The DeepSeek funding round is reportedly about 100 billion yuan (roughly $15 billion), with Tencent and battery maker CATL among the largest backers, and it arrives two days after the US ordered federal civilian agencies to remove DeepSeek technology. Bloomberg says the company has raised the money. Reuters, citing two people familiar with the matter, says the round is still being finalized, should close in the coming days, and could still change. As far as I can find, DeepSeek has not commented on either report.

The target moved up at least twice during talks, from 50 billion yuan to 80 billion and then to about 100 billion, according to Bloomberg. That doubling is the detail that stopped me. I can’t tell from the coverage whether any of those increases came after the ban surfaced, so I won’t pretend the order pushed investors in either direction.

Federal civilian agencies were never going to be DeepSeek’s big customers.

A customer base would show up in a September number: an annualized revenue run rate of about $1 billion, which CEO Liang Wenfeng reportedly told investors, along with the claim that higher prices for model access hadn’t driven customers away. That second claim is the one I care about, because in August I looked at V4-Flash costing about three cents to run Artificial Analysis’s index against $3.15 for Claude Fable 5. A lab that cheap, raising prices without losing customers, has a better pitch to Tencent and CATL than any leaderboard position. The revenue figure is secondhand, based on what Liang told investors, and I have no independent check on it.

Where the ban stops is clear. Meta’s MTIA 450 chip was already running DeepSeek’s models in September, and Mistral, which built Large 3 on DeepSeek’s blueprint and now hosts Chinese weights outright, sells to Western customers today. An order about agency networks doesn’t touch open weights that already sit on Western hardware.

The compute side is murkier. A Shanghai securities paper says the money goes to training, computing infrastructure and talent, and earlier reports put a large DeepSeek data center in Ulanqab, Inner Mongolia. How much of it can buy Nvidia-class hardware and how much has to go to domestic silicon like Alibaba’s Zhenwu V900 is a question nothing I’ve read answers. For scale, Bloomberg’s framing is that OpenAI and Anthropic have each raised more than $100 billion, so even a full $15 billion leaves DeepSeek far behind them in cash.

Reuters expects a close within days. Until DeepSeek confirms a number, I’m treating $15 billion as a reported figure and watching its API pricing instead of the valuation.

Sources

Investing.com (Bloomberg), Deepseek raises $15 bln in Tencent, CATL-backed funding round, October 22, 2026
China Daily (Reuters), DeepSeek nears $15 billion funding round backed by Tencent, CATL, October 22, 2026
Global Times, DeepSeek plans to raise $15b in new funding round to boost model training, computing power: report, October 22, 2026
AI Weekly, DeepSeek reportedly raises $15B with Tencent and CATL backing, October 22, 2026