· 供应链分析

分析超大规模云厂商数据中心模式差异及潜在合作逻辑

涉及标的:

中文翻译

我不会像某些特定的 $BMNR 或 $IREN 投资者那样盲目吹捧,认为街上的绿灯就意味着对公司利好。所以我倾向于认为这只是巧合。因为看起来 $AMZN、$GOOGL 倾向于偏好数据中心托管(Colo)模式(因为他们可以插入自己的 TPU、未来的 Trainium 芯片),而 $MSFT、$META(以及 99.9% 使用 $NVDA 且没有现成定制芯片的 AI 公司)则偏好 $NBIS、$IREN 类型的模式。所以 $AMZN 宣布在印第安纳州为其 AWS 数据中心部门投资 150 亿美元。但关键要注意的是,他们此前在那里已经花费了 313 亿美元,所以这并不算全新投入。你可能看到了 $NBIS 正在印第安纳波利斯附近建设 1000 多英亩的绿地数据中心。这只是一个有利于建设的区域,可能并非针对合作伙伴关系或像 Anthropic(最近承诺 400 亿美元资本支出)那样的另一笔超大规模云厂商交易。说实话,我现在可能更相信 Anthropic 的资本支出承诺而不是 OpenAI。话虽如此,也许如果 $AMZN 耗尽容量并达到与 $NVDA 的最大采购订单上限(因为他们正在插入自己的 $CIFR 芯片),他们可能会转而使用 Nebius。

英文原文

I'm not going to be a blind shill like some specific $BMNR or $IREN investors that can say a green light on the street means bullish for the company. So I would lean coincidental. Since it looks like $AMZN, $GOOGL tends to favor colo models (since they can plug in their own TPUs, Trainium chips in the future) while $MSFT, $META (and 99.9% of AI companies that use $NVDA and don't have custom chips readily built out) prefer $NBIS, $IREN type models. So $AMZN announced they're investing $15B in Indiana for their AWS DC segment. But key thing to note is they've already spent $31.3B there before so it's not exactly new. You probably saw how $NBIS is doing a 1000+ acre greenfield DC near Indianapolis. It's just a favorable area for buildout, probably not directed at a partnership and another hyperscaler deal like Antrhopic (who committed $40B in capex spend recently). I'd probably trust Antrophic more than OpenAI right now with capex spend lol. That being said, maybe if $AMZN runs out of capacity and hits max purchase order with $NVDA (since they're plugging in their own chips with $CIFR), they would use Nebius instead.

在 X 上查看原推 ↗