JPMorgan Raises Zhipu Target Price to HK$1,800; DeepSeek Price Hike Reshapes China's AI Competitive Landscape

Nashnova编辑部
Published todayAbout 13 min read

JPMorgan raised Zhipu's target price from HK$1,600 to HK$1,800 and lifted MiniMax to HK$260. This means → in China's AI model war, pricing power built on frontier intelligence is replacing low-cost strategies as the dividing line for investment value.

01

Why did JPMorgan adjust both companies at the same time?

In its August 16 note, JPMorgan raised Zhipu's target to HK$1,800 (maintaining Overweight) and MiniMax's target to HK$260 (maintaining Neutral).
The shared catalyst: Zhipu's release of GLM-5.3 and DeepSeek's API price hikes effective August 17.
This means → JPMorgan's core call is not "who is cheaper" but who holds pricing power at the frontier — only models strong enough to command prices can build sustainable businesses.
02

What is the "Pareto frontier" and why does JPMorgan use it here?

The Pareto frontier — a framework that asks "can anyone deliver the same capability at a lower price?" — is JPMorgan's primary tool for mapping China's AI model landscape.
In plain terms = if no rival can match a model's capability at equal or lower cost, that model sits on the Pareto frontier — safe from both performance comparisons and price wars.
This reflects JPMorgan's underlying investment logic: not who spends the most on training, but whose capability-price combination is hardest to displace.
03

What makes Zhipu's GLM-5.3 strong? How is the moat built?

GLM-5.3 uses the same base model as GLM-5.2. Gains in coding and agentic capability come primarily from reinforced post-training, not a new round of large-scale pre-training.
This means → the competitive edge shifts from "who has the biggest training set" to data quality, reinforcement learning, and engineering execution — capabilities far harder to replicate.
API pricing is essentially unchanged (input ¥8.00, cache-hit input ¥2.00, output ¥28.00 per million tokens). Stronger capability at the same price should lift adoption and retention. JPMorgan raised Zhipu's 2026–2030 revenue forecasts by 6%–10%.
04

How large are DeepSeek's price hikes? What do they signal?

V4 Pro peak hours: input rose from ¥3.00 to ¥9.00, output from ¥6.00 to ¥27.00. V4 Flash peak hours: input from ¥1.00 to ¥3.00, output from ¥2.00 to ¥9.00.
Short-term impact: cost-sensitive players — especially MiniMax — get breathing room as the cost gap narrows.
Yet JPMorgan argues DeepSeek's system-level efficiencies — its MoE architecture (a design that activates only part of the model's parameters to save compute), attention design, and KV-cache optimisation — remain its moat. In plain terms = DeepSeek raised prices, but it is still the industry's cost benchmark — rivals' cost structures have not fundamentally improved.
05

Where does MiniMax stand now? Why is M3.1 the make-or-break moment?

JPMorgan states plainly: MiniMax's current M3 model has not established a clear edge on either capability or cost-efficiency, leaving it squeezed between stronger models and DeepSeek's low-price strategy.
The breathing room from DeepSeek's price hike comes from a competitor's decision and could reverse at any time — it is not a self-built moat.
JPMorgan frames M3.1 as MiniMax's most important catalyst: it must either meaningfully improve capability or deliver standout cost-efficiency — claiming a position on either axis of the Pareto curve. A modest upgrade would have limited impact on the long-term thesis.
06

What drives the valuation gap between the two?

Zhipu's HK$1,800 target is based on a 20x 2030E P/E, discounted to December 2026 at a 15% WACC. The 20x premium reflects an expected 2026–2030 revenue CAGR of over 100%.
MiniMax's HK$260 target also uses a 20x 2030E P/E. JPMorgan raised its 2027–2030 revenue forecasts by 11%–21%, but remains cautious about MiniMax's ability to capture value as a standalone vendor — integrated platforms like ByteDance and Kuaishou can monetise across model, content, distribution, and advertising, while MiniMax relies more heavily on direct product and model-service revenue.
This reflects JPMorgan's overarching conclusion: endogenous capability improvement is worth more than passive gains from a friendlier environment. The valuation divergence between Zhipu and MiniMax is a direct expression of that judgment.

Content is for reference only, not financial advice.