DeepSeek seeks $71B valuation weeks after $7B raise

Key Takeaways

• DeepSeek closed a $7 billion round in late May at a $52 billion valuation; it's now in talks for a new round at $71 billion
• The capital will fund proprietary data centers, AI chip purchases, and development of DeepSeek's own inference chip
• Rock-bottom pricing on V4 models is driving rapid US enterprise adoption, but performance still trails GPT-5.6 Sol and Claude Mythos

DeepSeek is raising money again. The Chinese AI lab closed its first funding round at $7 billion just weeks ago. Now it's in early talks with investors for a new round that would value the company at $71 billion, according to the Financial Times.

The Hangzhou-based startup needs the cash to build its own data centers and stockpile AI chips. It's also developing a proprietary inference chip to reduce reliance on Nvidia and Huawei. Founder Liang Wenfeng contributed roughly $3 billion of the first round himself, making him the largest backer. Other investors include CATL, Tencent, JD.com, NetEase, and China's state-backed AI fund.

Why DeepSeek needs more capital so quickly

The funding appetite traces directly to DeepSeek's pricing strategy. The company recently released V4-Pro and V4-Flash, open-weights models with up to 1.6 trillion parameters. V4-Pro's pricing has been locked in permanently at roughly eleven times cheaper than GPT-5.5 on input tokens.

That strategy is working. US financial services firm Ramp, which tracks real spending across more than 50,000 companies, reported that DeepSeek ranked among the fastest-growing software vendors among US businesses in June. Ramp flagged security concerns, though, since companies send data directly through DeepSeek's platform.

Sustaining those prices requires infrastructure at scale. DeepSeek's efficiency gains, achieved through Mixture-of-Experts architecture and multi-head latent attention, only go so far when demand spikes. Owning data centers and custom silicon would cut per-inference costs and reduce exposure to chip export restrictions.

The performance gap with Western labs

DeepSeek's models punch above their price class but don't match the top Western offerings. OpenAI's GPT-5.6 Sol and Anthropic's Claude Mythos have reached a new tier that DeepSeek can't match yet. The gap in performance is real. But the gap in price is larger.

For many enterprise use cases, the tradeoff pencils out. A team running tens of thousands of inference calls per day might accept a 10% performance delta if costs drop by 90%. That math explains why US adoption is climbing despite the security flags.

Chinese rivals are closing in

DeepSeek isn't just racing OpenAI and Anthropic. Domestic competition has intensified. Zhipu AI released GLM-5.2, an open-source model that trails Anthropic's Opus by only a few percentage points on hours-long coding tasks. Enterprises are paying attention.

MiniMax is reportedly building a 2.7 trillion-parameter model slated for Q3. Moonshot AI, the company behind the Kimi assistant, is seeking fresh capital at a valuation of up to $30 billion. The race among Chinese AI labs has never been this intense.

This pressure explains the urgency. DeepSeek can't coast on its early-mover efficiency advantage. If a rival ships a comparably cheap model with better benchmarks, the calculus shifts overnight.

What the $71 billion valuation signals

Jumping from $52 billion to $71 billion in weeks is aggressive. It implies investors believe DeepSeek's usage growth will continue, that the pricing moat holds, and that chip restrictions won't cripple the roadmap. Those are big bets.

Building proprietary inference silicon is a multi-year project. Nvidia's H100 and Blackwell chips remain the industry standard for a reason. Huawei's Ascend chips offer an alternative, but export controls and software ecosystem gaps complicate that path. DeepSeek's chip ambitions are defensive, not a near-term performance play.

Security concerns for US enterprises

Ramp's security flag isn't trivial. When US companies send prompts and data through DeepSeek's servers, that data flows through Chinese infrastructure. For regulated industries, finance, healthcare, defense contracting, that's a non-starter. For startups optimizing burn rate, the calculus differs.

Some teams route non-sensitive workloads through DeepSeek while keeping proprietary data on Western providers. That hybrid approach captures cost savings without exposing the crown jewels. But it adds operational complexity.

Frequently Asked Questions

How much did DeepSeek raise in its first funding round?

DeepSeek closed its first funding round at approximately $7 billion in late May, valuing the company at $52 billion. Founder Liang Wenfeng contributed about $3 billion himself.

What is DeepSeek's new valuation target?

DeepSeek is in early talks with investors for a new funding round at a pre-money valuation of approximately $71 billion.

Why does DeepSeek need more capital so quickly?

The company wants to build its own data centers, purchase AI chips, and develop a proprietary inference chip to sustain its aggressive low-pricing strategy and reduce reliance on Nvidia and Huawei.

How does DeepSeek's pricing compare to OpenAI?

DeepSeek's V4-Pro is priced at roughly eleven times cheaper than GPT-5.5 on input tokens, and the company has made these prices permanent.

What are the security concerns with using DeepSeek?

US companies using DeepSeek send data directly through the company's Chinese-based platform, which raises data security and regulatory concerns for enterprises in regulated industries.

Source: The Decoder / Jonathan Kemper