Bloomberg reports that DeepSeek is close to securing at least 80 billion yuan, about $11.93 billion, in a new round that could approach $15 billion once signed term sheets are counted. The reported DeepSeek $12 billion funding round started with a $7.5 billion target at a roughly $75 billion valuation, then ballooned on investor demand, with battery giant CATL and Tencent writing the biggest checks. Nothing has closed yet, and Reuters notes it could not independently verify the report. But the direction is unmistakable: after the round, DeepSeek plans to restructure for an IPO in early 2027.

For anyone who has followed Liang Wenfeng, the interesting number is not $12 billion. It is zero. That is how many outside investors DeepSeek had for the first three years of its life.

The Man Who Did Not Want Your Money

Liang is the quant-finance founder who built DeepSeek inside his hedge fund High-Flyer with his own money and his own computing power. He studied engineering at Zhejiang University, co-founded High-Flyer in 2015, and started DeepSeek in 2023. When its R1 model matched top Western models in early 2025 at a fraction of the training cost, he became the most famous founder in Chinese tech without taking a single outside check.

He was explicit about why. Asked in a 2024 interview whether DeepSeek had fundraising plans, Liang said the company had no short-term plans to raise, adding that the real bottleneck had never been funding but the embargo on high-end chips. Industry watchers summed up his stance as three iron rules: no fundraising, no IPO, no commercialization. And it worked. The purity bought him elite talent, total focus, and a mystique no marketing budget could buy. He owned 84 percent of the company before outside money arrived, according to a TechCrunch analysis of corporate records, and he still put about $3 billion of his own assets into the June round.

Why the Math Changed

Then scale sent the bill. DeepSeek is reportedly building a data center in Inner Mongolia with at least 160,000 Huawei AI chips, developing its own inference chip to cut dependence on Nvidia and Huawei alike, and chasing AGI while every rival lab arms for the same prize. The investor frenzy follows the V4-Flash model, which set new benchmarks for cost and performance against Anthropic and OpenAI. A research skunkworks funded by a trading firm’s profits can do a lot. It cannot do all of that.

The business changed too. The Information reported in September that DeepSeek’s annualized revenue run rate had hit $1 billion, more than doubling in a few months after API price hikes of 2.3 to 4.5 times. Liang told investors the company devotes more than 70 percent of its computing capacity to training new models. A company with a billion-dollar revenue run rate, 160,000 chips on order, and an IPO on the calendar is not a side project anymore. It is a company, and companies need capital structures.

The IPO machinery is already moving. Reuters reported in September that DeepSeek hired CITIC Securities to prepare for a possible listing on Shanghai’s STAR Market, and that the company hired its first CFO, GL Ventures partner Yan Wentao. The June round, about $7.4 billion at a valuation above $50 billion, was the first crack in the wall. This round knocks the wall down. And DeepSeek is not alone in the stampede: rival Moonshot AI has filed confidentially for a Hong Kong listing, while Zhipu and MiniMax are already listed there.

What the DeepSeek $12 Billion Funding Round Teaches Founders

Here is the uncomfortable part for founders who admired Liang’s purity: he was right then, and he is right now. No fundraising was brilliant strategy when DeepSeek was a research lab and High-Flyer’s trading profits covered the bills. It is fiction when you need a 160,000-chip data center. Principles are strategy, and strategy changes when the facts change.

Founders fetishize consistency. They cling to old rules to protect their image while the company starves. Liang is doing the harder thing: rewriting the rules in public and taking the heat. He even paused this same round in July after his private remarks at an investor meeting leaked and went viral, then came back when the terms were right. That is not selling out. That is negotiating from strength, in the same compute arms race every AI founder is now fighting.

The real test is control. What made DeepSeek work was a founder with majority ownership answering to nobody. The question now is whether the Tencent and CATL money comes with the kind of influence Liang spent a decade refusing. The reported terms suggest he knows this: in a May meeting with prospective backers he demanded no poaching of his people and no encouragement of spinouts. Keep watching the governance, not the valuation.

What to Watch Next

First, whether the round actually closes at $12 billion or pushes toward $15 billion, and at what valuation. Second, the IPO timeline: early 2027 is the reported plan, and a STAR Market listing needs Chinese regulatory clearance, so the calendar can slip. Third, whether the research culture survives the money. DeepSeek has already lost key researchers to Xiaomi and ByteDance, and nothing kills a talent-dense lab faster than becoming a normal company with normal politics, in an industry where even OpenAI is shedding researchers.

In two years we will remember this as either the day DeepSeek became a company or the day the money broke it. Bet on the founder who changed his mind in public rather than in secret. And write your own rules for the company you are, while staying ready to rewrite them for the company you want to become.

Read more in our Entrepreneurship & Startups hub.

Sources: Reuters, The Decoder, Reuters, PYMNTS, Wikipedia

The post The Founder Who Swore Off Investors Is About to Raise $12 Billion appeared first on CEO Medium.

Xiaomi, best known for smartphones and electric cars, just released MiMo-V2.6-Pro, which debuted as the highest-scoring open-weight AI model in the world on Artificial Analysis’ Intelligence Index with a score of 46. That puts it ahead of DeepSeek’s best open model at 36, ahead of xAI’s Grok 4.6, and within touching distance of closed frontier models from Anthropic and OpenAI.

The model carries an MIT license. As a result, anyone can download it from Hugging Face, fine-tune it, and run it on their own hardware without paying Xiaomi anything.

Xiaomi MiMo vs DeepSeek: the cost story is the real story

Here is the number that should make every founder sit up: Xiaomi says it trained the flagship model for roughly $2.62 million, completing its reinforcement learning run in under six days. In addition, the smaller Flash variant cost about $850,000.

Not long ago, people discussed frontier AI training runs in terms of hundreds of millions of dollars. In contrast, Xiaomi did it for the price of a nice house in San Francisco. Fuli Luo, a former DeepSeek researcher, led the team, which tells you something about where the talent is flowing.

Moreover, API pricing is equally aggressive: $0.435 per million input tokens and $0.87 per million output tokens. Artificial Analysis measures the cost per task at roughly one-twentieth to one-sixtieth of leading international models.

What this means for your business

First, if you are building a company on top of AI APIs, your cost structure just got a preview of its future. Open-weight models at near-frontier quality, available for free, runnable on rented hardware. Labs releasing their work to the world are competing away the margin that proprietary API providers used to enjoy.

The strategic move for founders: build your product so the model underneath is swappable. The winners of the next phase will not be the companies with the best model. They will be the companies with the best product on top of commodity intelligence.

DeepSeek proved open models could compete. Finally, Xiaomi just proved they can win. Your AI bill is about to get a lot smaller, and your moat needs to be somewhere else.

Read more in our Entrepreneurship & Startups hub.

Sources: https://venturebeat.com/technology/better-than-deepseek-xiaomis-mimo-v2-6-pro-debuts-as-the-top-open-weights-model-in-the-world-alongside-cheaper-v2-6-flash

The post Xiaomi Just Dethroned DeepSeek in Open-Source AI. Your AI Bills Are About to Collapse appeared first on CEO Medium.