DeepSeek is already back for more. The Hangzhou-based AI lab is in early talks with investors for a new round at a $71 billion pre-money valuation, just weeks after closing its first-ever funding round at $52 billion. Founder Liang Wenfeng put roughly $3 billion of his own money into that initial raise, which brought in backers like Tencent, JD.com, NetEase, and China’s state-backed AI fund. The speed is striking — but so is the burn rate.
The reason for the quick return to market is straightforward: DeepSeek’s pricing strategy is a cash furnace. The company recently made rock-bottom prices permanent on its V4-Pro models, which come in at roughly eleven times cheaper than GPT-5.5 on input. That kind of pricing drives adoption — US financial services firm Ramp reports DeepSeek was among the fastest-growing software vendors for American businesses in June — but it demands enormous infrastructure. The fresh capital is earmarked for building proprietary data centers and buying AI chips, reducing reliance on rented compute. DeepSeek is also developing its own inference chip to lessen its dependence on Nvidia and Huawei.
Performance-wise, the picture is mixed. OpenAI’s GPT-5.6 Sol and Anthropic’s Claude Mythos have opened a genuine gap that DeepSeek can’t yet close. Still, the price-to-performance ratio makes that gap feel almost academic for many customers. If you’re getting close-enough capability at one-eleventh the cost, how many benchmarks do you really need to win?
The domestic pressure is just as intense as the Western competition. Zhipu AI’s GLM-5.2 is now within striking distance of Anthropic’s Opus models on long-duration coding tasks. MiniMax is reportedly targeting a 2.7 trillion-parameter model for Q3. Moonshot AI is shopping for capital at a $30 billion valuation. DeepSeek isn’t just racing OpenAI — it’s trying to outrun a whole pack of well-funded Chinese rivals who see exactly the same opportunity.
