Chinese startup Zhipu AI reaches $14m in ARR

Zhipu, a Beijing-based AI startup founded by Tsinghua University researchers, is now making over 100 million yuan (US$14 million) in annual recurring revenue from its AI development tools, according to co-founder Zhang Peng.

The company’s overall revenue reached US$42 million in 2024, with a person familiar with the matter saying sales are expected to more than double in 2025.

Zhipu’s subscription-based tools, including its GLM-4.5 model released in July, reportedly saw a tenfold increase in usage within two months, and its API platform now has more than 2.7 million paying customers.

The company introduced a coding subscription plan in September starting at 20 yuan per month, which has attracted over 150,000 users, according to a spokesperson.

Backed by investors such as Alibaba and Tencent, Zhipu hopes to go public and is trying to compete with global AI leaders OpenAI and Anthropic.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

Subscription revenue alone doesn’t prove GLM-4.5 can compete with frontier models

Zhipu reports 100 million yuan in recurring revenue and 2.7 million paying customers yet no independent checks confirm GLM-4.5 reaches OpenAI or Anthropic on enterprise benchmarks, so parity claims with frontier models remain thin.
Across 12 tests on agentic (autonomous agent), reasoning, and coding within a 355 billion parameter set, GLM-4.5 ranks third 1. Claude Opus 4.5 leads with top scores on SWE-Bench Verified (software engineering task completion) and Terminal Bench 2.0 (command-line agent performance) 2.
On BrowseComp web tasks, GLM-4.5 reaches 26.4% accuracy vs o4-mini-high at 28.3% 1. For tool calling (structured function invocation to use external tools), GLM-4.5 posts 90.6% vs Claude-4-Sonnet at 89.5% 1, which signals competitiveness, not dominance.

Infrastructure providers can capture Chinese enterprise generative AI (GenAI) spending across models

With 83% of Chinese organizations using generative AI 3 and regulators clearing more than one AI model per day over six months 3, third-party infrastructure firms can win budget. Buyers want benchmarking, integration, routing, plus compliance tooling.
Chinese teams cite fewer hurdles, with 31% naming lack of tools vs 47% globally 3. Another 21% cite lack of expertise vs 39% globally 3 so buyers look ready to fund supporting platforms.
Adoption runs high yet 64% remain in experimentation 3. Globally, 55% have governance frameworks in development (policies and processes for safe AI use) with 7% at high training levels (workforce readiness) 3.
Near-term revenue sits in Information Technology (IT) and operations 4. Marketing, customer services, and cybersecurity follow as enterprises scale from trials to production 4.