Z.AI, formerly known as Zhipu AI, is the Chinese lab behind the GLM family of open-weight models. It is also, by any honest measure, a machine that needs cash the way a data center needs power. The week of September 13, 2026 it moved another $5 billion through Hong Kong, part share placement and part convertible bond.
The $5 Billion Capital Machine Actually Happened
On paper, Z.AI just threw one of the biggest post-IPO funding moves in Chinese AI history onto the table. In a filing with the Hong Kong Stock Exchange on September 13, the Beijing lab behind the GLM models said it raised a combined $5 billion: roughly $2 billion from a placement of new Hong Kong-listed shares and about $3 billion from a concurrent convertible bond sale. Read the headline on its own and you would think China's answer to OpenAI just proved the skeptics wrong.
The problem is that a $5 billion raise has to be read in context, and the context is uncomfortable. The lab still loses billions every half. Its shares have fallen roughly 50 percent in under three months. And the cleanest way to understand this funding round is not as a vote of confidence, but as a machine that keeps turning because it has to.
- A placement of 21.97 million new Hong Kong shares at HK$714 (about $91) each, a 10 percent discount to the HK$793 Friday close.
- A zero-coupon convertible bond of 20.14 billion yuan (about $3 billion) due in September 2027, settled in U.S. dollars and issued at 100.5 percent of face value.
- An initial conversion price of HK$892.50, a 25 percent premium to the placement price, redeemable from February 18, 2027 once shares trade at or above 130 percent of conversion for 20 of 30 trading days.
Behind the Numbers, the Optics Turn Ugly
Strip the marketing gloss and this is the third trip to the capital markets in under a year. Z.AI listed in Hong Kong in January, raising around $4 billion. In July it placed another tranche, banking HK$31.4 billion by selling 19.78 million shares at HK$1,588 each. Now, roughly 60 days later, with the lock-up expired, it is back selling new shares at HK$714. Do the arithmetic: the same equity that fetched HK$1,588 a share in July is being marketed at less than half that in September.
The financials explain why. The lab's interim report showed revenue roughly quadrupling in the first half of 2026, but it still booked a loss on the order of two billion yuan. Analysts have trimmed valuation estimates, and the stock slid below HK$1,000 after six straight declines before this raise. A company that loses billions and keeps issuing discounted shares is not funding expansion from strength; it is funding survival from dilution.
- If the business is worth a 550 billion yuan valuation, why price fresh shares at a 10 percent discount to the last close rather than issuing at market?
- Why does a top line growing 400 percent still need a $5 billion injection three times in nine months?
- If the models are genuinely frontier, why is the flagship selling tokens on Alibaba's Tmall like phone top-ups, priced up to 1,078 yuan a month?
What the Proceeds Are For (and What They Are Not)
The company says roughly 60 percent of the net proceeds will fund research and development of next-generation models and its fully self-training system. That is the credible part: frontier inference and training are brutally expensive, and neutral on the question of profitability.
The rest is where a skeptical reader should slow down. A convertible bond maturing in September 2027 is a bill with a date on it. Z.AI is also pursuing a listing on Shanghai's Star Market, targeting up to 15 billion yuan, with regulatory tutoring done but no application publicly accepted. Every one of those channels is another tap on the same capital machine.
None of this means Z.AI is doomed. It could well be the lab that makes the compute bet pay off. But when a company raises $5 billion on the same week that it ships an open model and still sells every new share at a deepening discount, the honest way to read it is: the burn is real, the dilution is real, and the market is pricing in a gap between the valuation story and the unit economics. That gap does not disappear because the checks keep clearing.
For investors, the convertible bond itself deserves a hard look before anyone celebrates the headline number. Zero-coupon, dollar-settled, callable from February 2027: it is structured to convert, not to be repaid, and the 25 percent conversion premium only matters if the share price recovers from where the placement discounted it. Nothing in the structure protects holders from further slide, and the September 2027 maturity means the company must be worth more by then or the debt converts on worse terms.
None of this is to dismiss the genuine R&D ambitions. Fully self-training systems and next-generation GLM models are a credible use of research money. The skepticism is not about whether Z.AI can build models; it is about whether a forty-billion-dollar-plus capital structure built on relentless dilution can be repaid by a business that, on the latest interim numbers, has yet to turn a profit while funding it.
Comments