Sharon AI Holdings, an Australian provider of artificial-intelligence cloud services, said Thursday it signed a five-year agreement worth $1.32 billion to supply cloud-computing capacity to a global AI laboratory. The customer’s name wasn’t disclosed.
Under the deal, Sharon AI will deploy cloud solutions at a data center in New Zealand and expects to begin recognizing revenue in the first or second quarter of 2027. The lag between contract and revenue is typical of the AI cloud business, where operators sign multi-year commitments for capacity that takes a year or more to build and connect to power.
The contract is the latest in a string of large pre-sales that have come to define the specialist AI cloud industry. Sharon AI said its AI Factory network now has 132 megawatts of total capacity, with 116 megawatts already contracted to end customers. By the middle of next year, it expects to have more than 62,000 Nvidia graphics processing units deployed across its sites, a fleet large enough to rank the company among the bigger independent GPU cloud operators outside the United States.
The arrangement is a bet that AI laboratories, and the investors behind them, will keep renting computing power rather than build their own data centers. Specialist providers such as CoreWeave have built multibillion-dollar businesses on that premise, signing up AI companies for capacity that is paid for as it is used. Sharon AI is trying to replicate the model from the other side of the Pacific, using New Zealand’s land and power supply as the selling point.
Power, not chips, has become the binding constraint on AI infrastructure. Countries with cheap electricity, available land and fast permitting have turned into magnets for data-center investment, and New Zealand has been courting the industry with the promise of renewable generation. Analysts said the choice of location reflects a broader search for energy as hyperscale data centers in the United States run into grid limits and interconnection queues stretch for years.
The financing behind such projects has shifted too. Specialist cloud operators now raise money through a mix of private equity, debt and customer prepayments, with lenders taking comfort in signed contracts that guarantee future cash flow. The 116 megawatts of committed capacity gives Sharon AI a revenue backlog that banks can underwrite, which is how a company with a relatively short operating history lands a deal of this size.
The risks are as large as the opportunity. A single contract accounts for the bulk of the new revenue, leaving Sharon AI exposed to a customer that could renegotiate or walk away before the capacity is built. The provider also carries the usual execution risk of construction and chip delivery, and the financing burden of building 132 megawatts of data-center space on contracted revenue that won’t appear on the income statement until 2027.
The business model works until the capacity is built and the customer pays, one analyst who tracks the sector said. The pre-sales look great on paper, but the whole industry is running a race between construction schedules and AI demand, and some operators will lose that race.
The deal also shows how AI infrastructure money is spreading beyond the biggest U.S. technology companies. A year ago, cloud capacity deals of this size were signed almost exclusively by hyperscalers and a handful of American specialists. Now smaller operators in Australia, Europe and the Middle East are signing contracts of their own, funded by investors who see the GPU shortage as a generational opportunity.
The competitive field in the Asia-Pacific region is crowded. Amazon Web Services and Microsoft operate data centers in Australia and New Zealand, and local telecom companies have entered the AI cloud market with their own offerings. What the specialists bring is speed: a single-purpose operator can move faster than a hyperscaler balancing thousands of products, and can offer capacity at prices that reflect the shortage rather than the long-term cost of building.
The structure of the deal reflects how scarce computing capacity has become. Customers of specialist AI clouds typically sign up for GPU clusters months in advance, pay for them on a take-or-pay basis once delivered, and expect the provider to handle the procurement, cooling and maintenance. That puts Nvidia’s allocation decisions at the center of the business: a provider that cannot secure chips cannot sign contracts, and one that signs too many can find itself paying penalties on undelivered capacity. Sharon AI’s willingness to commit to 62,000 GPUs by mid-2027 implies it has already lined up supply agreements with Nvidia, though the company didn’t say so in its announcement.
Sharon AI’s announcement framed the agreement as proof that demand for AI compute remains intense even as investors rotate out of chip stocks and question whether the industry is overbuilding. The 62,000-GPU deployment target, set for mid-2027, suggests the company expects the boom to run for years. Whether that expectation survives contact with the realities of construction schedules, interest rates and a single large customer is the question investors will be watching.
