DeepSeek raises $7.4 billion as Liang reveals AGI strategy in 118-point Q&A

DeepSeek raised 50 billion yuan ($7.4 billion) in its first external funding round at a pre-money valuation of 367.5 billion yuan ($54.3 billion), with founder Liang Wenfeng contributing 20 billion yuan and Tencent, CATL, and JD.com among the backers.

"We started this company without thinking about how much money we would make or going to the capital markets," Liang Wenfeng, founder of DeepSeek, said during a four-hour investor Q&A that covered 118 questions. "We have a very great goodwill toward the world — this is something beyond money."

The fundraising marks a strategic shift for DeepSeek, which previously adhered to a "no financing, no IPO, no commercialization" principle. Liang personally invested 20 billion yuan, while Tencent committed 10 billion yuan, CATL 5 billion yuan, and NetEase, JD.com, and IDG Capital 3 billion yuan each. The National AI Industry Investment Fund contributed 1 billion yuan. The company is exploring a listing on Shanghai's STAR Market, with a potential IPO filing before year-end.

The capital injection helps DeepSeek narrow its compute gap with US rivals — Liang estimated the company trails American AI labs by 12 to 18 months while using one-twentieth the computing resources. The funding will accelerate GPU procurement and workforce expansion, particularly in AI agents and data center operations, as DeepSeek pursues its vision of achieving artificial general intelligence.

Liang's AGI Roadmap Prioritizes Continuous Learning Over Video

Liang outlined a five-stage path to AGI: chain-of-thought reasoning, agent capabilities, continuous learning, a self-iterating singularity, and finally embodied intelligence. He explicitly ruled out video generation and 3D modeling as distractions from the "main line" of intelligence research.

"Video generation has nothing to do with the intelligence roadmap," Liang said. "Commercially it's a good business, but we won't do it just because it's a good business."

DeepSeek's next major model will require continuous learning capability — the ability to improve without retraining — which Liang described as the breakthrough that would unlock general intelligence. Until then, the company focuses on reducing costs and improving inference speed. About half of DeepSeek's core researchers are currently working on data labeling, which Liang called the most important bottleneck in the current stage of AI development.

Domestic Chip Ecosystem Nears a Tipping Point

Liang expressed confidence that China's domestic AI chip ecosystem would be validated within a year, calling it a "historic opportunity" driven by US export restrictions on Nvidia hardware. DeepSeek's V3 model was trained using Nvidia GPUs but bypassed Nvidia's CUDA software stack entirely, using an in-house compiler called TileLang.

"Four Huawei cards can replace one Nvidia card," Liang said, referring to Huawei's 950 supernode. He estimated a hardware gap of "four times plus two years" between Chinese and American chips but said the software ecosystem gap would disappear entirely.

DeepSeek is collaborating closely with Huawei on chip adaptation, though Liang noted that production capacity remains the binding constraint for at least the next two to three years. He dismissed concerns about Nvidia's CUDA moat, arguing that the rise of AI-assisted coding makes it easier than ever to build competing software ecosystems.

For investors, DeepSeek's emergence as a well-capitalized AGI contender introduces a new variable into the global AI competitive environment. Tencent, JD.com, and CATL gain strategic exposure to China's leading independent AI lab at a valuation that, at roughly $54 billion pre-money, remains a fraction of OpenAI's reported $300 billion valuation. Liang's commitment to open-source models and API pricing set at a "reasonable profit" — defined as a 10-month hardware payback period — suggests a deflationary force in AI inference costs that could pressure margins across the sector.

This article is for informational purposes only and does not constitute investment advice.