In 2008, the world’s collateral failed. Mortgage bonds, the bedrock of the financial system, turned out to be worth less than paper. In 2026, we are watching the opposite experiment: a company borrowing US$356 million against a new kind of collateral, one that may turn out to be the most valuable commodity of the AI age.
SharonAI Holdings Limited (NASDAQ: SHAZ) closed a senior secured debt facility at a fixed rate of 9.95%, arranged by Jarden Australia with Goldman Sachs and private credit funds participating. The security backing the loan? A fleet of NVIDIA GPUs, some 68,000 of them expected to be operational by mid-2027, powering gigawatt-scale AI factories across the Asia-Pacific region.
How the machine works
The structure is worth understanding. The debt sits in a special purpose vehicle (SPV) that owns the GPU fleet, and the facility is secured against the chips themselves plus the cash flows from customer contracts. Sharon AI says it has an offtake book worth US$8.6 billion, a pipeline of customers waiting to rent the compute those GPUs will produce.
Chief executive James Manning has been on a tear. The company says it has raised US$2.6 billion in the last ten months, including a US$1.6 billion round in June. The new debt brings the total war chest to a level that would have been unthinkable for a regional data center operator five years ago. The APAC buildout, long talked about as the AI race’s second front, is now being financed like infrastructure: with debt, against collateral, at scale.
Why banks accepting GPUs as collateral is a milestone
For a bank to accept GPUs as collateral, it has to believe three things: that the chips will hold their value, that there will be customers to rent them, and that the operator can keep them running. All three are bets on the AI boom continuing, and on the idea that compute is now as fundamental as electricity or shipping.
It also changes who can build. Equity is expensive and slow. Debt is cheap and fast, at least when you have collateral the lender believes in. If GPUs are now bankable assets, the universe of companies that can build AI factories just got much bigger. The neocloud model, renting out AI compute without owning the whole stack, is graduating from venture-backed experiment to financed infrastructure.
The sovereign AI angle
There is a second story here, and it is about geography. Sharon AI is building across APAC, and governments across the region are racing to secure their own AI infrastructure. Sovereign AI, the idea that every country needs its own compute capacity, has moved from talking point to procurement strategy. A financed, collateral-backed buildout is exactly how you deliver it at speed.
The parallel to energy is hard to miss. Data centers are the new power plants, GPUs the new turbines. The companies that finance them like utilities, with long-dated debt against hard assets, may end up owning the 21st century’s most important infrastructure. Sharon AI is making that bet explicitly.
What this signals for the AI buildout
The neocloud model is graduating. When banks will lend against your GPUs, you are no longer a startup. You are infrastructure.
Watch the collateral math. The facility’s 9.95% fixed rate tells you what lenders think of the risk. That is high-yield territory, not investment grade. The bet is real, but so is the price.
APAC is the buildout’s second front. The US got the first wave of AI factories. The second wave is being built in Asia-Pacific, financed locally, serving sovereign demand.
Sovereignty sells. Governments want their own AI capacity. Companies that can finance and deliver it will find no shortage of customers.
The takeaway
Chips as collateral. It sounds like science fiction, but it is now a US$356 million fact. The AI buildout is entering its infrastructure phase, and the companies that master the financing will matter as much as the ones that master the models.
Celebrate the season: the city’s 2026 holiday tree and lights celebrations are worth a visit. And if you are hungry, these are the best lobster spots to try.
