Zhipu, MiniMax extend rally as AI optimism persists

Zhipu AI (HK:2513) jumped as much as 15% to a record HK$999.0 after releasing its GLM-5.1 model and raising model prices, while MiniMax (HK:0100) rose 5.4% to HK$1,053.0, both outpacing the Hang Seng's 0.3% decline. Zhipu is up ~650% YTD and MiniMax ~205% YTD as Chinese AI agent momentum and recent model releases from Meta and Anthropic bolster sector sentiment.

Analysis

Winners will be companies that control both model IP and the inference stack — that combination turns product hype into recurring revenue because it captures model price increases and reduces per-call cost. Second-order beneficiaries include GPU and HBM suppliers, cloud / colo providers serving inference workloads, and firms offering orchestration/agent management tooling; expect upstream demand to re-rate over a 6–18 month window as deployments shift from R&D to production.

Near-term price action is primarily a liquidity and sentiment story: momentum-fueled retail and margin-financed flows can extend rallies but also amplify intraday volatility. Key catalysts to watch over days–months are: (1) concrete metrics of monetization (ARPU per agent), (2) compute utilization and gross margin per model, and (3) policy actions — an export-control shock to high-end GPUs or tighter domestic data rules would shave obvious TAM and could compress multiples rapidly.

The crowd is pricing technological progress as synonymous with sustainable margins; that’s the weak link. Unit economics for agent products are still sample-dependent and highly sensitive to model size and context length; without clear evidence of structural margins (e.g., >40% gross margin on hosted agents) the current multiple expansion is fragile, creating both high-conviction long and short opportunities depending on evidence arrival over the next 3–12 months.

Key Decisions for Investors

• Long controlled exposure to model-to-inference winners: buy NVDA (or equivalent GPU exposure) or 9–12 month call spreads to capture the structural rise in inference demand. Rationale: 12-month horizon, asymmetric upside if compute scarcity persists; cap max loss to premium paid.
• Tactical long/hedged on selective Chinese AI names (e.g., leading IP owners) sized 2–4% NAV with strict risk controls: use 3–6 month call spreads or equity with a 20% stop. Pair with a short basket of momentum-driven small-cap HK AI names to neutralize broad market risk and isolate idiosyncratic execution risk.
• Event-driven short: buy 2–4 month put spreads on highly leveraged small-cap AI names after >30% intraday gap-ups or if margin-financing rates tighten materially. Rationale: high gamma events and retail exit windows often produce 30–60% drawdowns in these names.
• Long META (or US large-cap AI exposure) via 6–9 month call spreads to play durable monetization of generative features across ads/engagement. Keep position size modest and trim on a 25–35% realized move; monitor guidance for ARPU lift and agent monetization signals as the primary exit trigger.