Foreign banks, pension funds, sovereign wealth funds and insurers from Asia and the Gulf are increasingly funding AI computing capacity in North America, JPMorgan Asset Management’s Charles Wu told the SuperReturn conference in Singapore.
Capital expenditure in the sector stands at almost $1 trillion.
“Compute training assets are still largely in North America,” said Wu, the bank’s head of Asia Pacific alternatives institutional client strategy. “On the banking side and also on the asset management side, the capital providing the funding for these assets is largely becoming more global.”
Where the money comes from
US hyperscalers are funding much of their AI capital expenditure through operating cash flow, but are also raising roughly $250 billion from the bond market and about the same again from bank loans, according to Wu.
Everything leads back to five companies
Wu’s warning concerned what sits behind the paper.
“The reality is a lot of these funding securities, whether you’re talking about investment-grade bonds, 144a, private placements, even private credit structured deals — a lot of them seem to all triangulate to basically five key hyperscaler parties,” he said. “Either as a guarantor, as a tenant, as a funding source, or as a customer.”
He said his firm was also monitoring lease commitments and the risk that some financing is being raised off balance sheet.
Jean-Christophe Aubert, senior director of infrastructure investments at PSP Investments, framed the caution from the allocator’s side. “Concentration is a big point, but also for us, given the amount of capital that we would have to put to work, is trying to understand how can we ultimately realise those assets going forward.”
That is the exit question. Data centres built for a small number of tenants are difficult to sell if those tenants change strategy.
Why capacity should eventually spread
Several panellists expected Asia to follow North America closely, for a structural reason.
Inference — the actual use of AI models, as opposed to training them — benefits from proximity to end users. That drives demand for infrastructure wherever the users are.
“Asia is equally important as it plays an interesting role in both the demand side of the equation and the supply side through the physical stuff,” said Mohsin Pirzada, head of funds at the Qatar Investment Authority. “What we’re starting to see is a decentralisation, and that’s being driven by end consumer demand” as well as data sovereignty and security.
That is the mechanism by which compute moves outward: latency, plus national rules about where data can sit.
Energy is the constraint, and turbines are the specific problem
Panellists agreed power remains the major bottleneck. With gas the go-to source, a shortage of gas turbines is holding back data centre scaling — an unusually specific supply chain constraint, and one that bears on anywhere planning gas-fired capacity for AI.
Chenhua Shen, fund partner at I Squared, said Asia’s less liberalised power markets make utility access harder.
“It becomes a major obstacle for Asia to leapfrog,” she said. “But we are still very optimistic in a way because Asia enjoys very abundant renewables.”
Article originally featured on 2026 Bloomberg. Via Moneyweb.
