You’ve probably noticed the headlines screaming about a bloodbath in Chinese AI. The narrative has always been simple: Chinese companies have no moat, so they’re trapped in a vicious price war, slashing API costs to fractions of a penny just to buy market share. It’s a race to the bottom, right? Dead wrong.
The narrative that Chinese AI is just a cheap, discount-driven volume play is officially obsolete.
Over the last eight months, something shocking happened. China’s top AI model companies saw their annualized revenue jump from $4 billion to nearly $13 billion. But they didn’t achieve this by dropping prices. They did it by raising them—aggressively. Zhipu AI doubled their average API prices, yet token usage skyrocketed 40x. DeepSeek hiked prices on certain tiers by up to 12x, and usage barely flinched.
But the real twist isn’t just the pricing power. It’s where the money is coming from. MiniMax now pulls over 60% of its revenue from overseas. Kimi’s international earnings just surpassed its domestic take. Zhipu is hovering around 40% overseas. For the first time in tech history, China has built a genuinely massive software export engine.
For the first time, Chinese software companies aren’t just competing on cost—they are exporting high-value intelligence to the rest of the world.
Yet, if you look at how investors are valuing these companies, you’d think they were all playing the exact same game. They aren’t. The strategic divergence among Chinese AI players is now so extreme that lumping them together as “Chinese large models” is financially reckless.
Take DeepSeek. Rumors peg their valuation at $74 billion against a $1 billion ARR—a 70x multiple. Zhipu sits at a 36x multiple, while MiniMax trades at 20x. Why the massive spread? Because revenue quality dictates the multiple.
DeepSeek isn’t just another API vendor; it operates like a high-margin infrastructure utility. In the first seven months of the year, DeepSeek posted a staggering 82.9% gross margin—beating even Anthropic’s 60%. Meanwhile, Zhipu’s API margin sits at a measly 24.6%.
A dollar of DeepSeek’s revenue is worth four times a dollar of Zhipu’s revenue. ARR is no longer an equalizer; it’s a mirror reflecting the brutal reality of margin durability.
This happens because Chinese companies don’t have the infinite capital reserves of an OpenAI or Anthropic to pivot on a dime. They had to place early bets. Zhipu bet on enterprise APIs and domestic compute. MiniMax went all-in on multimodal and overseas markets. DeepSeek focused on structural efficiency. When the market shifts, these companies can’t just throw billions at the new trend—they are locked into their early strategic bets.
For product leaders, investors, and AI strategists, the writing is on the wall. The next phase of AI monetization will not be won by whoever slashes prices the hardest. It will be won by whoever commands pricing power, captures the overseas mix, and defends a gross margin that actually justifies their existence.
The era of blind ARR multiples is dead. Revenue quality is the only truth left.
FAQ
Q: Isn't this AI revenue boom just an unsustainable bubble fueled by enterprise FOMO?
A: No, because the growth is happening alongside massive price increases and widening margins. If it were pure FOMO, companies would be demanding discounts. Instead, they are paying 12x more for DeepSeek's API because the underlying model capability has become a non-negotiable utility.
Q: How should investors adjust their valuation models for Chinese AI companies?
A: Stop using a uniform ARR multiple. You have to segment companies by strategic optionalities and gross margin durability. A 20x multiple on a 24% margin API business is vastly different from a 70x multiple on an 82% margin infrastructure utility.
Q: Is the overseas expansion of Chinese AI actually a sign of domestic weakness?
A: Actually, it's the exact opposite. It proves that Chinese AI has finally broken out of the domestic price-war cage. By capturing international developer mindshare and exporting software globally, they are proving their models have world-class utility, not just localized cheapness.