In July 2026, the global AI industry witnessed a seismic shift. DeepSeek, the Chinese AI lab that shocked Silicon Valley with its cost-efficient models, closed a 51 billion yuan ($7.4 billion) maiden funding round — and its founder Liang Wenfeng personally contributed 20 billion yuan ($3 billion) of that, doubling his net worth to $36 billion overnight.
That makes Liang the wealthiest founder among the world's AI large model companies, surpassing Anthropic's Dario Amodei and OpenAI's Greg Brockman, according to the Bloomberg Billionaires Index.
A Funding Round That Rewrote the Rules
The numbers alone are staggering — China's largest-ever single AI funding round, pushing DeepSeek's valuation to between $50 billion and $59 billion. But what truly sets this deal apart is the structure:
Let that sink in. A company in its first external funding round rejected Alibaba and constrained Tencent. That kind of leverage doesn't come from storytelling — it comes from a genuine cost advantage. DeepSeek's API service achieves a theoretical cost profit margin of approximately 545%. For every $1 in API revenue, the gross profit approaches $5.45. That's among the highest in the global large model industry.
From Quant King to AI's Richest Founder
Liang's confidence comes from High-Flyer, the quantitative hedge fund he co-founded in 2016. As one of China's "Big Four" quant firms, High-Flyer manages over 70 billion yuan in assets, with a 2025 average return of 56.55% and a five-year average return of 114.35%.
When Liang founded DeepSeek in July 2023, he funded it entirely through High-Flyer's profits. For three years, the company followed a strict "no fundraising, no IPO, no commercialization" principle. Then in January 2025, R1 arrived — a low-cost, high-performance model that sent shockwaves through Silicon Valley and triggered a sell-off in U.S. tech stocks.
Today, Liang controls approximately 84% of DeepSeek through direct and indirect holdings (34% direct + ~50% via High-Flyer entities). Even after dilution, Bloomberg estimates his stake at approximately 78%.
The Secret Weapon: DeepSeek Is Building Its Own Chips
The funding and fortune are headline-grabbing, but the real bombshell is this: DeepSeek has been secretly developing its own AI inference chip, and the project was quietly launched about a year ago.
Reuters reported that DeepSeek has been recruiting chip design engineers through private channels — headhunters and internal referrals, never public job postings. The company has been in discussions with chip design firms, foundries, and memory suppliers about production plans.
Why inference chips, not training chips?
This mirrors a broader industry trend. OpenAI partnered with Broadcom to launch Jalapeño, its first custom inference chip, in just 9 months. Anthropic is reportedly in talks with Samsung for 2nm custom AI chips. The AI model race has evolved from "parameter wars" to full-stack integration of software and hardware.
Differential Pricing: The AI Industry Starts Counting Pennies
On June 29, DeepSeek announced that V4 would introduce peak/off-peak differential pricing — doubling API costs during business hours on weekdays. A Goldman Sachs research note interpreted this not as demand weakness, but as evidence of surging AI usage putting real pressure on compute infrastructure.
The era of "burning money for users" is ending. What replaces it is a more nuanced calculation: how efficiently can you convert compute into intelligence?
What This Means for AI Search and GEO
DeepSeek's rise has profound implications for the AI search and GEO (Generative Engine Optimization) landscape:
1. Open-source models democratize AI search: DeepSeek's commitment to open-source — V4-Pro surpassed 174,000 downloads in its first week — enables more teams to build AI search applications without vendor lock-in.
2. Falling inference costs accelerate AI search adoption: Custom inference chips will further compress token costs, transforming AI search from a premium service into essential infrastructure.
3. GEO becomes mission-critical: When AI-powered search is cheap enough and pervasive enough, being recommended in AI responses will matter far more than traditional SEO click-through rates. The brands that AI models cite and recommend will capture attention in ways that blue links never could.
A Stanford University report recently concluded that the performance gap between top U.S. and Chinese AI models has "effectively closed." DeepSeek's story proves that in the new AI era, technical independence beats capital bundling, and open beats closed.
The $7.4 billion round isn't the finish line — it's the starting gun for a new phase. Compute infrastructure, custom silicon, talent expansion: each is a multi-billion-dollar commitment. Liang Wenfeng is betting that the company that controls its own hardware stack will ultimately control the AI stack.
---
*This article is based on publicly available reporting. Sources: Bloomberg Billionaires Index, Reuters, Goldman Sachs Research, Sina Finance, CSDN.*
Frequently Asked Questions
How much did DeepSeek raise in its latest funding round?
DeepSeek raised approximately $7.4 billion (50 billion RMB) in its latest funding round, with founder Liang Wenfeng personally contributing $28 billion RMB, making it one of the largest AI funding rounds globally in 2026.
Why is DeepSeek building its own AI chips?
By designing custom inference chips, DeepSeek aims to reduce its dependence on NVIDIA GPUs and lower compute costs at scale. Self-developed silicon gives the company control over its entire AI stack from training to inference, which is critical as model serving costs become the dominant expense.
What does DeepSeek funding mean for the GEO industry?
DeepSeek $7.4B round signals that AI search is becoming the primary information discovery channel. As AI assistants like ChatGPT and DeepSeek themselves answer user queries directly, brands must shift from traditional SEO to GEO (Generative Engine Optimization) to ensure they appear in AI-generated responses.