DeepSeek nears $12 billion round from CATL, Tencent

By VK

Published 7 Oct 2026, 11:27 UTC

Reporting from The Decoder

DeepSeek is nearing a funding round of $12 billion or more, with CATL and Tencent the top backers, Bloomberg reports. An IPO is planned for 2027.

What it means for founders

Low prices look funded, not temporary. A war chest of $12 billion or more, plus compute it plans to own, makes it more likely that DeepSeek keeps undercutting US labs on API rates. If you route bulk work such as extraction, tagging or summaries to it, budget on today's rates holding for a while, but keep a second provider wired in.
Open weights are a stated intent, not a contract. Liang's commitment is a founder's preference, and an IPO adds new owners and disclosure pressure. If your product depends on self-hosting DeepSeek weights, watch the licence on each release and keep a fallback, for example a Western open model such as Reflection AI's Beam.
Procurement risk does not shrink with a bigger cap table. Ramp has warned that companies using DeepSeek's hosted platform send their data to it. Enterprise buyers and regulated customers will ask where inference runs. Self-hosting the weights on your own infrastructure answers that better than the hosted API does.
Signals to watch: the close of this round, any filing tied to the early 2027 listing, and whether the Huawei chip build changes the speed or price of DeepSeek's API.

The story

DeepSeek is close to securing at least $12 billion in new funding, with CATL and Tencent writing the biggest checks, Bloomberg reports, according to a summary by The Decoder published October 6. People familiar with the talks said the round could end up near $15 billion. The Hangzhou lab had first set out to raise around $7.5 billion, valuing the company near $75 billion. None of this is confirmed by the company: CATL declined to comment, and neither Tencent nor DeepSeek responded.

Why the DeepSeek funding round grew

The reported pull is DeepSeek V4-Flash, which The Decoder describes as resetting the price to performance bar against OpenAI and Anthropic models. V4-Pro and V4-Flash arrived as previews in April, and DeepSeek later locked in its 75 percent V4-Pro price cut for good. Cheap tokens need capital behind them. The company is reportedly building a facility that will hold 160,000 or more Huawei AI chips and designing an inference chip of its own to lean less on Nvidia and Huawei.

The round was not smooth. DeepSeek reportedly paused it after remarks by founder Liang Wenfeng about the company's dependence on Nvidia hardware spread widely online.

A second outside round, then a 2027 IPO

For most of its life DeepSeek ran without outside investors. That changed this year, when it closed its first external round, worth about $7 billion at a valuation above $50 billion. According to The Information, that deal routed investors through a limited partnership run by Liang, with no voting rights and a five-year lock-up; the Chinese government's AI investment fund was the one direct investor that kept a vote. CATL and Tencent were already among the largest backers then.

Once the new round closes, DeepSeek reportedly intends to restructure ahead of a listing in early 2027. Liang has said he wants to keep releasing models with open weights.

What we don't know yet

The final size, the valuation at close, and whether the new money carries voting rights or the same lock-up terms.
Where DeepSeek would list, and what the restructuring involves.
Whether IPO disclosure rules or public shareholders change how freely it publishes weights.
When V4-Flash and V4-Pro leave preview.