中国人工智能芯片巨头寒武纪科技周五公布，第三季度净利润同比激增147.7%，随着中国大力推动替代英伟达芯片，持续推高本土芯片制造商的高增长。
Chinese artificial intelligence chip giant Cambricon Technologies reported a 147.7 per cent surge in third-quarter profit on Friday, as the country’s aggressive push to replace Nvidia chips continues to fuel high growth among local chipmakers.
这家在上海上市的芯片制造商在截至9月30日的三个月里实现净利润14亿元人民币（约合2.1亿美元），高于去年同期的5.666亿元人民币。
The Shanghai-listed chipmaker generated 1.4 billion yuan (US$210 million) in net profit for the three months ended September 30, up from 566.6 million yuan during the same period last year.
根据其交易所披露文件，营收达到35亿元人民币，同比增长104.5%。这一增速快于公司第二季度75.8%的营收增幅。
Revenue reached 3.5 billion yuan, an increase of 104.5 per cent from a year ago, according to its stock exchange filing. That was faster than the company’s 75.8 per cent revenue increase seen in the second quarter.
寒武纪将营收激增归因于其持续的市场拓展和“推动AI应用落地”的努力。公司在文件中表示，这一营收增长直接带动了总利润和净利润的同比激增。
Cambricon attributed the revenue surge to its continuous market expansion and efforts to “facilitate the implementation of AI applications”. This top-line increase directly drove the year-on-year surge in total profit and net profit, the company said in the filing.
在周五业绩发布前，寒武纪股价收涨5.5%，报932.2元人民币。
Shares of Cambricon closed up 5.5 per cent at 932.2 yuan ahead of the earnings release on Friday.
本周早些时候，另一家国内领先的AI芯片制造商海光信息披露，第三季度营收59亿元人民币，同比增长45.5%。其季度净利润同比激增37.9%至10亿元人民币。
Earlier this week, Hygon Information Technology, another leading domestic AI chipmaker, posted third-quarter revenue of 5.9 billion yuan, an increase of 45.5 per cent year on year. Its net profit for the quarter surged 37.9 per cent year on year to 1 billion yuan.
寒武纪和海光的强劲势头建立在北京发展自给自足AI算力生态系统的倡议之上，随着中国加速部署由国产芯片驱动的AI基础设施，这一势头愈发明显。
The momentum at Cambricon and Hygon builds on Beijing’s initiative to develop a self-sufficient AI computing ecosystem, as the country accelerates the deployment of AI infrastructure powered by home-grown chips.
这两家公司被视为国内AI芯片领域的第一梯队企业，正迅速填补华盛顿出口管制造成的市场空白，这些管制切断了中国获取英伟达等美国芯片巨头先进AI加速器的渠道。
The two companies – regarded as top-tier players in the domestic AI chip sector – are moving rapidly to fill the market vacuum left by Washington’s export controls that have cut off China’s access to advanced AI accelerators from US chip giants like Nvidia.
分析师表示，寒武纪的盈利前景在很大程度上取决于其最新芯片的商业化进度，而后者又受到供应链压力的影响，尤其是AI数据中心所需的高带宽内存芯片短缺。
Analysts said Cambricon’s earnings outlook depended largely on the commercialisation progress of its latest chips, which was in turn affected by supply chain pressures, especially the shortage of high bandwidth memory chips required for AI data centres.
摩根士丹利分析师在近期的一份研报中写道：“寒武纪在第三季度向一家主要云服务供应商交付了MLU692用于测试，但我们预计大部分收入将在第四季度确认。由于高带宽内存成本上升挤压单位经济效益，MLU580的出货量也依然有限。”MLU692属于寒武纪的MLU690芯片系列，也称为思元690。根据摩根士丹利早前的一份报告，该系列在“每秒token数”这一指标上可比英伟达的H20高出50%至150%，该指标本质上衡量AI能处理并输出多少个词或其他数据单元。
“Cambricon delivered [its] MLU692 to a major cloud service provider in Q3 for testing, but we expect most revenue to be recognised in Q4,” Morgan Stanley analysts wrote in a recent research report. “MLU580 shipments also remain limited as higher high bandwidth memory costs pressure unit economics.”The MLU692 is part of Cambricon’s MLU690 chip series, also known as Siyuan 690. The series can outperform Nvidia’s H20 by 50 to 150 per cent when measured in “tokens per second”, a metric that essentially counts how many words or other data AI can process and spit out, according to an earlier Morgan Stanley report.
与此同时，海光面临着对其下一代中央处理器（CPU）及深度计算单元——一种专为AI设计的加速卡——的高期待。
Hygon, meanwhile, faces high expectations for its next-generation central processing unit and its deep computing unit – a type of accelerator card designed specifically for AI.
该公司周四表示，进展“按计划推进”，算力密度和互联带宽等关键指标将得到提升，但量产时间表仍待定。
The company said on Thursday that the progress was “on track” with key metrics such as compute density and interconnect bandwidth set to be upgraded, though mass production timelines remained pending.