The numbers hit the timeline like a bomb: five Chinese AI startups—Zhipu, DeepSeek, Kling, Moonshot, MiniMax—collectively pulling in an estimated $2.6 billion in annual revenue. A chorus of bullish tweets erupted, painting a picture of a Chinese AI sector finally stepping out of the lab and into the market. But as a narrative hunter, I don't see a success story. I see a ghost—a revenue ghost, haunting a space where unit economics are sacrificed for narrative dominance, and where the real story is buried beneath a layer of optimistic estimates and strategic leaks.
Chasing the ghost in the blockchain's gray matter, I recognize the pattern. It's the same one we saw during the 2017 ICO boom: token prices skyrocketing on promises of world-changing adoption, while the underlying product was still a whitepaper. Here, the token is replaced by the API endpoint, the whitepaper by the press release. The narrative of 'Chinese AI is real' is being manufactured through carefully leaked revenue figures, sourced from a Menlo Ventures partner’s back-of-the-napkin calculation. Let's treat this not as a financial disclosure, but as a narrative artifact. We need to dissect it with forensic precision.
Context: The Historical Narrative Cycles
Two years ago, the dominant narrative was 'China is behind in AI.' Then came DeepSeek, an open-source model from a quant hedge fund spin-off, and suddenly the narrative flipped to 'China is overtaking.' The story hinged on capability: benchmarks, model size, inference speed. But capability narratives have a short shelf life. They require constant validation through new tests. The smarter play is to shift the narrative to market penetration—revenue. Revenue is sticky, tangible, and reassuring to investors. By injecting these $2.6B estimates into the discourse, the narrative engineers are attempting to move the conversation from 'Wow, their model can write a poem' to 'Wow, their model can write a check.'
Where code meets the human heartbeat: The human heart wants stories of victory. The Chinese AI narrative is perfectly tailored to that desire. But we must ask—how much of this revenue is real recurring business, and how much is a one-time government contract, a cloud subsidy, or internal transfer pricing? From my experience tracking the 'narrative debt' in crypto projects—where hype outran actual usage—I've learned that revenue narratives are the easiest to fabricate without real-world verification.
Core: The Narrative Mechanism and Sentiment Analysis
The core insight here is not the $2.6B itself, but the emotional protocol it triggers. The number is ambiguous enough to be exciting but vague enough to never be disproven. Let's break down each startup's revenue component through a narrative lens:
- Zhipu (est. $1B): Tied to the Chinese government and state-owned enterprises. Revenue likely from custom deployments and AI 'new infrastructure' projects. Revenue resilience is high, but growth is capped by bureaucratic timelines. The narrative here is 'trusted partner for national AI.'
- DeepSeek (est. $500M): The champion of open-source and extreme pricing. Their API costs a fraction of OpenAI's. To hit $500M, they'd need enormous volume. But at those margins, the unit economics are bleeding. The narrative is 'disruption through cost,' but the reality is a subsidy-fueled land grab.
- Kling ($500M): A video generation model from Kuaishou. This revenue likely includes internal usage for Kuaishou's own content creation. It's not a pure product company. The narrative is 'new medium for creators,' but it's really a cost center turned into a revenue line.
- MiniMax ($400M): Focused on multimodal and social. Their revenue is opaque—likely from API calls and some app subscription. The narrative is 'next-gen social AI,' but traction is unverified.
- Moonshot ($200M): Their Kimi model is famous for long-context processing. $200M in revenue suggests decent enterprise adoption, but still early. The narrative is 'the best context window in China.'
Reading the invisible signals of digital identity: Each revenue stream carries a different narrative weight. Zhipu's is political, DeepSeek's is populist, Kling's is internal, Moonshot's is technical. But none of these narratives address the fundamental question: Is the business model sustainable, or is it a narrative debt waiting to collapse?
Contrarian: The Contrarian Narrative – The Revenue is a Mirage
The contrarian angle is not that the numbers are wrong (though they likely are), but that the entire framework of 'revenue as validation' is flawed in the current AI landscape. We've seen this in blockchain: a DeFi protocol can generate $1B in trading fees, but if 90% of that is wash trading by its own team to inflate metrics, the narrative is hollow. Here, the revenue metrics are equally fragile.
Unraveling the tapestry of digital mythologies: Consider the unit economics of DeepSeek. At $0.5 per million input tokens (versus OpenAI's $15), DeepSeek needs 30x the volume to match OpenAI's revenue. Training costs are similar across models. So DeepSeek is essentially burning capital to acquire market share. The narrative of 'successful commercialization' masks a potential Ponzi-like dynamic: later-stage investors subsidizing earlier customers, with no clear path to profitability.
Moreover, the $2.6B aggregate may include double-counting if any of these companies trade among themselves (e.g., Zhipu buys model services from DeepSeek). Without a blockchain to trace the cash flows, these narratives are unverifiable.
Takeaway: The Next Narrative to Watch
Narratives don't die; they just replace themselves. The next narrative will pivot from raw revenue to unit economics. Watch for these startups to start touting 'gross margin' and 'customer retention rates.' The signal to look for is whether they can attract real, paying enterprise clients outside of government and related-party transactions. The ultimate test is when they have to survive without access to cheap compute or favorable policy.
The artifact holds the memory we forgot: We forgot that in the early days of crypto, projects celebrated 'total value locked' as a sign of health, only to see it evaporate when incentives were removed. The same will happen here if the narrative hygiene isn't maintained. The ghost in the revenue is the assumption that growth equals sustainability. It doesn't. Not in crypto, not in AI.
As we watch the narrative unfold, the only reliable compass is data that is on-chain—or in this case, on the balance sheet and audited. Until then, treat every $2.6B as a work of speculative fiction. The real question isn't how much they earned, but how many of those dollars came from someone who actually needed the product, not just a narrative.
