In March, Junyang Lin said goodbye to the AI models he had spent years building at Alibaba. By November, investors had decided his new company was worth $5 billion.

It hasn’t released a product yet.

Pragmatik Labs, the Shanghai startup Lin founded after leaving Alibaba, has closed a new funding round that values it at $5 billion, Bloomberg reported on November 20, 2026, citing people familiar with the deal. The round raised tens of millions of dollars. In August, the company’s first financing valued it at $2 billion.

That’s two and a half times the price in about three months.

A valuation is the price investors agree the whole company is worth when they buy their shares. Tens of millions of dollars against a $5 billion price means the newest backers bought only a small slice.

Ex-Alibaba Qwen chief’s Pragmatik Labs doubles valuation to $5B in new round

The man who built Qwen

Lin, born in 1993, came to AI through language. He studied linguistics at Peking University and joined Alibaba’s DAMO Academy, the group’s research arm, in 2019.

At Alibaba he became the technical lead of Qwen, the company’s family of large language models. Alibaba released many Qwen models as open source, which means anyone can download them and build on them for free. Lin was the public face of that work, and he became one of the youngest engineers to reach P10, a senior rank at Alibaba.

Then, on March 4, 2026, he left. He walked away from Qwen with a short goodbye post and nothing announced to replace it. That meant giving up one of the most visible jobs in Chinese AI to start over with no product, no customers and no public name. Investors moved fast anyway. Reports in early summer said he was raising several hundred million dollars at a $2 billion valuation, and in mid-August he made the company official on X. Gaorong Ventures and HSG, the firm formerly known as Sequoia China, co-led that first round, and Tencent and a Shanghai government-backed fund joined in.

The name points back to his studies. Pragmatics is the branch of linguistics that looks at how context changes what words mean. The company’s Chinese name, 语用, means the same thing.

What Pragmatik is building

Pragmatik builds AI agents for both digital and physical settings. An AI agent is software that plans a task, carries out the steps and adjusts based on what happens.

The digital side covers knowledge work and business operations. The physical side is about machines that sense real environments and act in them, the kind of AI that could one day run a robot. According to an account of the August launch, the company’s website also lists frontier research and turning that research into products.

So far, no product has been shown, and no customers or revenue have been disclosed.

Physical AI is attracting big checks right now. Genesis AI recently raised money at a $3.5 billion valuation and plans to mass-produce a robot called Eno.

Why a carmaker is on the list

The new investors are HyT Capital, CDH Investments and the parent company of carmaker Geely Automobile Holdings.

Geely is the name that stands out. A carmaker builds machines that have to read their surroundings and decide what to do next. That’s close to what Pragmatik says its physical agents will do.

Chinese companies are also funding the hardware underneath agents. In one recent deal, Meituan led a round in a startup designing CPUs for AI agents.

The earlier backers include Tencent, one of China’s largest tech companies. So Pragmatik now counts a gaming and social media giant, a group of venture firms, a city government fund and a car group among its owners.

What happens next

Pragmatik, HyT, CDH and Geely did not respond to requests for comment. The company has not confirmed the round publicly.

It also hasn’t announced a product, a launch date or any other next step. For now, the $5 billion price rests on a reported deal and on what Lin did at Qwen.

Laura is a business historian and researcher who writes about how companies began and grew. At Companies History, she works primarily from original sources such as annual reports, regulatory filings, patents, and archived news coverage. Her articles often correct popular myths about famous founding stories.