Hook
$700 million raised. $2.7 billion valuation. Zero benchmark scores published. Zero customer revenue disclosed. Zero mention of model architecture.
That’s not a technical company—that’s a capital allocation vehicle dressed in AI clothing. And the market is buying it.
Baichuan Intelligence, founded by former Sogou CEO Wang Xiaochuan in 2023, just closed what it calls an A round. In any other industry, a $700M A round would be an anomaly. In China’s AI gold rush, it’s now the norm. The company also announced plans for an IPO by 2027—a timeline that aligns perfectly with the estimated cash burn rate of $1-2 million per month that a 7-figure GPU cluster requires.
Let’s be clear: this is not a technology story. It is a capital story. And as someone who audited The DAO in 2016 and watched the Ethereum panic sell, I’ve learned that the loudest funding rounds often mask the weakest technical foundations.
— Root: Auditing the DAO and Ethereum
Context
Baichuan AI is one of half a dozen Chinese large language model (LLM) startups that emerged after ChatGPT’s launch. The company initially open-sourced its Baichuan 2 series (7B, 13B, 53B parameters) to build developer mindshare, then switched to a closed-source model with Baichuan 3—reportedly hundreds of billions of parameters. This open-to-closed pivot mirrors the strategy of almost every Chinese AI startup: use open source to attract talent and ecosystem, then lock the product behind an API paywall.
Today, Baichuan’s valuation sits at approximately $2.7B, placing it alongside Moonshot AI (Kimi) and Zhipu AI in the first tier of Chinese AI unicorns. Its investors include Alibaba, Tencent, and Xiaomi—strategic backers who can offer cloud compute or distribution channels. But here’s the rub: Moonshot AI has a consumer product with 20+ million monthly active users. Zhipu AI has government contracts with State Grid and China Mobile. Baichuan has… a press release.
Core: The Funding Is Real. The Technology Is Black Box.
Let’s dissect what the news does and does not say.
First, the numbers. $700 million is a massive injection. At typical Chinese AI startup burn rates of $10-20M per month (compute + talent + salaries), this gives Baichuan roughly 2.5 to 4 years of runway—exactly enough to reach the claimed 2027 IPO target. The investors are betting that by then, the company will have built enough revenue to justify a public listing. But revenue from where?
The article mentions no customer contracts, no API pricing, no annual recurring revenue (ARR). Not even a single enterprise deal worth over ¥1M. Compare this to the transparency of public AI companies like OpenAI (reported $2B ARR in 2024) or even Anthropic ($1B+ run rate). Baichuan is operating in a data vacuum.
Second, the technical gap. Independent benchmarks like OpenCompass, MMLU, and C-Eval consistently rank Baichuan models below GPT-4, Claude 3, and Zhipu’s GLM-4. In some tests, newer models from DeepSeek (a cheaper, open-source alternative) have surpassed Baichuan’s flagship. The company has not published scores for its latest version, Baichuan 3. When you stop releasing benchmarks, you are hiding something.
Third, the compute dependency. A $700M raise implies a significant portion will go to GPU clusters. But China is under US export controls on H100/H800 chips. Baichuan likely relies on Alibaba Cloud for compute, and may have purchased A800 or domestic Huawei Ascend 910B chips. The problem is that these chips have reduced memory bandwidth and slower interconnects, making large-scale training less efficient. Without self-owned data centers or guaranteed capacity, Baichuan’s training costs are higher and scalability is constrained.
— Root: Auditing the DAO and Ethereum
Contrarian: This Is Not About AI Leadership—It’s About Exit Strategy
Every bull market has its narrative. In 2017, it was ICO whitepapers. In 2021, it was DeFi yield farming. In 2024-2025, it’s Chinese AI startups. The common denominator? Investors are buying stories, not code.
Baichuan’s 2027 IPO plan is the most telling detail. It signals that the current investors—Alibaba, Tencent, etc.—need an exit. They are not long-term believers in open-source AI; they are betting that a public market will absorb their shares at a premium. This is the same pattern we saw with “startups” that raised billions for electric vehicles or biotech. The technology might work, but the primary goal is to create an exit vehicle.
We farmed the yields until the protocol farmed us. In this case, venture capitalists are farming retail IPO buyers.
Furthermore, the narrative that “China is leading in AI” is convenient for marketing, but the reality is that Chinese AI companies face severe regulatory constraints. The Cyberspace Administration requires all public-facing models to pass safety reviews (algorithms filing, generative AI registration). Any compliance slip—a politically incorrect response, a data leak—can suspend the service and delay the IPO. Baichuan has not publicly disclosed its safety alignment methods (RLHF, DPO, red teaming). That’s a regulatory time bomb.
— Root: Auditing the DAO and Ethereum
Takeaway: Actionable Price Levels for Your Attention
If you are an institutional investor or a crypto-native fund manager watching this space, treat Baichuan’s $700M A round as a signal, not a buy order.
Here’s what to watch:
- Model performance: If Baichuan releases open benchmarks for Baichuan 3 (MMLU >85, HumanEval >70, long-context >128K), the technology story gains credibility. If they stay silent, assume the gap is widening.
- Enterprise revenue: Any disclosure of >10 enterprise customers with ¥1B+ TCUs would validate the business model. Right now, it’s zero.
- Compute infrastructure: Announcements of self-owned data centers or binding GPU leases (especially with Huawei or Alibaba) would indicate they’ve solved the compute bottleneck. Otherwise, they are renting by the hour, and margins will be thin.
- Regulatory clearance: Confirm whether Baichuan has obtained the official “Algorithm Filing” and “Generative AI Service Filing” from the CAC. Without those, IPO is impossible.
The contrarian play is not to short Baichuan—that’s not possible yet. It’s to short the narrative that massive funding equals technological dominance. In a market where capital is abundant, the real competitive advantage is code that works, models that outperform, and customers who pay. Baichuan has proven the first. The other two remain under a black box.
— Root: Auditing the DAO and Ethereum