Chinese AI startup MiniMax is building a 2.7 trillion-parameter, open-weight large language model that Reuters says could arrive as early as the third quarter. If it ships as described, MiniMax would jump past China’s current heavyweight models, including Meituan’s 1.6 trillion-parameter LongCat-2.0 and AI lab DeepSeek’s 1.6 trillion-parameter V4-Pro. Bigger models can improve “reasoning” and handle longer, multi-step tasks, but only if they’re practical to run, which is where Mixture of Experts aims to cut the computing power needed per answer. Open-weight models are becoming popular outside China partly because they’re cheaper to use than proprietary rivals. For markets, MiniMax’s HK$4.8 billion IPO turns AI scale into a cash-burn test, as training and running a 2.7 trillion-parameter model can rack up huge bills. Reuters says the firm raised HK$4.8 billion ($614 million) in a Hong Kong IPO in January and is planning a second listing on Shanghai’s STAR Market to fund computing needs while trying to turn open-weight traction into steady revenue.