Hayes Warns AI Data Center Buildout Could Test Demand in 2027–2028
Редакция ЦОД:Главное
Arthur Hayes, co-founder and CIO of Maelstrom and former BitMEX CEO, said the current AI data center investment cycle could produce substantial overspend and excess compute capacity. Speaking at the Gamma Prime investment summit in Singapore, he identified late 2027 through 2028 as the likely stress period as facilities now under construction are delivered and contracted compute commitments come due.
- Hayes expects late 2027 through 2028 to be the main demand and contract-risk test for the current AI data center buildout.
- His several-trillion-dollar overspend estimate is an investor opinion, not a verified measure of AI data center CAPEX.
- The central risk is that newly delivered capacity could exceed the ability of customers to pay for previously contracted compute.
- Flop is planned as a payments and spot-market project for AI-agent compute, with a first-quarter 2027 launch expected.
Arthur Hayes has issued a bearish assessment of the AI infrastructure buildout, arguing that investment in AI data centers could result in several trillion dollars of inefficient spending and leave the market with excess compute resources.
At the Gamma Prime investment summit in Singapore, Hayes said the key test would likely emerge in late 2027 through 2028, when a large share of facilities now being built is expected to enter service. In his view, infrastructure providers will then seek payment from AI companies for compute capacity they had previously committed to procure, exposing whether demand and customer economics can support those commitments.
Hayes argued that a broad delivery wave could make compute capacity cheaper and more abundant. He named SpaceX, OpenAI and Anthropic as major prospective end users of compute resources, and stated that they are not yet profitable. That assertion was presented as Hayes’s view; the source provides no financial reporting to independently assess it.
The implications for data center operators and infrastructure investors would depend on whether capacity additions outpace paying demand. If Hayes’s forecast proves correct, the market could see short-term pressure on compute prices, utilisation at new facilities and the payment capacity of some AI-infrastructure customers. Conversely, lower-cost available compute could support wider deployment of AI applications. Both outcomes are analytical possibilities, not confirmed market results.
Hayes also introduced Flop, a planned crypto project positioned around payments and a spot market for AI-agent compute resources. Its launch is expected in the first quarter of 2027.
The comments highlight a potential mismatch between the rapid expansion of AI infrastructure supply and end users’ ability to pay for contracted capacity. If Hayes’s scenario materialises, new AI-oriented capacity could face near-term pressure on compute pricing, utilisation and customer credit quality, although this remains an investor’s thesis rather than evidence of an established oversupply.
- — / 7 октября 2026 г. · КитайскийПервоисточник →
Материал подготовлен редакцией ЦОД: Главное на основе указанных первоисточников.