SpaceXAI: Faster, Cheaper Compute for Google & Anthropic #shorts

演讲者声称，SpaceX AI 在构建和扩展人工智能计算能力方面独树一帜，比其他任何公司都更迅速、成本更低、更系统地进行。 作为证明，演讲者强调了两个关键例子： 1. **谷歌**，尽管拥有“几乎无限的资源”并拥有“自己的AI芯片”，但据报道正在向SpaceX AI支付费用以租赁AI算力。 2. **Anthropic**，“最近筹集了超过600亿美元”，也正在向SpaceX AI支付费用以获取算力访问权限。 演讲者解释说，谷歌和Anthropic本可以自行构建算力基础设施，但那样会： * **成本更高。** * **上线时间会大大延长。** “未能及时获得用于模型训练和推理的算力所造成的机会成本意味着这些公司会损失收入。” 因此，对他们来说，向SpaceX AI付费是一个“不假思索的决定”，因为SpaceX AI能够“更快、更便宜地将算力投入使用。” 这最终使得像谷歌和Anthropic这样的公司“大幅提升其模型的能力”以及“他们创造的全部收入”，因为他们获得了更多“用于推理的计算能力、更多的token（文本单元）以及其产品更多的使用量。”

The speaker asserts that SpaceX AI uniquely excels at building and scaling AI compute more rapidly, affordably, and coherently than any other company. As proof, the speaker highlights two key examples: 1. **Google**, despite its "essentially unlimited resources" and owning its "own AI chip," is reportedly paying SpaceX AI to lease AI compute. 2. **Anthropic**, which "recently raised $60 plus billion," is also paying SpaceX AI for compute access. The speaker explains that Google and Anthropic could have built their own compute, but it would have: * **Cost them more.** * **Taken way longer to get online.** The "opportunity cost of not having access to that compute for training and inference means lost revenue" for these companies. Therefore, it's a "no-brainer" for them to pay SpaceX AI because SpaceX AI can "get the compute online faster and cheaper." This ultimately results in companies like Google and Anthropic "massively increasing the capabilities of their models" and "all the revenue they're producing," due to increased "access to more compute for inference, more tokens, [and] more use of their products."