SemiAnalysis 报道称，SpaceX 计划在 2027 年建设 6-8GW 的增量数据中心，可能超过 10GW，并有望实现每年 3000 亿美元的经常性收入。这项雄心勃勃的扩张得益于 AI 推理的高盈利能力，像 OpenAI 和 Anthropic 这样的公司每年每个 GW 可以产生超过 1000 亿美元的收入。微软也是一个关键参与者，在过去八个月中增加了 10GW 的容量，并受益于直接访问 OpenAI 模型而无需收入分成。该报告强调了 SpaceX 快速建设这些设施的潜力，并引用了快速建设数据中心和发电厂的例子。 AI
影响 此次扩张标志着 AI 计算能力将大幅提升，可能加速各行业的 AI 开发和部署。
排序理由 一家主要科技公司发布了重要的基础设施和商业战略公告，对 AI 计算能力产生了影响。
AI 生成摘要 · Google Gemini · 来自 5 个来源。 我们如何撰写摘要 →
完整方法见我们的编辑标准。
So where do the 2027 sites go? We evaluated every site suitable for SpaceX for our Datacenter Model subscribers, and our Energy Model tracks the gas equipment available quarter by quarter across 30+ suppliers. Come find out before the market does: (5/5) https://t.co/4HzSGA7ui3
Can SpaceX actually build it? Colossus 1's 300MW took just 122 days. The Southaven power plant grew from 27 turbines (~495MW) in February to 69 turbines (1.7GW) in July. MiniHard will likely hit 450-500MW in ~5 months. It's speed above all. (4/5) https://t.co/hjnXoyaMdx
And there's a third company that can print these economics: Microsoft. Full access to OpenAI models, none of the training costs, and no more 20% revenue share. In the last 8 months, Microsoft has added 10GW of new planned capacity! and a SpaceX deal with 90-day cancellation is ht…
Why would anyone pay for that much compute? Inference is unbelievably profitable. Our Tokenomics Model and Inference Simulator show OpenAI and Anthropic can each generate over $100B/GW/year selling API inference on a GB300 cluster, vs ~$12B/GW/year of rental costs. (2/5) https://…
Elon shocked the world again on SpaceX's first earnings call: 6-8GW of incremental datacenters in 2027 alone, possibly well above 10GW. Sounds impossible, but we believe it. We see SpaceX on track for ~10GW by year-end 2027, and a path to $300B ARR. (1/5)🧵 https://t.co/X6qtIKVREZ