Ares Management LP is steering clear of financing the high-performance processors that power advanced AI models, and is instead putting money into the data centers that house those chips.
According to a person familiar with the matter, the asset manager worries that investing in GPUs could leave it stuck holding assets that become obsolete shortly after they hit the market. The person asked not to be identified because the information is private.
CEO Michael Arougheti, speaking on an analyst call Thursday about data center financing versus GPU chip financing, said: "Frankly, due diligence on data center projects is simpler, and the relative returns are quite attractive." He added that this is not a rejection of the GPU financing business, but that opportunities in data centers and the accompanying power infrastructure are more appealing to them.
Arougheti noted that the data center financing market is enormous, while the capital available to deploy is relatively limited. The firm estimates that AI-related capital expenditure demand is around $5 trillion, of which about $2 trillion goes to physical AI facilities and supporting infrastructure, and $3 trillion to purchasing GPU chips.
His remarks come as investor anxiety over the AI sector intensifies. In recent weeks, concerns that massive AI spending will not match future revenue have sent related stocks and corporate bonds lower. Nearly all major US banks and alternative asset managers have joined this AI financing boom, with deals including Broadcom's roughly $60 billion chip debt financing (the chips to be leased to Anthropic) and SpaceX's plan to raise about $40 billion to buy Nvidia chips.
Chips cost far more than the data centers that house them, yet typically depreciate over just five to six years. On the earnings call, Arougheti said he cannot conduct due diligence to assess the value decay and depreciation curves of such technology. "At least for now, we choose to deepen our presence in the half of the market we know best and where we have the deepest accumulated expertise," he said.
Prominent compute cloud provider CoreWeave Inc., which rents GPUs to AI developers, saw its bonds plunge this month, with prices now in distressed territory; the cost of insuring its debt has also soared, reflecting market worries about the growing scale of debt underpinning the AI boom.
Comments