Goldman Sachs Upgrades Zhipu Rating to Buy with Target Price of HK$1,560
nashnova research
Goldman Sachs upgraded Zhipu (02513.HK) from Neutral to Buy with a DCF-based target of HK$1,560, raising its year-end 2026 ARR forecast to US$3.2 billion — above the company's own US$3 billion goal — signaling that this Chinese AI-model company is monetizing faster than the market expected.
What did Goldman see to upgrade at this level?
The core trigger is a clearer monetization path: GLM-5.5 and GLM-6 are scaling to larger pre-training parameter sizes, and the data flywheel — a virtuous cycle where more usage feeds more data feeds better performance — is starting to turn.
This means → Zhipu is crossing from "burning cash on training" into "model capability directly converting to revenue." Goldman raised its year-end 2026 ARR forecast from US$2.7 billion to US$3.2 billion, above the company's own US$3 billion target.
Two drivers: strong token demand, plus new commercial terms with Chinese and global hyperscale cloud providers effective from October 2026, bringing incremental high-margin revenue.
Where is the money coming from, and where is it going?
Zhipu has raised close to US$10 billion in new equity financing year-to-date, including IPO proceeds, significantly strengthening its balance sheet.
The spending plan is clear: annualized R&D expenditure is projected at US$1.3 billion for H2 2026 and US$1.9 billion for 2027, all converted into compute capacity — 50% for training, 50% for inference.
In plain terms = nearly every dollar raised is going straight into compute expansion, split evenly between training and inference, betting that the chain of "more compute → stronger models → more revenue" will hold.
How big is the profit opportunity in inference?
Goldman believes Zhipu's model-pricing power combined with a cost-efficient inference architecture can drive sustained inference gross-margin expansion.
This means → inference is shifting from a cost center to a profit engine — Goldman projects Zhipu could reach group-level profitability by 2029.
A supporting factor is the harness/Co-Work product line: as context windows expand and cost efficiency improves, Chinese AI models have reached the capability level needed to deliver virtual employees and white-collar task automation at competitive cost.
Is the valuation expensive? How does it compare to peers?
Zhipu currently trades at roughly 12× year-end 2026 ARR / 10× 2027 ARR, compared with MiniMax at 9× / 6×.
Goldman sees the premium as justified: Zhipu ranks first among Chinese AI-model companies by ARR run rate and ships both frontier models and smaller flash models, giving it a leading competitive position.
Goldman's three-scenario valuations: base case US$98 billion, bull case US$157 billion, bear case US$32 billion — implying +149% upside to the target price, a bull-case price of HK$2,500, and a bear-case price of HK$510.
Where are the risks? What would make this call wrong?
Goldman explicitly flags four risks: intensifying competition among Chinese AI models, market concerns around ZCode data retention, shareholder dilution from roughly US$9 billion in equity financing since July 2026, and pre-IPO share lock-up expiry pressure starting early January 2027.
This reflects a core tension: the fundraising fuels compute expansion, but it also dilutes existing shareholders — whether the math works depends on the pace of ARR acceleration.
In plain terms = Goldman's logic is "revenue grows fast enough to outrun dilution," but if monetization disappoints, lock-up expiry pressure and rising competition will squeeze the stock from both sides.
市场有风险，内容仅供研究参考，不构成投资建议。