Retreat of Short Sellers Opens Valuation Window for Hong Kong’s AI Leaders

Hong Kong’s two AI large‑model leaders MINIMAX‑W (00100.HK) and Zhipu (02513.HK) staged modest rebounds on April 29, rising 6.31% and 0.90% respectively, as short‑selling pressure eased sharply. For MiniMax, shorted shares fell from over 900,000 on April 24 to under 200,000 by April 28, with shorted value dropping from HKD 715 million to HKD 145 million. Zhipu showed a similar pattern, with shorted shares halved and shorted value reduced by more than half, signaling a clear retreat of bearish sentiment.

Investor confidence is being reinforced by new buy recommendations. Citi initiated coverage of MiniMax Xiyu Technology with a “Buy” rating and a HKD 1,330 target price, projecting 128% revenue CAGR from 2025 to 2030 and profitability by late 2029. The valuation is based on 30x estimated 2028 sales, with the forthcoming M3 series model highlighted as a catalyst to unlock monetization potential and advance the company’s “digital employee” strategy. Morgan Stanley’s Asia‑Pacific team raised its target price from HKD 990 to HKD 1,100, noting that while US firms lead in frontier LLM breakthroughs, Chinese AI companies are closing the gap through engineering efficiency. Under computing constraints, leading Chinese and US models are nearly level on the Artificial Analysis index, with MiniMax M2.7 and Zhipu GLM‑5.1 scoring in the 50–54 range, narrowing the gap to just 3–6 months.

China’s cost‑performance advantage is emerging as a global differentiator. Morgan Stanley emphasized that Chinese models achieve equivalent intelligence at 15%–20% of US inference costs, driven by architecture innovation, post‑training optimization and inference infrastructure. DeepSeek‑V4 exemplifies this trend. OpenRouter data show Chinese models’ token share surged from 5% in April 2025 to 32% in March 2026, while US models fell from 58% to 19%. MiniMax, Zhipu and Alibaba saw token usage rise 4–6x between December and February–March, propelled by breakthroughs in coding and agentic workflows.

Far from commoditization, frontier models are demonstrating pricing power. Zhipu’s prices have doubled year‑to‑date, and MiniMax’s KV cache prices have also doubled. Morgan Stanley expects MiniMax to raise prices again after the M3 upgrade, catalyzing ARR growth. Forecasts suggest leading models could reach USD 1–1.5 billion ARR by end‑2026 and USD 2.5–5 billion by end‑2027, representing 3–5x annual growth. MiniMax’s ARR reached USD 150 million in February 2026, while Zhipu’s hit USD 250 million in March, underscoring strong momentum.

AI is also reshaping Hong Kong’s market structure. Zhipu and MiniMax are expected to join the Hang Seng Tech Index on June 8 with a combined weight of 5%–7%, triggering USD 1.25–1.75 billion in passive inflows. More broadly, technology accounts for 40% of IPO fundraising year‑to‑date and 43% of pipeline projects, positioning AI as a long‑term structural force reshaping index composition, liquidity and capital flows.