Anthropic’s Mythos preview spurred DeepSeek’s $7.4B raise

Chinese AI lab DeepSeek closed a $7.4 billion fundraising round — its first ever external capital raise — after Anthropic's Mythos model preview alarmed its leadership about a growing capability gap between U.S. and Chinese frontier AI, according to The Information Tencent Holdings Limited.x+1

The report, published this week, said CEO Liang Wenfeng concluded after seeing the Mythos preview that China's top AI lab needed far more money to remain competitive. The decision marked a reversal for DeepSeek, which had built a reputation for producing frontier-quality models at a fraction of the cost of U.S. rivals, all without taking a dollar of outside investment.x+1

The Deal

The round values DeepSeek at more than $50 billion, making it China's most valuable AI startup. Liang himself wrote the largest check, contributing roughly 20 billion yuan (about $2.85 billion), followed by Tencent with 10 billion yuan and battery maker CATL Contemporary Amperex Technology Co., Limited with 5 billion yuan. JD.com ↘ 0.97%, NetEase ↘ 0.60%, and IDG Capital each contributed 3 billion yuan.cnbc+3

Under the deal's unusual structure, all investors except China's National Artificial Intelligence Industry Investment Fund must place their capital into a limited partnership managed by Liang rather than directly into DeepSeek. Those investors face a five-year lockup and hold no voting rights. Only the state fund invested directly and retained both voting rights and freedom from the lockup.michaelparekh.substack+1

What Spooked DeepSeek

Anthropic debuted its Mythos preview in early April as part of Project Glasswing, a cybersecurity initiative involving partners including Amazon Amazon.com, Inc. ↘ 1.54%, Apple ↗ 0.36%, Microsoft ↗ 1.75%, and CrowdStrike ↗ 20.50%. The model autonomously identified thousands of critical zero-day vulnerabilities across major operating systems and browsers — some dating back decades. Anthropic restricted public access, sharing Mythos only with select partners to patch software before attackers could exploit similar AI capabilities.techcrunch+2

The model's performance represented a leap in agentic reasoning and security capabilities that, according to The Information's reporting, convinced Liang that DeepSeek's lean, self-funded approach was no longer tenable.x+1

A New Phase

DeepSeek is also reportedly planning to double the size of all departments as it scales to compete with both domestic Chinese rivals and global leaders. Liang has told investors the company will prioritize pioneering AI research over immediate commercialization while continuing to advance open-source initiatives.siliconrepublic+1