Management Corp. 正避开为运行复杂 AI 模型的高性能处理单元提供融资交易，转而选择投资容纳这些单元的数据中心。
Management Corp. is avoiding financing deals to back the high-powered processing units that run sophisticated AI models, opting to invest in the data centers that house them instead.
据一位接近该公司的人士透露，这家资产管理公司担心，如果为图形处理单元提供资金支持，可能会持有在发布不久后就可能过时的资产。该人士因信息属私密而要求不具名。
The asset manager has concerns that if it backs graphics processing units, it could get stuck holding assets that may be obsolete not long after they’re released, said a person close to the firm, who asked not to be identified because the information is private.
“老实说，与为被称为 GPU 的计算机芯片提供融资相比，为数据中心提供融资是一种更简单的承销，且具有非常有吸引力的相对价值，”首席执行官 Michael Arougheti 周四在与分析师的电话会议上表示。
“It’s frankly a simpler underwriting with really attractive relative value,” Chief Executive Officer Michael Arougheti said on a call with analysts Thursday, referring to financing data centers versus the computer chips known as GPUs.
他补充道：“无意贬低 GPU 融资，Ares 正在数据中心及相邻电力基础设施中发现更具吸引力的投资机会。”
“Without casting aspersions on GPU financing,” Ares is finding more attractive investments in data centers and adjacent power infrastructure, he added.
据 Arougheti 称，相对于可用于支持数据中心的资本，数据中心融资市场规模巨大。该公司估计 AI 资本支出需求约为 5 万亿美元，其中约 2 万亿美元用于 AI 实体建筑及相邻基础设施的建设，3 万亿美元用于 GPU。
The market for financing data centers is huge relative to the capital available to back them, according to Arougheti. The firm estimates AI capital expenditure needs to be around $5 trillion, with roughly $2 trillion needed for the build-out of physical AI buildings and adjacent infrastructure and $3 trillion for GPUs.
Arougheti 的言论正值投资者对 AI 行业焦虑加剧之际，近几周因担心该行业的巨额支出可能超过未来收入，股票和公司债券均遭受重创。
Arougheti’s remarks come at a time of heightened investor anxiety around the AI sector, with both stocks and corporate bonds taking a beating in recent weeks on concerns that the massive spending on the sector may outstrip future revenues.
几乎每一家美国银行和另类资产管理公司都参与了 AI 融资热潮，相关交易包括博通公司为将租赁给 Anthropic PBC 的芯片提供约 600 亿美元债务融资，以及 SpaceX 谈判筹集约 400 亿美元用于采购 Nvidia Corp. 芯片。这些芯片的成本是容纳它们的数据中心的数倍，但通常在大约五到六年内折旧。
Almost every US bank and alternative asset manager has participated in the AI financing boom, on deals including Broadcom Inc.’s roughly $60 billion debt financing for chips that will be leased to Anthropic PBC and SpaceX’s talks to raise about $40 billion for Nvidia Corp. chips. These chips cost multiples more than the data centers that house them, but typically depreciate over roughly five to six years.
Arougheti 在公司的业绩电话会上表示，他不知道“如何对那种技术的衰退和折旧曲线进行承销”。“至少目前，我们有点坚持只做我们最了解、且我们能力组合最深厚的那半个市场，”他说。
Arougheti said on the firm’s earnings call that he doesn’t know “how to underwrite the decline and depreciation curves on that type of technology.”“At least for the time being, we’re kind of sticking to the half of the market that we understand best and where our capability set is the deepest,” he said The bonds of CoreWeave Inc., the most well-known of the so-called neoclouds that lease GPUs to AI developers, plunged this month to levels that usually indicate distress.
所谓“新云”公司中最知名的 CoreWeave Inc. 的债券本月暴跌至通常预示困境的水平。该公司向 AI 开发者出租 GPU。为其债务购买保险的成本也飙升，反映出市场对支撑 AI 繁荣的借贷规模不断扩大感到担忧。
The cost to insure its debt has also spiked, reflecting concerns about the growing amount of borrowing underpinning the AI boom.