Why China’s Answer to Surge AI Got a $1 Billion Valuation On Just $30 Million in Orders
Juro Osawa (with Jing Yang) reports that Chinese AI data/evaluation startups UniPat and Humanlaya have reached ~$1 billion valuations despite still-small order books, positioning as China’s analogues to U.S. post-training data firms Surge, Mercor and Scale. Dimension Capital flagged the cohort after a China research trip, noting a shift from rudimentary annotation toward evaluation/verification infrastructure; it singled out Beijing-based UniPat, whose specialized benchmarks are used to evaluate U.S. and Chinese models. UniPat (founded Dec.; co-founder Li Kuan, ex-Alibaba RL infrastructure) recently closed a round at a $1 billion pre-money valuation (person familiar). Competitor Humanlaya (co-founder Jasper Jiao, ex-ByteDance/Xpert) closed a new round at nearly $1 billion post-money (two people familiar). Caveat on multiples: as of end-May UniPat had only $30 million in customer orders (person familiar), so the $1B mark looks extreme vs forward revenue; Mercor by comparison had >$2B gross revenue run rate in June at ~$10B valuation. VCs argue the premium reflects China’s still-nascent domestic high-end data/post-training market. Both sell human-expert and synthetic data plus domain benchmarks (coding, biomedical, visual). Major Chinese model makers—including Alibaba, ByteDance, Moonshot and DeepSeek—have bought from them (people familiar). Some Chinese labs also buy from U.S. firms like Surge; UniPat/Humanlaya would struggle to sell to U.S. labs given China’s outbound data-transfer rules. UniPat: >100 employees in Beijing (~half researchers from Chinese AI labs); planning Bencher expert platform later this month. Humanlaya: ~80 employees (~half domain experts); TalentsAI platform claims >100,000 expert sign-ups across 300 industries (exam still required). Both backed by HSG (formerly Sequoia Capital China). Labor-cost gap: e.g. U.S. medical doctor ~$150/hr vs Humanlaya ~$45/hr in China; Chinese market revenue for such training data is internally estimated at roughly a tenth of the U.S., with labs able to hire interns <$50/day for synthetic data—so external spend is still warming up.
Why it matters
Named Chinese Surge/Mercor analogues with valuation, order-book, customer, founder, headcount and investor (HSG) detail—directly usable to create/update Dealroom profiles for UniPat and Humanlaya and to track China’s post-training data stack vs U.S. peers (Surge, Mercor, Scale).
Executive takeaways
- UniPat: ~$1B pre-money round; $30M orders end-May; co-founder Li Kuan (ex-Alibaba); >100 staff Beijing; Bencher platform imminent.
- Humanlaya: ~$1B post-money; co-founder Jasper Jiao (ex-ByteDance Xpert); ~80 staff; TalentsAI >100k expert sign-ups (exam-gated).
- Customers named: Alibaba, ByteDance, Moonshot, DeepSeek (and ByteDance both competitor via Xpert and customer).
- Both funded by HSG (ex-Sequoia China); Dimension Capital LP letter highlighted the cohort.
- Valuation caveat: $1B on $30M orders is a very high multiple; Chinese high-end data spend still ~1/10 U.S.; lower expert rates ($45 vs $150/hr doctors).
What The Information may be missing
Exact round sizes, lead investors beyond HSG, ownership %, revenue (not just orders), and which model versions/benchmarks drove wins are not disclosed. No side-by-side pricing vs Surge/Mercor contracts, nor confirmation whether ByteDance’s Xpert is scaling as a captive substitute. Dimension Capital’s LP letter is paraphrased, not quoted at length.
