How Sharon AI funds 56,000 NVIDIA GPUs

At a glance

• Sharon AI (NASDAQ:SHAZ) is deploying up to 40,000 NVIDIA GB300 GPUs under its six-year NVIDIA deal, lifting its AI fleet to 132MW, 102MW already contracted.

• The chips cost well over US$2 billion, far more than Sharon AI could buy outright.

• NVIDIA and lenders fund them for a share of the revenue, now a named program with Sharon AI and Firmus its first two partners.

• Unlike IREN and Firmus, which largely build offshore, Sharon AI builds inside Australian data centres.

• Money is not the constraint. Power and grid connection are.

What Sharon AI is building

Sharon AI is a neocloud company that buys large fleets of NVIDIA chips, installs them in data centres, and rents the computing power to businesses, startups and researchers. Its collaboration with NVIDIA, announced on 12 June 2026, lifts planned capacity to 132MW and tens of thousands of GPUs.

The hardware bill is the real challenge. The chips alone cost well over US$2 billion, more than Sharon AI could fund from its own balance sheet, so they are never bought outright. How that spend gets funded is the substance of the deal, and it follows a template now well established overseas.

How the buildout gets funded

Neoclouds do not pay cash for their chips. Vendors, lenders and investors fund the hardware in exchange for a share of what it earns.

NVIDIA’s involvement is the clearest example. It backs Sharon AI’s commitment with credit support and takes a share of the cloud revenue the chips generate, on top of selling the hardware. Sharon AI avoids pre-funding the whole bill from its own balance sheet, and NVIDIA earns more only when the chips are well used. We set out the mechanics in our coverage of the deal. In early July NVIDIA packaged this arrangement into a named revenue-share and credit-support program, presenting Sharon AI and Firmus's 360MW Batam campus as its first two public partners, together committing more than 430MW and up to 210,000 GPUs across Australia and Indonesia.

Vendor support is one structure among several. The global neocloud sector has built a menu of ways to fund GPUs, and most large operators use more than one.

NEOCLOUD FINANCING STRUCTURES

Source: Primary disclosures and Financial Times reporting, June 2026.

The scale of this is hard to overstate. Roughly US$120 billion of AI infrastructure debt now sits in these separate vehicles worldwide. Australia is in the picture as a financier too, not just an operator: Macquarie is among the lenders backing global neoclouds, and Firmus is built on a US$10 billion facility from Blackstone and Coatue.

Sharon AI is now adding the investor leg. The Australian Financial Review reported on 14 June 2026 that it has hired Goldman Sachs to raise about $1 billion through a combination of new shares and convertible debt, sold privately to institutional investors, to fund the NVIDIA buildout. The terms had been sent to investors, and the raising had not been announced publicly at the time of writing.

Where the capacity actually lands

The financing model is not new for Australian-founded operators, and Sharon AI is not the only one building at home. Firmus is developing 1.6GW across five Australian sites, and ResetData runs sovereign capacity domestically. The real divide is between those building in Australia and IREN, the largest Australian-founded neocloud, which runs its halls offshore.

AUSTRALIAN-FOUNDED NEOCLOUDS

IREN, Sydney-headquartered and one of the larger players globally, runs its AI halls in the United States and Canada, scaling on its own multi-billion-dollar NVIDIA neocloud deal. Much of the Australian-founded neocloud capacity built so far has been built abroad, for offshore customers. Sharon AI’s 132MW is going into existing Australian data centres, anchored by its cluster at NEXTDC’s M3 facility in Melbourne, with 102MW already contracted to customers.

Power is the real constraint

Financing solves whether the chips can be bought. It does nothing for the harder problem, which is getting them powered and switched on.

A chip order ships in months. A data centre takes three to five years to build, and in Australia, connecting it to the grid alone can take around two years, on AEMO’s own pipeline numbers. With 102MW of Sharon AI’s 132MW already contracted, customer demand is clearly there. What would speed Australia’s AI buildout is not more money chasing the sector, which is plentiful, but faster grid connections and firmer power.

What to watch

The reported Goldman Sachs raising is the first thing to watch. The AFR put pricing as soon as 17 June, with the ASX debut expected in July, and formal terms would confirm how the chips get paid for. Beyond that, Sharon AI has not named the sites for its new capacity, and its next quarterly results and ASX listing documents are where the site detail and build timing should appear.