Moore Threads Lushan (Fifth-Generation "Huagang" Architecture · Graphics Rendering + AI)
Product Overview
Lushan is Moore Threads' high-performance graphics rendering chip built on the fifth-generation "Huagang" architecture, with launch planned within 2026.
Lushan's key breakthrough is the first hardware-level ray tracing acceleration engine on a domestic GPU — previously, Moore Threads' ray tracing capability relied mainly on software paths and general-purpose compute unit emulation. Its sister product "Huashan", also on the Huagang architecture and focused on unified AI training and inference, will also reach mass production within the year; together they form the fifth-generation architecture's "graphics + compute" two-track layout.
At its 2026 interim results briefing, Moore Threads stated that "Lushan"-related GPU products are expected to launch within the year, subject to the company's official product release information. The company's H1 R&D spend was RMB 769 million (+38.16% YoY), with the R&D focus precisely on the Huashan and Lushan chips under the new-generation "Huagang" architecture and related applications.
Core Specifications
| Parameter | Value |
|---|---|
| Architecture | Fifth-generation "Huagang" architecture |
| Process Node | Not disclosed |
| Positioning | High-performance graphics rendering + AI compute |
| Graphics Features | Hardware-level ray tracing acceleration engine (a first for Moore Threads) |
| FP16/BF16 Compute | Not disclosed |
| Memory Capacity | Not disclosed |
| Memory Bandwidth | Not disclosed |
| TDP | Not disclosed |
| Software Stack | MUSA architecture (fifth generation) |
| Launch | Launch within 2026 (in development) |
⚠️ Specification Notes: Lushan is currently in development and Moore Threads has not published complete specifications. Except for "Huagang architecture" and "hardware-level ray tracing", all fields in the table are undisclosed. The widely cited "64x AI compute performance improvement" is a relative multiplier under the same architecture, not an absolute compute figure.
Moore Threads Full-Function GPU Roadmap
| Generation | Architecture | Representative Product | Status |
|---|---|---|---|
| Fourth generation | MUSA | MTT S5000 | In mass production (cloud training/inference mainstay) |
| Fourth generation | MUSA | MTT S4000 / S3000 / S2000 | In mass production |
| Fifth generation | Huagang | Lushan (graphics rendering) | Launch within the year |
| Fifth generation | Huagang | Huashan (unified AI training and inference) | Mass production within the year |
Key Insight: Moore Threads is one of the few domestic GPU vendors covering both graphics rendering and AI compute. Lushan fills the ray tracing graphics gap while Huashan fills the unified training-and-inference compute gap — the two lines together support the company's "full-function GPU" positioning.
Moore Threads H1 2026 Business Results
| Parameter | Value | YoY |
|---|---|---|
| Operating revenue | RMB 1.736 billion | +147.42% (already exceeding full-year 2025) |
| Cloud product revenue | RMB 1.693 billion (~97.5% share) | — |
| Net loss attributable to parent | RMB 11.5631 million | Narrowed by 95.73% |
| R&D spend | RMB 769 million | +38.16% |
| Overall gross margin | 56.95% | Down (raw material price increases + delivery mix changes) |
The company stated that in H1 it onboarded key customers in areas such as leading internet companies and telecom operators, and will continue advancing an integrated layout of high-performance domestic GPUs, advanced interconnect networks, cluster software stacks, and compute scheduling platforms, building the technical foundation to support a 100,000-card-scale intelligent computing cluster.
Vendor Information
| Parameter | Value |
|---|---|
| Company | Moore Threads Intelligent Technology (Beijing) Co., Ltd. |
| Headquarters | Beijing |
| Founded | 2020-10 |
| Listing | STAR Market (688795) |
| Core Product Lines | Full-function GPU (graphics + AI compute), Kuae intelligent computing cluster |
| Software Stack | MUSA architecture |
Use Cases
- ✅ Professional graphics rendering / digital twins / 3D design (ray tracing capability)
- ✅ Graphics + AI fusion workloads (rendering + inference on one card)
- ✅ Domestic workstation / graphics terminal substitution
- ❌ Large-model training mainstay (left to Huashan / the MTT S5000)
- ❌ No product currently on sale (in development)
Related Cards
- Moore Threads Huashan — Same-architecture unified AI training and inference
- Moore Threads MTT S5000 — Current cloud flagship on sale
- Moore Threads MTT S4000 — Previous-generation cloud product
- Full Comparison Table
References
- Moore Threads 2026 interim results briefing (progress on the new-generation GPUs "Lushan" and "Huashan")
- Enflame's IPO and the second half for the domestic GPU "Four Little Dragons": next-generation product roadmap
- WAIC 2026: Moore Threads AI factory and domestic GPU cluster showcase
- Moore Threads official website