Chinese AI lab DeepSeek is in advanced talks to raise $1.5B at a $71B valuation, just one month after closing a $7B round at $50B. The move funds the runway toward a 2027 IPO that could pull forward to late 2026, at a moment when the lab runs on Huawei chips and already captures nearly a quarter of the enterprise AI tokens processed by Vercel.
Key Takeaways
- DeepSeek is in talks for a fresh $1.5B round at a $71B valuation, up from $50B last month.
- The IPO is targeted for 2027 with a possible pull-forward to late 2026, with Tencent and the Beijing National AI Industry Investment Fund leading the round.
- DeepSeek captured close to 23% of enterprise AI tokens processed by Vercel in June, on Huawei chips despite US export restrictions.
Have an AI Sum Up This Article
ChatGPTA Round That Doubles the Valuation in One Month
The deal in motion covers $1.5B of fresh capital. The Chinese lab had already closed a $7B round the previous month at a $50B valuation. The new round rewrites the mark to $71B, which is a 40%+ valuation jump in thirty days for one single AI startup.
The round is led by two strategic investors. Tencent, the Chinese digital giant, takes the lead alongside the Beijing National Artificial Intelligence Industry Investment Fund. The setup sends a clear signal: Chinese capital is closing ranks around a national AI champion to fund the next stage, IPO included, without depending on Western capital pools.
The IPO target is officially 2027, but internal investors flag a possible pull-forward to late 2026. The listing venue has not been named, which leaves open the Hong Kong versus Shanghai debate. That choice is a political one: a Hong Kong listing opens the stock to Western flows, a Shanghai listing keeps it inside the domestic circuit.
DeepSeek’s funding trajectory traces a classic Western AI unicorn path, but compressed into a shorter tempo. OpenAI and Anthropic each took four years to move from early rounds to eleven-figure valuations. DeepSeek covers the same ground in two years, with a leaner cash burn and Huawei silicon instead of Nvidia. That hardware choice sits at the core of its strategy, DeepSeek cutting its Nvidia dependency.
Why Enterprises Are Rerouting Their Tokens to DeepSeek
The number that explains the valuation does not come from an academic benchmark. In June 2026, DeepSeek captured close to 23% of the enterprise AI tokens processed by Vercel. A quarter of the AI traffic on a major US cloud platform already runs through a Chinese lab. That performance is measured in real production volume, not in lab scores.
The driver behind that adoption is cost. DeepSeek prices its tokens well below OpenAI and Anthropic for a level of performance that is good enough on most production workflows. The pricing breaks the budget habits of teams that assumed they would live indefinitely with an Anthropic or OpenAI ticket. The lab is also pushing speed, dSpark accelerating inference by 85%.
The typical entry point stays US infrastructure. Vercel, Cloudflare Workers AI and several other platforms carry DeepSeek in their catalog next to the American labs. The enterprise customer never negotiates directly with a Chinese vendor: they toggle an option in a dashboard and their workflow reroutes. That commercial abstraction unlocked enterprise adoption much faster than any direct contract would have.
The priority use cases are the cost-sensitive workloads: email generation, structured extraction, bulk summaries, support ticket triage. Workloads where the brand of the model matters less than the bill. On these workloads, DeepSeek is settling as the new implicit standard, which is also why Anthropic has admitted that Claude runs office work more than code: the code segment, more demanding on quality, holds up better against the Chinese pricing pressure.
More articles on Horizon
- OpenAI Preps a Screenless AI Speaker That Moves
- Grok 4.5 vs Muse Spark 1.1: Our Test Verdict
- 16 Nobel Laureates Say Prepare the Economy for AI Now
US Labs Feel the Pressure on Pricing, Not on the Headlines
The response from US labs flows through pricing. OpenAI opened GPT-5.6 in a worldwide rollout last week with a tariff grid tuned lower on high-volume tokens. Anthropic extended its Fable 5 promo through July 19 to hold its pro conversion tiers. Each player pushes in the same direction, under the same constraint: keep a quality gap that justifies the residual price premium.
The Huawei chips are the angle that makes the story even more delicate for Washington. DeepSeek runs on Chinese silicon despite US restrictions on Nvidia exports to China. The performance achieved shows that the Huawei fabric can carry a frontier lab, which undermines the central argument behind the restrictions: cutting China off from cutting-edge chips to slow down its AI output. The compute is moving forward anyway.
The IPO question opens another front. If DeepSeek lists in Hong Kong in 2026 or 2027 at a triple-digit-billion valuation, Western institutional investors will have to choose between entering the capital through a HK float or bending to US restrictions on capital directed at Chinese tech. The trade-off will be political as much as financial.
The timing is tight. A late 2026 IPO forces the audit, the financial roadmap and the investor storytelling to close in less than six months. The $1.5B round funds exactly this IPO-prep hardware: CFO hiring, compliance structuring, measured international expansion so the thesis holds up in front of bankers. DeepSeek is entering the phase where the lab becomes a machine that tells its story to the markets, on top of training its models.
Follow the story on Horizon.
