埃隆·马斯克是“新云”公司的悲观论者。这位SpaceX首席执行官周日在X上发帖，对“新云”——即CoreWeave和Nebius等专门为人工智能提供计算能力的云服务公司——的未来采取了强硬立场。他写道，债务市场不愿为任何几乎完全依赖人工智能数据中心获取收入的公司提供融资。他表示，SpaceX已获得400亿美元债务用于购买图形处理器（GPU），但与“新云”公司不同，他的公司通过“星链”和火箭发射业务获得收入。
Elon Musk is a neocloud doomer. The SpaceX CEO took a hard stance on the future of neoclouds — cloud companies like CoreWeave and Nebius, that specialize in providing computing power for AI — in a Sunday X post. He wrote that debt markets are reluctant to finance any company that relies almost entirely on AI data centers for revenue. He said SpaceX has secured $40 billion in debt to buy GPUs, but unlike neoclouds, his company earns revenue from its Starlink and rocket-launch activities.
“这意味着200多家‘新云’公司将面临小鱼级别的死亡和整合，”他写道。“有些能挺过去，但不会有太多。”马斯克的帖子是回复风险投资家查马斯·帕里哈皮蒂亚的，后者在X上写道，“新云”公司“组织混乱且资本不足”。在回复另一条猜测他是否试图压低“新云”公司股价并低价收购的X帖子时，马斯克回应称：“SpaceX不会收购任何‘新云’公司。毫无意义。”
"That means the 200+ neoclouds are going to see guppy-level death and consolidation," he wrote. "Some will make it through, but not many." Musk's post was in reply to one by venture capitalist Chamath Palihapitiya, who wrote on X that neoclouds are "disorganized and poorly capitalized." Replying to another X thread speculating if he was trying to drive down the prices of neocloud stocks and scoop them up for cheap, Musk replied, "SpaceX is not buying any neoclouds. Pointless."
此前，由于对高债务水平和缺乏其他收入来源的担忧，这些“新云”公司的股价近期遭受重创。CoreWeave的股价在过去一个月下跌了约37%，而Nebius的股价同期下跌了30%。然而，英伟达首席执行官黄仁勋坚持认为，对人工智能云基础设施的需求“高得离谱”。英伟达是CoreWeave、Nebius、Iren、Lambda等“新云”公司的主要支持者。
This comes as the stocks of neocloud companies have taken hits recently amid fears of high debt levels and a lack of other revenue streams. CoreWeave's stock has dropped about 37% in the past month, while Nebius' has dropped 30% in the same time. Nvidia CEO Jensen Huang, however, has maintained that demand for AI cloud infrastructure is "off the charts." Nvidia is a major backer of neocloud companies like CoreWeave, Nebius, Iren, Lambda, and others.
马斯克在周五的另一条X回复中补充说，该行业将出现大量“闲置在部分完工数据中心内的计算能力”。马斯克所说的“闲置”，是指那些已经准备好使用但因数据中心尚未具备启用条件而无法部署的计算能力。正如《商业内幕》的尤金·金上周所写，亚马逊云科技（AWS）预计将接收价值50亿美元的英伟达GB300机架，但该公司数据中心的电力不足以支持这些设备。
Another point that Musk added, in a separate X reply on Friday, was that the industry would see a lot of "stranded compute in partially complete data centers." By "stranded," Musk was referring to compute that is ready to use but can't be deployed because data centers aren't yet equipped to turn it on. As Business Insider's Eugene Kim wrote last week, Amazon Web Services is expected to receive $5 billion worth of Nvidia GB300 racks, but the company does not have enough power in its data centers to power them.
AWS将把这些设备存放在仓库中，直到其数据中心准备就绪。数据中心也面临一些挑战。伯恩斯坦研究公司最近的一份报告预测，美国计划中的数据中心容量中只有约35%能够建成。报告预测，像亚马逊、谷歌、Meta和微软这样拥有建设和运营数据中心经验的超大规模企业，很可能会成功完成其计划中中心的建设。但伯恩斯坦研究公司表示，像名人投资者凯文·奥利里这样承诺建设15吉瓦数据中心容量的新手，其项目被称为“吹牛瓦特”（bragawatts），不太可能实现。
AWS will be storing the units in warehouses until its centers are ready. Data centers, too, are running into some challenges. A recent Bernstein Research report projected that only about 35% of planned data center capacity in the US will be built. Hyperscalers like Amazon, Google, Meta, and Microsoft, which have experience building and operating data centers, will likely successfully complete construction of their planned centers, the report predicted. But Bernstein Research said newcomers, like celebrity investor Kevin O'Leary whose company has pledged to build 15 gigawatts of data center capacity, were called "bragawatts," and their projects are unlikely to be realized.
在人工智能行业面临清算之际，不仅是“新云”公司，小型人工智能模型开发初创公司（即“新实验室”）也面临困境。这些公司包括前OpenAI首席技术官米拉·穆拉蒂的Thinking Machines Lab和OpenAI联合创始人伊利亚·苏茨克维的Safe Superintelligence。这些实验室资金雄厚，但面临来自Anthropic和OpenAI等前沿人工智能巨头的激烈竞争。几位风险投资人最近告诉《商业内幕》的本·伯格曼，这些“新实验室”被过度炒作，投资它们是一个充满争议的话题。
It's not just the neoclouds facing a troubled stretch amid an AI reckoning; neolabs are, too. Neolabs are small AI model development startups, like ex-OpenAI CTO Mira Murati's Thinking Machines Lab and OpenAI cofounder Ilya Sutskever's Safe Superintelligence. These labs are heavily funded and face strong competition from frontier AI behemoths like Anthropic and OpenAI. Several VCs told Business Insider's Ben Bergman recently that neolabs were overhyped and investing in them was a contentious topic.