Ares Management is taking a cautious approach to financing graphics processing units (GPUs) and chips used in artificial intelligence infrastructure, citing uncertainty over the technology’s depreciation and the complexity of underwriting such transactions.
The alternative asset manager estimates that AI and digital infrastructure will require more than $5 trillion in capital expenditure through 2030, including about $3 trillion for GPU and chip financing and $2 trillion for physical data centres and adjacent power generation and transmission infrastructure.
But Ares sees the latter as a more attractive investment opportunity, given the relative simplicity of underwriting physical assets and its existing capabilities in the sector.
“I, for one personally, do not know how to underwrite the decline and depreciation curves on that type of technology,” its management said during the firm’s third-quarter earnings call on Thursday (Oct.29), referring to GPU and chip financing.
The firm also pointed to the complexity and concentration risks involved in such transactions, as investors weigh how quickly chips could lose value as technology advances and newer generations of processors enter the market.
By contrast, Ares said the $2 trillion opportunity in physical infrastructure represents a large addressable market relative to available capital, offering attractive risk-adjusted returns across debt and equity investments.
The firm is targeting data centre investments in markets including Tokyo and Osaka, alongside other global hubs, through pre-leased projects with investment-grade hyperscaler counterparties and 12- to 15-year leases.
“With that risk boxed, we feel like we can generate really attractive returns regardless of the ultimate, you know, technology outcomes here,” management said, referring to its approach to physical infrastructure.
The comments come as financing demand for AI infrastructure grows, while investors become more discerning about counterparty risks and pricing amid market volatility and higher funding costs.
