Let's get the number on the table first: at least $12 billion. That's what DeepSeek, the Hangzhou-based AI lab, is set to raise in a new private round led by Tencent and battery maker CATL, according to Bloomberg's reporting (Investing.com). People familiar with the talks say signed term sheets could push it as high as 100 billion yuan — roughly $15 billion.

Here's the part that makes the round interesting: DeepSeek didn't ask for that much. It went out looking for about 50 billion yuan, around $7 billion. Demand came back "far bigger than initially expected" (TNW), and the target nearly doubled on the way to close.

A company raising more money than it planned isn't news on its own. A company raising it in this market, at this size, with these backers, right before a planned IPO — that's the story.

The thesis: the price of a credible frontier-model lab just got quoted out loud, and it has a lot of zeros.

🧠 Why This Matters

DeepSeek is the lab that spent most of 2025 forcing everyone else to answer an uncomfortable question about cost. Its R1 model, released as open-weight early that year, performed close to the expensive closed models while running far cheaper. Its V4 and V4 Flash models, out in 2026, did it again.

Bloomberg's sources described the V4 line in blunt terms:

"[V4 models were] widely lauded for having performance comparable to advanced models from OpenAI and Anthropic while operating at a fraction of their costs." — Bloomberg, as reported by Investing.com

That is the pitch investors are buying. Not a bigger model — a cheaper one that keeps up. When your unit economics are a fraction of a rival's and your weights are open, the money tends to find you. This round is that dynamic turning into a term sheet.

📊 Deep Dive

The raise reportedly builds on a previous private round earlier in 2026 that valued DeepSeek somewhere in the $52–59 billion range (TNW). That earlier round had a distinctive cap table: founder Liang Wenfeng put in roughly 20 billion yuan himself, with Tencent at about 10 billion, CATL around 5 billion, and participation from JD.com, NetEase, and China's national AI investment fund.

How the new round stacks up against the rest of the AI fundraising board right now, per CoinDesk's tally (CoinDesk):

  • DeepSeek — raising at least $12B (up to ~$15B); IPO reportedly targeted for early 2027.
  • Moonshot AI — valued at $50B after its final private round; Hong Kong listing eyed for Q1 2027, potentially raising up to $5B.
  • OpenAI — reportedly raising $30B+ at a $1.4 trillion pre-money valuation.
  • Anthropic — IPO reportedly targeted for around November, at a valuation near $2 trillion.

Two things jump out. DeepSeek's $12–15B raise is enormous by any normal standard and still a rounding error next to the trillion-dollar figures attached to the largest US labs. And the clock matters: DeepSeek, Moonshot, OpenAI, and Anthropic are all lining up capital ahead of public listings inside roughly the same 18-month window.

⚠️ The Catch

Start with what isn't confirmed. The round is "near," not closed. The $12 billion floor and the $15 billion ceiling both come from people briefed on live negotiations, and live negotiations move. Treat the range as a range.

Valuation is the murkier number. Reports of the prior round's price don't fully agree — you'll see figures from the low $50-billions up toward the $70-billions depending on the source — which is a reminder that private marks on fast-moving labs are estimates, not quotes on a screen.

Then there's the hardware question every lab now faces. DeepSeek's edge has always been doing more with less, but "less" still means a great deal of compute, and the company has said it intends to expand capacity. Export controls shape what chips Chinese labs can buy, and that constraint doesn't go away because the fundraising went well.

🎯 What Happens Next

Watch for the close. Once the round is final, the headline figure and the valuation stop being reported ranges and become the number DeepSeek carries into its IPO conversations. The company has reportedly engaged CITIC Securities for a possible listing on Shanghai's STAR Market (TNW), with early 2027 floated as the window (CoinDesk).

Watch the backers, too. CATL is a battery company, not an obvious AI investor. Its presence — alongside Tencent writing another check — says the capital lining up behind DeepSeek reaches well past the usual tech crowd.

And watch the next model. The whole bull case rests on DeepSeek shipping capable models at low cost. The round buys runway; the next release has to keep the cost-performance claim true.

🧩 Bigger Picture

Step back and the shape is clear. A handful of AI labs — some in the US, some in China — are raising billions each within months of one another, most of them pointed at a public listing. DeepSeek and Moonshot are both stacking capital ahead of 2027 IPOs; OpenAI and Anthropic are chasing far larger figures on their own timelines (CoinDesk).

What DeepSeek adds to that picture is the open-weight, low-cost variable. Its models are free to download and cheap to run, which is exactly why they spread — and exactly why investors are willing to fund the next round at a size the company didn't even ask for. The market has decided that a lab proving frontier-level results don't require frontier-level budgets is worth a premium.

Whether that holds depends on the thing no term sheet can buy: the next model being as good, and as cheap, as the last one.

DeepSeek went out for $7 billion and investors tried to hand it twice that. The hard part was never raising the money — it's staying cheap enough to deserve it.

Sources