DeepSeek's decision to raise outside capital is not just a funding story. It is a signal that the self-sufficiency model the thing that made the lab genuinely singular has reached its limits.

Founded in 2023 by Liang Wenfeng, the founder of Chinese quant hedge fund High-Flyer Capital Management, DeepSeek operated for two years as one of the only serious AI labs in the world with no venture capital and no corporate backers. High-Flyer, which reportedly posted a 56.6% return in 2025, funded the research entirely. Approaches from prominent Chinese VC firms, major tech companies, and state-linked funds were all declined. The independence was deliberate and, at the time, almost missionary in its conviction: that good AI research should be free from the commercial pressures that come with outside money.

What ended that era was the R1 model, released in January 2025.

R1 was trained using a mixture-of-experts architecture and reinforcement learning rather than the more expensive supervised fine-tuning used by Western frontier labs and typically activates only 37 billion of its 671 billion parameters during inference. The reported training cost was approximately $6 million. The result matched OpenAI's best models on most benchmarks. Running R1 costs around $0.55 per million input tokens roughly 96% less than OpenAI's equivalent. The models were released open-weight under the MIT licence, available to anyone.

The market reaction was seismic. Nvidia's stock fell sharply in the days following R1's release as analysts reassessed assumptions about AI infrastructure spend. Silicon Valley spent weeks interrogating whether the hardware moat they had assumed was structural had been quietly neutralised by a lab with fewer chips and a different training philosophy. As of March 2026, Stanford University's annual AI Index report showed the performance gap between the top US model and its best Chinese competitor had narrowed to just 2.7 percentage points.

The decision to raise now reflects the reality that maintaining that position costs more than it used to. Reasoning models, agent systems, and the next generation of infrastructure require capital at a scale that even a profitable hedge fund cannot comfortably sustain indefinitely. The initial reported target was at least $300 million at a $10 billion valuation, a figure that placed DeepSeek among China's most valuable AI startups, though a tiny fraction of its US peers' marks. That valuation has since reportedly climbed toward $45 billion as investor interest intensified, according to reporting from the Financial Times and Bloomberg in May 2026.

What the round cannot buy is the thing that made DeepSeek's first two years remarkable: the freedom to research without commercial pressure. How that culture survives outside money and whether it produces the next R1 is the question that matters more than the valuation.