News
Research Report
The AI Transformation 100: Concrete Ideas for Fixing How We Work
Key Takeaways
- DeepSeek raised about $7.4 billion at a valuation above $50 billion in its first funding round.
- The deal structure keeps control with CEO Liang Wenfeng while giving most investors no voting rights.
- DeepSeek’s rapid model releases and aggressive pricing continue to pressure Western AI rivals.
Chinese AI startup DeepSeek raised more than 50 billion yuan (about $7.40 billion) at a valuation exceeding $50 billion in its first funding round, according to a report from The Information on June 16, 2026. The deal used an unusual structure designed to preserve founder control.
The funding required investors to place capital into a limited partnership managed by DeepSeek CEO Liang Wenfeng rather than into DeepSeek itself, the report said. Most investors are subject to a five-year lock-up and hold no voting rights.
China's National Artificial Intelligence Industry Investment Fund was the only exception, investing directly in DeepSeek and retaining both voting rights and freedom from the lock-up.
Earlier in June, Wenfeng committed 20 billion yuan of his own money, while Tencent is considering 10 billion yuan and battery giant CATL is weighing 5 billion yuan, which would make them the largest external investors.
Table of Contents
Breaking Down the DeepSeek Deal Structure
The funding round's architecture concentrates authority with DeepSeek's founder while limiting investor influence.
| Deal Feature | How It Works |
|---|---|
| Limited partnership vehicle | Investor capital flows to an entity managed by CEO Liang Wenfeng |
| Five-year lock-up | Most investors cannot withdraw capital for five years |
| No voting rights | Most investors hold no voting power over company decisions |
| National AI Fund exception | China's National AI Investment Fund invested directly in DeepSeek |
| Founder capital commitment | Liang Wenfeng committed 20 billion yuan of personal funds |
Recent DeepSeek Developments
DeepSeek has maintained a relentless model release cadence while dismantling the pricing assumptions underpinning Western AI vendors. The Hangzhou-based lab's R1 reasoning model — trained for a claimed $294,000 using 512 Nvidia H800 chips — delivered reasoning at costs estimated 20 to 50 times cheaper than comparable OpenAI offerings.
August 2025 brought V3.1, a 685-billion-parameter hybrid unifying chat, reasoning and coding in a single open-source system that benchmarked at parity or better versus OpenAI and Anthropic. By December 2025, V3.2 variants introduced a Sparse Attention mechanism that cut long-context AI inference costs by 50 675%, and April 2026 saw the preview of V4-Pro and V4-Flash, a 1.6-trillion-parameter open-weight family with a 1-million-token context window.
The pricing pressure reached a new threshold in April 2026 when DeepSeek launched V4-Pro with a 75% price reduction. V4-Pro was also optimized for Huawei Ascend infrastructure, a deliberate pivot from Nvidia that triggered a scramble among Chinese tech firms to secure Huawei AI chips.
DeepSeek's momentum has not come without friction. In February 2026, Anthropic accused the lab of running more than 24,000 fraudulent accounts and 16 million exchanges to distill Claude's capabilities, allegations that landed amid active US debate over chip export controls. German regulators separately sought to delist DeepSeek from Apple and Google app stores over unlawful data transfers to China.
DeepSeek Background
Founded in 2023 in Hangzhou, China, DeepSeek targets software developers, AI and machine learning engineers, startups, research labs and enterprises seeking advanced AI capabilities. The company develops large language models and related infrastructure, with products including public chat tools, a mobile app, a developer platform and APIs.