legal high

China’s ‘Nvidia alternative’ Moore Threads posts 147.42% growth in H1 revenue amid GPU sector’s expansion_我的网站

A |      8 月 26 日消息，OpenAI 昨日（8 月 25 日）发布博文，展示其首款自研 Jalapeño 芯片，并表示无论是单用户 token 处理量，还是每千瓦吞吐量，都超过了目前市面上的顶尖推理处理器。在官方博文中，OpenAI 使用半导体研究公司 SemiAnalysis 开发的公共基准测试工具 InferenceX，通过 GPT-OSS 120B、DeepSeek R1 670B 和 Kimi K2.5 1T 三款模型测试。    

    A logo of Moore Threads at the 2025 World Artificial Intelligence Conference in Shanghai on July 28, 2025 Photo: VCG
Chinese chipmaker Moore Threads, often described as "China's Nvidia" or "Nvidia alternative," reported fast revenue growth in the first half of 2026, which exceeded its entire 2025 full-year figure. The growth accompanied other domestic graphics processing unit (GPU) companies' gains, marking a clear shift toward profitability.
Moore Threads, China's GPU leader, reported 1.736 billion yuan ($257 million) in first-half revenue, up 147.42 percent year-on-year. Gross profit reached 989 million yuan, up 103.78 percent, while net losses attributable to shareholders narrowed sharply by 95.73 percent, according to the financial results released on Sunday.
The company attributed the strong growth to booming artificial intelligence (AI) demand for full-function GPUs and the accelerated commercialization of its Kuae intelligent computing clusters. 
Moore Threads maintained heavy research and development (R&D) spending of 769 million yuan in the first half, up 38.16 percent year-on-year, bringing cumulative investment since 2022 close to 5.9 billion yuan. As of the end of June, it filed 2,167 patent applications and held 788 authorized patents, which ranked among the top ranks of domestic GPU enterprises, the company said.
As a result of its investment in R&D, Moore Threads' flagship MTT S5000 GPUs have realized large-scale sales, with deployments in Beijing, Wuxi, Hangzhou and other cities. The first batch of intelligent computing clusters passed national security and reliability evaluations and achieved multiple "domestic chip training domestic models" breakthroughs. Its MUSA software stack now supports more than 800,000 developers.
Moore Threads also announced plans for an IPO in the Hong Kong market. It was listed on Shanghai's STAR Market on December 5, 2025.
"Chinese firms are increasingly turning to domestic suppliers, accelerating demand for local AI accelerators. Under the combined forces of surging inference computing needs and continued capital expenditure by major cloud providers, the domestic GPU and AI chip sectors are moving from an investment phase into a commercialization and profit-delivery phase," Yang Delong, chief economist at Shenzhen-based First Seafront Fund, told the Global Times on Sunday.
Biren, Moore Threads, MetaX and Enflame Technology are dubbed China's "four little dragons" in the GPU sector. Apart from Moore Threads, the other companies also showed rapid growth in their recent financial results.
In June, MetaX reported first-quarter revenue of 562 million yuan, up 75.37 percent year-on-year, and said that it had begun volume sales of its next-generation general-purpose GPU C600, representing further breakthroughs in the areas of domestic adaptation and supply chain autonomy control.
Enflame Technology reported 1,474.85 percent year-on-year growth in its revenue in the first quarter. The company estimated that its first-half revenue would grow 258.68-289.13 percent, likely to reach the level of the entire year of 2025. 
Domestic GPU company Biren in March released its first financial results after being listed in Hong Kong. It reported 207.2 percent full-year growth in revenue in 2025, with gross profit up 210.8 percent. 
Fellow AI chip leader Cambricon also delivered equally robust results on Friday. Its first-half revenue hit 5.996 billion yuan, up 108.13 percent year-on-year, while net profit attributable to shareholders rose 122.61 percent to 2.311 billion yuan. 
Growth was driven by deeper partnerships with leading financial and internet firms and scaled commercial deployments. Cambricon said that it completed adaptation support for major Chinese open-source AI models including GLM, DeepSeek, Qwen, Kimi and MiniMax, strengthening its domestic AI computing ecosystem.
Ma Jihua, a veteran industry insider, told the Global Times that multiple Chinese companies are advancing along different technological routes covering high-, mid- and low-end products. 
"With rapidly expanding capacity, they are better meeting domestic needs and beginning to reach global markets, thereby promoting more accessible AI," said Ma.
"The AI technology sector's high prosperity is expected to persist, remaining a core investment theme with substantial medium- and long-term opportunities," said Yang.
。测试显示对比市面上领先的 AI 系统（GB200 和 GB300），每瓦能够完成的 AI 吞吐量是对比系统的 1.5 至 1.9 倍，端到端延迟则约为对比系统的 28% 至 59%。混合 Token 吞吐量与现有 AI 加速器的最快解码速度对比3 款模型峰值单用户解码吞吐量各平台最佳运行点的混合 Token 峰值吞吐量OpenAI 表示，Jalapeno 在运行 GPT-OSS 120B 时，每千瓦的峰值吞吐量比 GB200 系统高出约 1.9 倍。在 DeepSeek R1 上，它的每瓦性能比 GB300 高出约 1.7 倍；在 Kimi K2.5 上，它的每瓦性能比 GB300 高出约 1.5 倍。Jalapeno 的额定功率为 700 瓦，但在测试工作负载期间，其持续功耗保持在 550 瓦或以下。附上相关图片如下：延伸阅读Jalapeño 是 OpenAI 首款自研 AI 推理芯片，由 OpenAI 与博通（Broadcom）合作开发，台积电采用 3nm 制程代工制造，于 2026 年 6 月 24 日首次对外展示。该芯片是一颗 ASIC（专用集成电路），采用 systolic array（脉动阵列）架构，配备 HBM 高带宽内存，专为大语言模型推理场景设计，不用于模型训练。芯片及配套服务器系统均不会对外销售，仅供 OpenAI 内部使用，旨在降低自身 GPU 采购和算力运营成本。OpenAI 硬件负责人 Richard Ho 领导了该芯片的研发，他此前在 Google 工作近九年，是 Cloud TPU 项目的核心工程师。Jalapeño 从架构设计到完成流片仅耗时约 9 个月，OpenAI 称这是高性能先进半导体领域有史以来最快的 ASIC 开发周期之一。OpenAI 自身的 AI 模型也深度参与了芯片设计流程，帮助加速设计优化。

B | 目前第二代 Jalapeño 芯片已进入后期开发阶段，预计未来数月内完成流片；第三代芯片也已启动概念设计。OpenAI 计划于 2026 年年底前小规模部署 Jalapeño，2027 年逐步扩大部署规模。需要注意的是，Jalapeño 此次测试对比的对象为英伟达 GB200 和 GB300 系统，尚未与英伟达最新一代 Vera Rubin 芯片进行对比测试。OpenAI 表示英伟达仍是其重要合作伙伴，未来将持续采用多供应商策略以满足庞大的算力需求。相关阅读：。

Current article:http://d8enc.yaoyaotoutuanmieshaideizun.buzz/1w4xboi/20260826/4397.html

Published on:00:00:00