Deepseek Funding Round Nears $12B Ahead of 2027 IPO

Deepseek's latest fundraise has grown well past its original target. According to Bloomberg, the Chinese AI startup is close to securing at least $12 billion, and people familiar with the talks say the final figure could get near $15 billion. The company first set out to raise about $7.5 billion at a valuation of roughly $75 billion.

Two backers stand out. Battery maker CATL and internet giant Tencent are putting in the largest amounts. CATL declined to comment on the report. Tencent and Deepseek did not respond to requests for comment.

Why investors are lining up

The demand traces back to one model: V4-Flash. Deepseek released V4-Flash and V4-Pro as preview versions in April. V4-Flash has since set new benchmarks for the ratio of cost to performance when measured against models from Anthropic and OpenAI.

That matters because cost is now a central part of the AI competition. A model that delivers strong results at a lower price is attractive to developers and businesses that run large workloads. For investors, it also signals that a lab outside the US can compete directly with the best-funded companies in the field.

A round that stalled, then grew

The path to this round was not smooth. Deepseek had put the fundraise on hold after remarks by founder Liang Wenfeng about the company's dependence on Nvidia chips spread widely online. The pause did not last, and the round is now larger than planned.

This is not Deepseek's first time raising outside money. In June, the company closed its first external round at about $7.4 billion, with a valuation above $50 billion. The new round would come only months later, at a higher target valuation and with more capital on the table.

The road to an IPO

Once this round closes, Deepseek plans to restructure itself in preparation for an initial public offering in early 2027. That would put the company on a similar timeline to several US labs that are also moving toward the public markets, including OpenAI's own pre-IPO fundraising.

Three points shape the plan:

1. Compute. Deepseek is building a data center with at least 160,000 AI chips from Huawei.
2. Chip independence. The company is also developing its own inference chip. The goal is to reduce its reliance on both Nvidia and Huawei.
3. Open models. Liang wants Deepseek to keep releasing open models with freely available weights, even as the company takes on large outside investors and heads toward a listing.

The chip question has followed Deepseek for some time. Its large Huawei order fits with earlier work such as the open-source tools it shipped for Huawei Ascend chips. An in-house inference chip would go one step further and give the company more control over the hardware that runs its models.

The open-weight commitment

Liang's stance on open weights is notable in the context of an IPO. Many labs that raise large sums move toward closed models to protect their products. Deepseek's founder is signaling that he wants to keep the opposite approach. Open weights let anyone download and run the models, which helped build Deepseek's reputation with developers worldwide.

How that approach holds up under pressure from public shareholders remains an open question. For now, it is part of the pitch.

The Bigger Picture

This round suggests that investor appetite for Chinese AI labs is strong, at least when they can show competitive models at low cost. The backing from CATL, a battery maker, and Tencent, a tech platform, also points to broad interest from large Chinese companies outside the core AI industry.

For readers following the global AI race, the story ties together three trends. First, cost efficiency has become a key way to compete with US labs, and V4-Flash appears to be Deepseek's strongest argument there. Second, hardware remains a weak spot. The viral comments about Nvidia reliance, the Huawei data center and the custom inference chip all show how much chip access shapes strategy. Third, open-weight models are now a competitive field of their own, with US startups such as Reflection taking on Chinese rivals directly.

It is worth watching whether Deepseek can keep its open-weight policy once it becomes a public company. It is also worth watching whether its own chip moves past development, and whether the Huawei data center can deliver the compute it needs. The final size of the round, somewhere between $12 billion and $15 billion if the reports hold, will be an early signal of how much confidence investors place in that plan.