Zhipu shares jump on second AI price hike this year

Zhipu AI (HK:2513) shares jumped 16.2% to HK$904.50 after the company launched its GLM-5.1 model and raised access prices by 8%–17%; this is the second price increase since its IPO. Management reported 2025 revenue more than doubled, and shares are up nearly 600% year-to-date; peer MiniMax (HK:0100) rose 5.6% and is up ~200% YTD, signaling sector-wide optimism on Chinese AI monetization.

Analysis

The willingness of leading model providers to raise access prices signals durable monetization power for proprietary LLMs, but the clearest second-order beneficiary is the inference hardware and systems stack. Every incremental dollar charged for API access increases the break-even for cloud vs on-prem inference; large verticals (finance, gaming, search) will accelerate purchases of dedicated servers and appliance-grade networking to avoid recurring API spend, compressing time-to-revenue for hardware OEMs over 6–18 months.

Risks that can reverse the move are concentrated and time-bound: a rapid shift to high-quality open-source models or aggressive wholesale discounts from hyperscalers would compress provider ARPU within 3–9 months, while renewed US/China export controls on accelerator silicon would create short-term supply squeezes that spike hardware pricing but choke volume growth thereafter. Talent flow and data-sovereignty regulation in major Chinese provinces create asymmetric outcomes where domestic incumbents keep pricing power but international expansion stalls for years.

Consensus attention is on headline revenue acceleration; it misses margin mix and customer concentration — top enterprise accounts can account for 30–50% of incremental revenue but are also the first to negotiate lower rates or build internal alternatives. That creates an asymmetric payoff: hardware and systems names capture durable spend even if SaaS multiples derate, so positioning should favor capex-exposed suppliers with scalable service revenues and avoid pure-play API/consumption names that face elastic demand shocks.

Key Decisions for Investors

• Long SMCI (core equity or 9–12 month call spread): allocate 1–2% of portfolio to capture accelerated on-prem inference demand. Target +30–60% upside in 6–12 months if cloud-to-onprem migration accelerates; set a hard stop at -25% and trim on +40% gains.
• Buy 6–12 month SMCI calls financed with a smaller short position in high-multiple SaaS (replaceable by options if names unavailable): 2:1 skewed risk—expect hardware upside to outpace SaaS re-rating if API pricing persists. Position size 0.5–1% TIP risk, exit on signs of open-source LLM adoption accelerating (weekly GitHub/model deployment metrics spike).
• Event hedge: buy 3–6 month puts on exposure to China AI pure-plays (or short a basket) equal to 25–35% of long SMCI notional to protect versus rapid model commoditization or regulatory bans; target protection to trigger full re-evaluation if protection doubles in value.
• Monitor triggers and timeframes: reduce aggression if (a) major open-source LLM achieves parity in 3 months, (b) China introduces heavy data-sovereignty rules limiting foreign cloud providers, or (c) export control announcements expand — reprice positions within 48–72 hours of such events.