JPMorgan Raises Zhipu Target Price to HK$2,000
Alina Collins
JPMorgan raised Zhipu's target price from HK$1,800 to HK$2,000 while cutting MiniMax's from HK$400 to HK$300 — the core logic is that open-weight monetization hinges on model quality, and the winner takes most.
What is JPMorgan's "winner-takes-most" framework really saying?
Open-weight models — releasing model parameters so anyone can download and deploy them — live or die by model quality when it comes to monetization.
This means → a capability leader can give its weights away and still pull users back to the paid official channel; a weaker model gets dragged into price wars and traffic diversion.
The market reads "open weights" as revenue leakage. JPMorgan says that is only half right — official APIs retain systematic advantages in latency, caching, throughput, feature support, and reliability. An open-weight release is a published checkpoint; the official endpoint is a continuously evolving product.
How much cheaper is the official channel than third parties?
The report uses DeepSeek V4 Pro as an example: the official path, with lower list pricing and a 93.5% cache hit rate, costs roughly $24–41 per month for a standard workload.
Some third-party routes run $85–196, a gap of roughly 6–12×.
In plain terms = for the same model, the official route can cost one-tenth of a third party — that is the mechanism behind "even though the model is open, money flows back to the provider."
Why was Zhipu's target price raised?
The core reason is that GLM-5.2 cemented Zhipu's position at the frontier, validating JPMorgan's thesis that open-weight commercialization creates meaningful option value for leading model providers.
Zhipu's strategy: use permissive access to scale adoption, while positioning the official path and premium tiers (such as the GLM-Turbo series) for quality-sensitive demand.
GLM-5.2 ships under an MIT license — one of the most permissive open-source terms — and is already distributed on AWS and Microsoft Azure, with coverage still expanding.
How much upside is left for Zhipu?
JPMorgan notes the market has largely priced in Zhipu's year-end $1 billion ARR (annual recurring revenue) guidance.
This means → remaining upside depends not on expanding Zhipu's own GPU stack, but on whether open-weight models can scale through external cloud platforms and distribution channels.
Key watch point: how GLM-5.5/6 benchmarks against Kimi K3 and DeepSeek V4.1 — whichever model proves stronger converts open distribution into paid inflow.
Why was MiniMax's target price cut?
The core issue: M3 has not yet shown sufficient evidence of model-driven pricing power. JPMorgan treats M3's permanent 50% discount as a telling signal.
This means → M3 has not earned a capability premium against China's leading models. The discount props up short-term volume but erodes market confidence in monetization.
MiniMax is not without strengths — it retains relevance in multimodal AI, overseas adoption, and agentic workflows, and OpenRouter call volumes show meaningful developer uptake. But JPMorgan argues that workflow monetization requires stronger model pull; coding or agent products must deliver a large enough improvement in task completion to shift user habits.
What risk do both companies share?
Both are in a capital-intensive phase. JPMorgan expects each to raise two more rounds in 2026 and 2027.
Faster model iteration, larger overseas deployments, or higher-than-expected inference costs could all force them to seek additional outside capital.
In plain terms = the model race is a cash-burning game. Whether Zhipu can maintain its model lead is the decisive prerequisite for its "winner-takes-most" option value to materialize; if MiniMax cannot close the quality gap, its funding pressure only grows.
Content is for reference only, not financial advice.