors, or that the absence of a plan is the plan. The transcript supports the second reading, and the reason is mechanical.

Open source is not charity at this scale. It is a commoditization play. If model weights are free, the model layer stops being where value sits. Value moves up the stack to cloud distribution, application integration, and the geopolitics of who runs the underlying compute. A lab whose strategy is to keep pushing the model layer toward zero price benefits from every other layer becoming more expensive. Liang's 'no commercial intent' language, in this reading, is not a description of the company's relationship to money. It is a description of which layer the company plans to make money from. The four-hour meeting did not need a written vision, because the strategy is the open-source stance itself.

The compute piece sharpens the case. The transcript covers DeepSeek's reliance on Huawei chips and the structural gap with US compute access, both of which are constraints rather than choices. Open source reduces the cost of being compute-poor relative to better-funded US labs, because anyone can run the model on whatever silicon is available. A free model that runs on domestic Huawei hardware is more strategically useful inside China than a closed model that needs Nvidia silicon the lab cannot reliably import.

A 2027 IPO will be the public test of whether the open-source strategy actually works as a business model, not just as a posture. The cap table says investors bought the story. The strategy says they bought the absence of a different strategy. If DeepSeek reaches its 2027 listing and the IPO prices inside the reported $71–74 billion range, the strategy holds: open source at this scale is a way to capture value at the layers above the model, and Liang's 'no plan' language is honest about that.

If the listing prices below the range, or if the company has to close weights to hit the number, the strategy fails. The transcript then becomes a record of a founder who told investors what they wanted to hear, in a meeting the cap table paid for. The next data point is the close of the reported $1.5 billion follow-on. That will set the floor for the IPO conversation, and it will tell the market whether the 'no commercial intent' line still travels at $74 billion.