Zero Coupons, Full Commitment: Z.AI's $5 Billion Raise Bets the Balance Sheet on GLM

On Friday the stock closed at HK$793. On Monday morning, Z.AI — the Beijing-based developer of the GLM model family, formerly known as Zhipu AI — told the market it had completed one of the largest capital raises ever executed by a Chinese AI company after its IPO: roughly US$5 billion, assembled from two concurrent instruments filed in Hong Kong.

The mechanics are straightforward and, to anyone who has followed this year’s AI financing wave, instantly recognizable. The company placed approximately 21.97 million new H shares at HK$714 each — a 9.96% discount to Friday’s close — raising about US$2 billion. On top of that, it sold US$3.01 billion of zero-coupon convertible bonds, yuan-denominated but settled in US dollars, issued at 100% to 100.5% of face value, which works out to a yield between negative 0.5% and zero. Holders can convert into roughly 26.37 million shares at HK$892.50 per share — a 25% premium to the placement price and about 12.6% above where the stock traded before the deal leaked.

Investors, in other words, are paying Z.AI for the privilege of lending to it, in exchange for a shot at equity upside if the GLM franchise keeps compounding.

What the money is for

According to the filing and the company’s own statements, about 60% of net proceeds will fund research and development of next-generation GLM models and Z.AI’s fully self-training infrastructure. The remainder targets computing capacity, expansion, and potential acquisitions.

That allocation tells you where the pressure sits. Frontier model training has become a capital furnace — not just for the GPUs themselves, but for the multi-year power contracts, data pipelines, and reinforcement-learning infrastructure wrapped around them. Z.AI’s GLM-5.3, released in August, was billed as the most capable open-weights coding model available, with a 50% improvement over GLM-5.2 on the company’s in-house code bench. Staying on that trajectory requires training runs that cost more than most companies’ entire R&D budgets.

The second raise in two months — and the selloff

This is where the story loses its shine. Z.AI went public in Hong Kong in January and pulled in about $4 billion from a follow-on share sale in July. This $5 billion raise is its second mega-financing in roughly eight weeks, and the market’s verdict was swift: shares tumbled more than 10% on Monday, hitting a five-and-a-half-month low. By midweek the stock hovered near HK$668 — down roughly 73% from its post-IPO peak, compressing a market cap that had briefly flirted with $128 billion during the summer’s open-weights euphoria.

The dilution math explains the pain. Between the placement and full conversion of the bonds, Z.AI is adding nearly 48 million new shares — around a tenth of its outstanding base — within months of a prior raise of similar scale. Retail holders who bought the IPO narrative are now holding a stake that has been diluted twice before the ink on the first raise dried.

The fundamentals underneath the fundraising

The uncomfortable backdrop is that Z.AI’s revenue, while growing explosively, remains tiny against its capital consumption. Its most recent interim report showed six-month revenue of 953.89 million yuan (about US$142 million) — up nearly five-fold year-over-year — with net losses narrowing to 2.07 billion yuan. Bloomberg reported in July that the company was on track to become the first independent Chinese AI developer to hit $1 billion in annualized revenue, though its August results missed lofty analyst estimates as China’s AI price war bit into margins.

A $5 billion raise against a $142 million half-year revenue base is not a financing; it’s a wager. Z.AI is effectively pre-funding several years of frontier-scale training runs on the expectation that GLM’s open-weights ecosystem, enterprise contracts (the GLM Enterprise 100 program), and its aggressive API pricing keep pulling it up the revenue curve faster than the cash burns.

Why zero-coupon convertibles are the AI-sector tell

Z.AI is hardly alone in discovering the convertible bond. The instrument has become the signature financing tool of the 2026 AI buildout — Nebius raised $3.75 billion via convertible loan in March, and US convertible issuance hit roughly $34 billion in the first four months of the year, more than double the prior-year pace, before accelerating through the summer. Alibaba, for its part, announced an HK$80 billion (US$10.2 billion) share placement in August, with all proceeds earmarked for AI.

For companies with volatile, narrative-driven stock prices, zero-coupon convertibles are near-perfect instruments: they impose no cash interest burden during the heaviest capex years, and if the equity story works out, bondholders convert and the debt evaporates into dilution the market has already priced. If the story doesn’t work out, the company has taken in $3 billion of essentially free capital that must eventually be repaid — a problem for a future management team.

The terms here are notably aggressive even by the sector’s standards. Z.AI can redeem all — but not part — of the bonds starting February 18, 2027, if shares trade at or above 130% of the conversion price for 20 out of 30 trading days. That clause is a bet within a bet: the company is wagering that GLM’s next generation forces the stock up more than 30% above HK$892.50 within eighteen months, letting it retire the debt cheaply while keeping the equity story intact.

The open question

Strip away the deal mechanics and the strategic picture is this: China’s most valuable public AI pure-play is trading down 73% from its peak while simultaneously raising capital at a pace that signals either confidence or desperation, depending on your prior. The GLM-5.3 release cycle proved Z.AI can ship frontier-adjacent open weights that the global developer community actually uses. The $1 billion ARR trajectory proves enterprises will pay.

What no financing document can prove is whether the next training run — the one this $5 billion funds — produces a model compelling enough to justify the roughly $130 billion market cap investors briefly ascribed to the company this summer. The bondholders who accepted a negative-to-zero yield have already voted with their wallets. The equity market, selling the stock down 10% on announcement, has voted the other way. Over the next eighteen months, one of those votes gets settled.