Another mega-round, but this time the numbers do the talking.

On September 13, Zhipu closed a ~$5 billion funding round. The deal has two parts: ~$2 billion in share placements (priced at HK$714/share, roughly 10% discount to last trade) and ~$3 billion in zero-interest convertible bonds (conversion price HK$892.50, ~25% premium, maturing September 2027). Just months after the previous ~$4 billion round, Zhipu's capital market "blood-making speed" is turning heads.

Revenue Numbers: From Storytelling to Math

What actually gives this raise credibility is the operating data Zhipu disclosed alongside it.

In H1 2025, Zhipu's MaaS (Model-as-a-Service) revenue hit 825 million yuan, a 2,736% year-over-year increase. What does that number mean? A year ago, MaaS revenue was negligible; now it's Zhipu's core revenue engine.

The more critical metric: gross margin turned positive for the first time at 24.6%. LLM companies burning cash is the norm — a positive gross margin signals Zhipu's commercialization path is shifting from "subsidized growth" to "sustainable business." During the same period, token usage grew 40x, while average API pricing rose 101% — users aren't just consuming more tokens, they're willing to pay higher prices for better models. ARR has reached $1.6 billion.

Deal Structure: The Zero-Interest CB Signal

The structure of this deal deserves a closer look.

The $3 billion zero-interest convertible bond means investors collect no interest until maturity — they're purely betting that Zhipu's share price rises above HK$892.50 for conversion. The 25% premium threshold signals: investors believe Zhipu's stock has at least 25% upside, and this bet must pay off before September 2027.

Zero-interest CBs are the preferred financing tool for high-growth companies — they don't dilute current shareholders, add no interest burden, but give investors upside exposure. Zhipu choosing this structure signals confidence in its own growth, and hints that the next round or IPO may already be on the horizon.

Where the Money Goes: Self-Training and Compute

Zhipu says proceeds will fund three directions: next-gen GLM model R&D, a fully self-training pipeline, and compute infrastructure expansion.

The "fully self-training pipeline" is the key item. It means Zhipu is reducing reliance on external data and training resources, building end-to-end autonomous capability from data cleaning through pre-training, post-training, and deployment. In the current geopolitical climate, this "fully autonomous" approach isn't just a technical choice — it's a survival strategy.

Just months after the last $4B round, Zhipu's capital market cadence is accelerating. When revenue data starts backing the valuation story, the LLM race shifts from "who raises more" to "who spends smarter."

See you tomorrow.