Asia-Pacific’s AI infrastructure builders have a financing problem. The hardware they need — racks of Nvidia H100s and their successors — costs hundreds of millions. Traditional lenders balk at the depreciation cycle.
A GPU bought today is worth half as much in three years; a data centre shell holds its value better. That gap is the reason a new joint venture between USD.AI and BSQ Capital Partners exists.
The venture is small for now. It targets $300 million in senior secured, GPU-backed financings across the region. The structure matters: the hardware itself is collateral, not just an operator’s cashflows.
That gives lenders a tangible asset to seize if a borrower defaults. For operators, it means they can fund clusters without tying up equity.
If performance milestones are met, the venture can scale to $1 billion. BSQ Capital Partners is the regional anchor. Based in Singapore, it calls its focus “next-generation” digital infrastructure — AI data centres, interconnection facilities, the power and cooling systems that make them run.
It knows the markets from Japan to Southeast Asia, where demand for training and inference hardware is climbing fast. USD.AI built a platform to underwrite and manage compute assets. The two firms are betting that the financing architecture can catch up to the hardware buildout.
That buildout is uneven across the region. Some markets have aggressive government incentives for AI compute.
Others have cheap power but weak interconnection. All of them face the same capital constraint: banks that lend against real estate, not graphics cards. The joint venture is designed to fill that specific hole.
The $300 million is an initial deployment. It is not a final number.
The path to $1 billion depends on hitting performance milestones — a standard structure in venture-style finance, but less common in hardware-backed lending. That detail matters. It means the venture is testing its underwriting model at a moderate scale before committing larger sums.
For lenders, the appeal is clear. Senior secured status puts them first in line. GPU-backed collateral is liquid — there is a secondary market for H100s and their successors.
That is not true of most infrastructure assets. An empty data centre shell is hard to sell quickly.
A cluster of H100s can be redeployed within weeks. For operators, the venture offers a way to avoid diluting equity or taking on unsecured debt at punishing rates. The capital outlay to stand up a cluster runs into the hundreds of millions.
Few operators have that cash on hand. The joint venture gives them a path to fund the hardware without tying up their balance sheets.
Asia-Pacific is a battleground for AI compute capacity. The financing mechanisms to support that buildout have lagged behind. This venture is one attempt to close the gap.
Whether it scales to $1 billion depends on whether the first $300 million performs. That is a question of underwriting, not ambition.
