马斯克秀 AI 算力:计划年底前配 110 万张英伟达 GB300,目标约 6 个月跻身行业领先Musk shows off AI computing power: plans to deploy 1.1 million NVIDIA GB300 cards by year-end, targeting industry leadership in about 6 months


AI free token/AI免费积分

IT之家 9 月 25 日消息,全球首富埃隆 · 马斯克(Elon Musk)今天(9 月 25 日)在 X 平台发布推文,详细披露了旗下 SpaceXAI 公司的算力部署情况。

SpaceXAI 公司旗下当下主要运营 Colossus 1 和 Colossus 2 两大计算集群,其中 Colossus 1 包含 15 万张英伟达 H100、5 万张 H200 以及 3 万张 GB200;而 Colossus 2 集群包含 11 万张 GB200 以及 44 万张 GB300。

马斯克表示下周还将投运 22 万张 GB300,并计划 2026 年 11 月再投运 22 万张 GB300,如果一切部署顺利,在 2026 年 12 月月底计划再投入 22 万张 GB300,意味着 Colossus 2 计算集群到今年年底将配备 110 万张 GB300。

IT之家注:GB200/GB300 是英伟达(Nvidia)推出的新一代 GPU 芯片型号,属于 Blackwell 架构后续机型,专为 AI 训练与推理优化。GB 系列相较 H100/H200 在能效与算力密度上显著提高,当前成为 xAI 新建集群的主力芯片。

马斯克秀 AI 算力:计划年底前配 110 万张英伟达 GB300,目标约 6 个月跻身行业领先

对于普通使用者而言,在这些庞大英伟达计算集群算力支撑下,SpaceXAI 可以构建更聪明、更强大的 Grok 等 AI 机型,训练更先进的人工智能模型,提高机型感受,预估带来更精准的回答、更强的逻辑推理能力、更自然的语音交互等。

在谈及 AI 行业竞争,马斯克明确表示,SpaceXAI 的 AI 研发只有 3 年历史,而 Anthropic 和 OpenAI 等竞争对手已有 6 到 10 年的积累。他提出如果发展势头保持强劲,SpaceXAI 有望在约 6 个月内达到领先地位。

马斯克秀 AI 算力:计划年底前配 110 万张英伟达 GB300,目标约 6 个月跻身行业领先

本文来源:https://www.ithome.com/1/007/138.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

IT Home News, September 25: World's richest man Elon Musk today (September 25) posted a tweet on the X platform, detailing the computing power deployment of his SpaceXAI company.

SpaceXAI currently mainly operates two computing clusters: Colossus 1 and Colossus 2. Colossus 1 includes 150,000 Nvidia H100s, 50,000 H200s, and 30,000 GB200s; while the Colossus 2 cluster contains 110,000 GB200s and 440,000 GB300s.

Musk stated that another 220,000 GB300s will be put into operation next week, with a plan to deploy another 220,000 GB300s in November 2026. If all deployments go smoothly, another 220,000 GB300s are planned to be invested by the end of December 2026, meaning the Colossus 2 computing cluster will be equipped with 1.1 million GB300s by the end of this year.

IT Home Note: GB200/GB300 are new-generation GPU chip models launched by Nvidia, belonging to the follow-up models of the Blackwell architecture, specifically optimized for AI training and inference. The GB series significantly improves in energy efficiency and computing power density compared to H100/H200, and is currently the main chip for xAI's new clusters.

Musk shows off AI computing power: plans to deploy 1.1 million NVIDIA GB300 cards by year-end, targeting industry leadership in about 6 months

For ordinary users, with the support of these massive Nvidia computing clusters, SpaceXAI can build smarter and more powerful AI models like Grok, train more advanced artificial intelligence models, and improve the model experience, which is expected to bring more accurate answers, stronger logical reasoning abilities, and more natural voice interactions.

When discussing competition in the AI industry, Musk explicitly stated that SpaceXAI's AI R&D has only a 3-year history, while competitors such as Anthropic and OpenAI have already accumulated 6 to 10 years of experience. He proposed that if the development momentum remains strong, SpaceXAI is expected to reach a leading position within approximately 6 months.

Musk shows off AI computing power: plans to deploy 1.1 million NVIDIA GB300 cards by year-end, targeting industry leadership in about 6 months

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry