# Compute

> Free compute support for research and startups; the application follows the accelerator process.

Source: https://www.miracleplus.com/compute/
Markdown: https://www.miracleplus.com/compute.md

Compute Power for Free

Empowering Research and Innovation

### 200+ Petaflops

GPU clusters and TPUs for large-scale model training and inference

### ¥1 Million

Up to ¥1 million in free compute for top research and startup projects

### Safe

Enterprise-grade security, data privacy, 24/7 technical support

### Fast Deployment

One-click deployment: launch fast. Focus on innovation

### High-performance compute that drives innovation

Backed by MiraclePlus's high-performance compute, a new generation of high-potential, deep-tech projects is on the rise. We offer free compute for AI teams from universities, research institutions, and startups-with a transparent process, easy application, and the support you need to move fast.

Whether you're a pioneer in frontier tech or an entrepreneur bringing innovations to life, we welcome you to be part of our community.

### Program Overview

Who Can Apply

Open to universities, research institutes, tech companies, startups, and independent developers. If your project has a clear AI focus and aims to turn technology into real products, you're eligible to apply. We prioritize innovative proposals with strong social impact.

Review Timeline

Decisions in 1-2 months. Approved projects receive compute resources and technical onboarding.

Compute Support

Up to ¥1M in resources (≈8x H800 GPUs for 15 months). We'll work with you to assess needs and progress.

Technical Support

Full technical and integration support provided. We support major AI frameworks and hardware setups. Contact us during application for configuration details.

### Application Process

The compute application follows the same process as the MiraclePlus Accelerator. Just click "Apply for Free" to submit your application.

### Priority Projects

Research Project

MiraclePLus supports frontier AI project -- from scaling laws to RL and AI agents. If you're pushing limits, we're your partner.

Open-Source Project

AI research and entrepreneurship come with significant challenges — but open-sourcing can unlock wider support. If you're looking to open-source a key AI initiative, MiraclePlus Compute is here to support your journey.

Startup Project

Startups are driving real-world impact through innovative products and models. We supports teams with strong tech and market potential. If you're building the next breakthrough, we welcome your application.

### Jun Wang Lab, CAS Institute of Microbiology

Project Overview

A team from the Institute of Microbiology, Chinese Academy of Sciences is using natural language methods to build large models for protein functions and interactions. Their goal is to develop next-generation peptide-based drugs at scale for key therapeutic targets, addressing major chronic diseases such as metabolic disorders and cancer.

Team Introduction

Jun Wang from the Institute of Microbiology, Chinese Academy of Sciences, earned his Ph.D. from the Max Planck Institute for Evolutionary Biology in Germany. His earlier work on using AI to discover antimicrobial drugs was published in Nature Biotechnology and selected as one of The Guardian’s “Top 10 Scientific Breakthroughs of 2022.”

### Mooncake – Scalable, Fast Inference for Large Language Models

Project Overview

Mooncake is an open-source, next-gen distributed inference architecture for LLMs. With a high-performance data transfer engine and KV cache pooling, it accelerates inference and improves GPU utilization. Mooncake integrates with projects like SGLang, vLLM, Dynamo, and LMCache, and is deployed in production by Moonshot AI (Kimi), Alibaba, and Ant Group.

Team Introduction

·Feng Ren – Project Initiator / Ph.D. in Computer Science, Tsinghua University. Focused on high-performance memory storage systems.

·Mingxing Zhang – Project Advisor / Assistant Professor, Department of Computer Science, Tsinghua University. Research focuses on distributed systems and storage optimization.

### MidReal – Deep Reasoning AI Fiction

Project Overview

System2 Research developed the MidReal.ai platform based on its proprietary long-range reasoning architecture, System2 Agent. Focused on post-training and multimodal intent alignment, MidReal.ai has attracted over 3 million users in a year and leads global benchmarks in complex plot generation. Its high interpretability and industrial-grade reasoning efficiency lay the groundwork for film adaptations and interactive storytelling.

Team Introduction

·Our core team hails from top institutions including MIT, Princeton, Duke, Tsinghua, and OpenAI. Collectively, we've published over 180 papers at leading AI conferences, with more than 16,000 academic citations.

·The team has built two products with 10M+ users and led three multimillion-dollar ventures. Backed by a global network of 20+ top labs and companies, we’re scaling the System2 Agent ecosystem.