The Short Sellers' Verdict on China's AI Price War

Short interest in two of China's most prominent AI startups hit an all-time high this week. The market's message is unambiguous: Zhipu AI and MiniMax are bleeding in a price war they cannot win. As of Tuesday's close, bearish bets against both companies exceeded any previous level since their public listings. Let's parse this signal before the narrative machine spins it into something it is not. The price war in China's large language model sector is not new. It started in May 2023 when ByteDance cut API prices to near zero, forcing every player to follow. What is new is the market's willingness to express a verdict in capital terms. Short sellers are not journalists. They put money behind their conviction. The record positioning against Zhipu and MiniMax represents a structural opinion: these two companies, despite their technical pedigree, are trapped in a commodity business with no exit. Let me establish context. Zhipu AI, backed by Tsinghua University, built its reputation on the GLM series. MiniMax, founded by former SenseTime executives, made its mark with the MiniMax-01 model and a strong play in consumer-facing AI applications. Both raised billions in private capital at valuations that implied they could compete with Baidu, Alibaba, and Tencent. The market is now saying those valuations were fiction. Here is the core issue: short sellers are betting on a specific failure mode. They see a market where API prices have collapsed by over 90% in two years. They see inference costs that remain stubbornly high. They see a customer base that is price-sensitive and disloyal. And they see a competitive landscape where the three largest cloud providers can subsidize their AI divisions with profits from unrelated businesses. Zhipu and MiniMax have no such luxury. My own experience with on-chain data and market microstructure tells me to look at the order flow, not the headlines. The short position here is not retail speculation. It is institutional, coordinated, and based on a specific thesis about unit economics. Let me walk through that thesis in detail. The first element is revenue quality. Zhipu and MiniMax generate most of their revenue from API calls. This is the most commoditized segment of the AI stack. When a developer chooses an LLM provider, they compare price per million tokens, latency, and context window. They do not care about the model architecture or the alignment technique. They care about the invoice. In a price war, this means every competitor can only win by cutting prices further, which destroys margins for everyone. The second element is cost structure. Training a frontier-level model requires tens of thousands of GPUs. Inference at scale requires a similar investment. Zhipu and MiniMax do not own their data centers. They rent capacity from cloud providers who are also their competitors. This creates an inherent cost disadvantage. When Alibaba cuts its API prices, it can absorb the loss because its cloud division benefits from higher utilization. When Zhipu cuts prices, it simply burns more cash. The third element is differentiation. Both companies have tried to build moats. Zhipu has pushed into government and enterprise contracts, leveraging its academic connections. MiniMax has focused on consumer applications, particularly in short-form video and gaming. But these attempts at differentiation have not translated into pricing power. Government contracts are notoriously slow and margin-poor. Consumer apps are subject to fickle user preferences and high acquisition costs. Neither provides a defensible position against a well-funded giant with a superior distribution network. The contrarian angle here is worth examining. The short thesis is compelling, but it may be premature. Short sellers often confuse a bad business model with a failing company. Zhipu and MiniMax have real technology. Their models rank in the top tier of Chinese LLMs. They have talented teams and strong institutional backing. The Chinese government has designated AI as a strategic priority, and there is a political dimension to their survival. If Beijing decides that having two independent AI champions is important, it can provide support through state-linked funds, procurement preferences, or regulatory barriers to foreign competitors. There is also the possibility that the price war ends sooner than expected. The current situation is unsustainable for everyone. Even Alibaba and ByteDance are losing money on their AI divisions. The market may be approaching a point of consolidation where weaker players exit and pricing stabilizes. If that happens, Zhipu and MiniMax, as the last independent players standing, could emerge stronger. But I would not bet on that scenario. My four years of ledgers in this industry have taught me that capital markets are rarely wrong about the direction of travel. The short sellers are not betting on a quarterly miss. They are betting on a structural transformation that makes these companies irrelevant. The code whispered what the whitepaper hid: the unit economics of the Chinese AI market do not support three independent competitors. They barely support two. The key metric to watch is not the short interest itself but the cash runway. Both companies have raised substantial war chests, but they are burning through them at an alarming rate. Zhipu's burn rate is estimated at over $100 million per quarter. MiniMax is not far behind. At current levels, they have 12 to 18 months of runway. That is not enough time to achieve profitability in a market where prices are still falling. The second metric to watch is customer churn. The short sellers have access to private data that suggests enterprise clients are already diversifying away from Zhipu and MiniMax toward the larger providers. If this trend accelerates, revenue growth will stall even as costs continue to rise. The third metric is the GPU supply chain. Both companies are dependent on NVIDIA chips, which are subject to export controls. They have made progress with domestic alternatives like Huawei's Ascend, but the performance gap remains significant. If they cannot access the latest hardware, they will fall behind in model quality, which will further erode their competitive position. Here is my takeaway for the next quarter. The short interest will likely remain elevated until we see evidence of a strategic pivot. That pivot could take the form of a merger between Zhipu and MiniMax, a major government-backed investment, or a radical shift in business model away from API commoditization. Absent such a move, the bearish thesis will play out. The market is not being cruel. It is being rational. Whale tails flicker in the shadows of the Chinese AI market, but this time the signal is not about accumulation. It is about distribution. The smart money has made its choice. The question is whether Zhipu and MiniMax can prove the market wrong before the cash runs out. The data says no. The narrative says maybe. I trust the data.