MiraclePlus was founded in 2019 in Beijing by Lu Qi, the former Microsoft executive vice president and Baidu group president, as the independent successor to YC China after Y Combinator withdrew from the market. It runs the same shape of programme as its predecessor — two accelerator batches a year, a standard $300,000 cheque for 7% equity, three months of weekly partner office hours, and a Demo Day in front of more than a thousand investors — but applies it to Chinese hard-technology founders rather than consumer software. Ten batches in, the firm reports 612 funded startups, roughly 1,500 alumni founders and a combined portfolio valuation near RMB 90B, with a portfolio that skews heavily toward embodied intelligence, AI infrastructure and agent products.
The firm states that over 99% of its investments sit in frontier innovation fields, and it deliberately funds projects other investors consider too early or too technical — buying deep-tech risk at a fixed pre-seed valuation rather than competing in priced rounds.
The programme's stated objective is to get a company ready to face investors within three months. Partners including Lu Qi run weekly one-on-one office hours — around 100 hours per project on average — focused on the single constraint blocking traction.
The most concentrated cluster in the portfolio: mobile robots, tactile sensing, actuators and simulation data. It reflects a view that China's manufacturing base gives embodied AI a cost and iteration advantage no software-only market can match.
Alongside hardware, the firm backs the inference and agent layer — SiliconFlow for model serving, HPC-AI Tech for distributed training, and consumer agent products such as Fellou and Macaron built by alumni founders.
MiraclePlus-backed AI companies highlighted by the firm, spanning model infrastructure, embodied intelligence and agent products.
Model inference and serving platform giving developers low-cost access to open-source and Chinese foundation models.
Creator of Colossal-AI, an open-source system for distributed large-model training and inference.
Synthetic data and simulation platform supplying training data for robotics and autonomous driving models.
Multi-dimensional tactile sensors and humanoid robots built around touch as the primary modality.
Autonomous mobile robots for factory and warehouse logistics in electronics manufacturing.
Agentic browser that plans and executes multi-step web tasks on the user's behalf.
Consumer agent that generates small personal apps and workflows from natural-language requests.
Direct-drive motors and actuator modules used in robot joints and precision motion systems.
High-performance networking and interconnect for AI training clusters.
Analogue in-memory computing chips for always-on, low-power edge perception.
Domain-specific processors targeting data-centre workloads and AI acceleration.
Consumer AI hardware and interaction devices built by an alumni founding team.
No rounds led by MiraclePlus are in the tracker yet. Deals will appear here as they are recorded.