AI Startup Economics: Zhipu and MiniMax Losses Exceed Revenue by 3–6 Times
Zhipu AI and MiniMax listed on the Hong Kong Stock Exchange in January 2026, becoming the first publicly traded developers of foundational AI models. In 2025, Zhipu boosted revenue by 132% to $105 million, but net losses surged 60% to $680 million—a loss-to-revenue ratio of 6.5:1. MiniMax saw revenue grow 159% to $79 million with adjusted losses of $251 million, around 3.2:1.
These figures reveal the true cost of scaling AI without the veil of private funding rounds. The main driver: investments in R&D, which swallow the lion's share of the budget.
Expense Structure: R&D Dominates
Zhipu's R&D spending hit 3.18 billion yuan ($460 million)—4.4 times its revenue. The R&D-to-revenue ratio of 8.4:1 is far worse than competitors':
- OpenAI: ~1.6:1
- Anthropic: ~1:1
Key cost items:
- Renting compute power for model training.
- Expanding the developer team.
MiniMax optimized its infrastructure, lifting gross margins from 12% to 25%. Still, the overall cost structure is similar: priority on compute and personnel, not marketing or operations.
In a cash crunch, the companies went public out of necessity. Zhipu was burning $40 million monthly with 6–7 months of runway left. MiniMax was spending even faster. The IPO raised $558 million for Zhipu and $600 million for MiniMax, extending their 'runway'.
Price Wars and Margins
Zhipu's gross margin fell from 56% to 41% due to growth in cloud business and dumping by competitors like ByteDance and Alibaba. Pricing examples:
- Zhipu coding tool — 20 yuan/month (~$3).
- Anthropic's Claude — 7 times more expensive.
Zhipu Chairman Liu Debing predicts U.S. players will join the price race. This ramps up margin pressure and accelerates cash burn.
Transparency as a Benchmark for the Industry
Going public exposed the true costs of foundational AI models. Private firms like OpenAI and Anthropic disclose data selectively. Market reaction:
- Zhipu shares: -5.45% on report day.
- MiniMax shares: +500% since IPO.
Investors focus on growth potential, ignoring current losses. The reports highlight the gap between model benchmarks and sustainable business.
Key Takeaways:
- Zhipu's R&D is 8.4 times revenue—the main loss driver.
- IPO as a last resort: 6–7 months of cash.
- Dumping cuts margins to 41%, sparking a price race.
- Public data serves as a benchmark for the entire AI industry.
- MiniMax boosted margins to 25% via infrastructure.
These cases illustrate the challenges of monetizing frontier models: high GPU CAPEX, low API prices, and the need for constant iterations.
— Editorial Team
No comments yet.