Tsinghua team's Qujing raises Series A, past $1.4B total in six months
What's the deal? Beijing Qujing Technology closed a Series A round in July 2026, led by Henan Investment Group Huirong FundDealroom has a profile for this one. Try Dealroom →, with existing backers including Zhenzhi Capital, Shangshi Capital, Xinglian Capital, Shanghai Guofang Innovation, Honghui Fund, Huakong Fund, and Hangzhou Fucheng making follow-on investments.
The round caps a rapid fundraising run: Qujing has raised more than $1.4 billion in total over roughly six months. It closed an angel++ round backed by Paratera in February and a Pre-A round in May, co-led by Xinglian Capital and Huakong Fund.
Why now? Founded in late 2023 by a team from Tsinghua University's high-performance computing institute, Qujing bet early on AI inference rather than model training. That wager is paying off as demand for AI tokens surges — China's daily token usage topped 140 trillion by March 2026, more than 1,000 times the level two years earlier, according to the National Data Bureau.
What's the endgame? Qujing positions itself as a high-quality "AI token factory," selling Token as a Service through its ATaaS platform. It runs two models: producing tokens directly on leased compute for model makers and enterprises, and co-operating token factories built for clients with their own compute.
Chief executive officer Ai Zhiyuan frames the strategy as "few models, deep optimisation" — concentrating resources on a handful of high-value models rather than broad compatibility. "Training is a cost item; inference is a profit item," he said.
By the numbers: Since the 2026 Lunar New Year, Qujing says AI token output per machine rose more than three-fold and total high-quality token production climbed more than 30 times. Revenue in June 2026 alone exceeded its full-year 2025 total, and the company says its next round is already underway.
What could go wrong? The AI infrastructure space is crowding fast, and rivals can match individual capabilities — low latency, high concurrency, stable output — in isolation. Qujing's bet is that few competitors can deliver all of them at once, at production scale and low cost. Sustaining that edge as compute-rich incumbents move in remains the open question.
The signal: As tokens become the currency of the AI era, the once-hidden layer beneath model builders and applications is emerging as its own high-growth business. Qujing's quick re-raise — and repeated top-ups from existing investors — reflects a broader shift in capital toward inference infrastructure, where efficiency, not raw GPU count, increasingly decides who wins.
Image credit: jurvetson
