Chips, China's DeepSeek & AI Hacker Agents: Inside This Week's Valuation Surge

Three AI companies with almost nothing in common — a US chip startup, China's most closely watched AI lab, and a cybersecurity firm built by the founder of Mandiant — are all commanding valuations that have doubled or more in a matter of months. The throughline isn't which company has the smartest model. It's who controls the infrastructure underneath the model, and who defends it once it starts acting on its own.

DeepSeek is closing in on a funding round of at least 80 billion yuan (roughly $12 billion), nearly doubling its original 50 billion yuan target, with Tencent and battery giant CATL committing the largest stakes. People familiar with the deal say the final figure could approach 100 billion yuan, and the round could value DeepSeek at a minimum of $75 billion — up sharply from the roughly $50 billion valuation it set in its first external round back in June. The surge in demand followed the release of DeepSeek's V4-Flash model, which investors view as matching Western frontier labs at a fraction of the cost. Notably, the company reportedly paused this round earlier in the year after founder Liang Wenfeng's remarks about DeepSeek's reliance on Nvidia chips went viral — and DeepSeek is now building a data center with at least 160,000 Huawei chips while separately developing its own inference chip to cut dependence on both Nvidia and Huawei. The capital raised here isn't just fueling model development; it's fueling a chip-independence project.

Armadin, founded by Mandiant creator Kevin Mandia, closed a $255.5 million Series B co-led by Andreessen Horowitz and Accel at a valuation above $2.5 billion — just seven months after emerging from stealth, and bringing its total funding to $445 million. Rather than governing or monitoring AI agents already deployed inside a company, Armadin's platform runs swarms of its own AI agents that behave like attackers, probing an organization's systems to find chains of small vulnerabilities a human red team would miss. The thesis, as Mandia puts it, is that "offense is uniquely advantaged right now" because AI lets real attackers find and chain weaknesses faster than defenders can respond — so the only credible defense is running the same kind of AI-speed attack against yourself first.