Zhipu AI, MiniMax shares slide after DeepSeek unveils new open-source model

DeepSeek unveiled preview versions of its V4 open-source AI model, including a high-performance Pro version and lower-cost Flash variant, with up to a 1 million token context window. The release intensified competitive pressure in China’s AI sector, sending Zhipu AI (Knowledge Atlas Tech) down 8% to HK$946 and MiniMax Group down 4.7% to HK$818.5. The move reinforces a low-cost, open-source strategy that could pressure pricing and margins across rival AI firms.

Analysis

The market is starting to price Chinese AI as a network-effects business rather than a pure model-quality race, and that is the real pressure point for the public names. An open-weight release with high performance plus lower-cost access shifts bargaining power toward developers and enterprise buyers, compressing monetization assumptions for smaller domestic model vendors that still need to spend aggressively on training and inference. The second-order loser is not just the listed AI firms cited here, but the broader ecosystem of GPU resellers, model fine-tuning service providers, and cloud integrators that were expecting scarcity rents from proprietary access.

The important timing issue is that this is a margin story before it becomes an adoption story. In the next 1-3 quarters, the easiest read-through is multiple compression on any company whose equity case rests on premium API pricing or model exclusivity; the upside from broader usage will lag because enterprise switching tends to reprice contracts slowly. If the new release materially improves developer lock-in, then the competitive damage extends to adjacent application layers, since application vendors will be forced to compete on UX and distribution rather than model access.

The contrarian view is that open source can accelerate category growth fast enough to offset pricing pressure, especially if the largest players respond by bundling distribution, enterprise support, and vertical solutions. That means the immediate selloff may be an overreaction for the best-capitalized platforms, but likely underreaction for the weaker private-to-public conversion stories that need future funding rounds. The key tail risk is a local price war that forces another 12-18 months of heavy capex with no visible path to operating leverage, which would be a meaningful negative for sentiment even if benchmark performance keeps improving.

Key Decisions for Investors

• Short a basket of China-listed AI pure plays for 1-3 months; favor names with the highest valuation-to-revenue multiples and weakest cash runway. Risk/reward is attractive if the market starts marking down future funding needs rather than current product momentum.
• Pair trade: long the best-capitalized China tech platform with distribution advantages, short a smaller model vendor. The long leg should benefit if open-source adoption drives usage while the short leg is exposed to pricing compression and higher burn.
• Buy near-dated puts on the most overextended HK-listed AI names into any relief rally. Use 20-30% downside target over 4-8 weeks with defined premium risk; the thesis is multiple reset, not fundamental collapse.
• For longer-duration exposure, wait for a second-leg selloff before adding to any China AI long. Entry should come only if the market over-penalizes the infrastructure and platform winners along with the pure plays, creating a better risk/reward entry below prior support.
• Monitor cloud and semiconductor suppliers as a hedge: if this triggers broader usage, the best hardware beneficiaries may outperform model vendors over 6-12 months.