Neocloud operator Sharon AI has signed a $373 million infrastructure deal for 2026, marking its expansion into the Australian market.

A new breed of cloud provider is quietly building physical infrastructure inside Australian data centres, and it isn’t AWS, Microsoft, or Google. Neoclouds—specialised operators that rent out GPU capacity purpose-built for AI training and inference—pulled in more than US$25 billion in global revenue during 2025 and are now expanding at a pace that has caught even seasoned cloud analysts off guard, according to Synergy Research Group. In Australia, the clearest proof point is Sharon AI, a Nasdaq-listed neocloud that just signed a US$373 million five-year AI cloud services deal on August 4, 2026, its infrastructure running inside NEXTDC data centres in Melbourne and Sydney.

For Australian IT buyers accustomed to choosing between AWS, Azure, and Google Cloud, this represents a genuinely new fork in the road. Neoclouds do not attempt to be everything; they offer no managed databases, serverless functions, or global SaaS ecosystems. Instead, they focus on a single product: renting Nvidia GPUs at scale. They are doing so inside Australian soil precisely when hyperscaler AI capacity is tight and tightening further. This examination outlines what neoclouds actually are, why Australia became their proving ground, what the numbers reveal about pricing and growth, and what enterprise cloud teams should monitor next.

A neocloud is a cloud provider built around a single product: renting GPU compute for AI training and inference, rather than offering the broad service catalogue typical of hyperscalers. CoreWeave, Nebius, Lambda, Crusoe, and Sharon AI dominate 2026 market reports. None sell equivalents to S3 storage buckets, Lambda functions, or managed Kubernetes control planes the way AWS does. What they provide are racks of Nvidia H100, H200, B200, and Blackwell Ultra B300 GPUs, wired into high-bandwidth networking and rented by the hour, month, or under multi-year committed contracts.

The category emerged because hyperscalers could not supply enough frontier GPU capacity quickly enough once the AI training boom accelerated. Gartner now predicts neocloud providers will capture 20% of the US$267 billion AI cloud market by 2030—a forecast that would have sounded implausible three years ago when the term “neocloud” barely existed outside specialist circles. Synergy Research Group’s tracking confirms the segment’s rapid ascent: revenues hit US$9 billion in the fourth quarter of 2025 alone, a 223% year-over-year increase that pushed the full 2025 total past US$25 billion.

Market size estimates vary depending on methodology, but every recent report points the same direction. Mordor Intelligence pegs the 2026 neocloud market at US$35.22 billion, up from US$24.07 billion in 2025, with a 46.37% compound annual growth rate projected through 2031. Research and Markets separately estimates 2026 revenue at US$42.17 billion, rising to US$253.79 billion by 2030. The spread between estimates reflects how new and fast-moving this category is, not disagreement on its trajectory.

The single most concrete Australian data point in this story is Sharon AI Holdings, a Nasdaq-listed neocloud operator (ticker SHAZ) that announced a US$373 million five-year AI cloud services agreement with a global AI platform customer on August 4, 2026. The initial deployment under that deal uses 2,048 Nvidia Blackwell Ultra B300 GPUs, and the company says it plans to scale its total AI Factory fleet from 62,000 to 64,000 GPUs worldwide by mid-2027, according to the official PRNewswire announcement.

What makes this an Australian story rather than a generic US neocloud press release is where the hardware physically sits. Sharon AI’s Australian GPU capacity is built inside NEXTDC’s M3 facility in Melbourne and its S3 facility in Sydney, with reported contracted capacity of 132MW total, 102MW of it already under contract, and more than 55,000 GPUs due for deployment in the country by mid-2027. As recently as September 2025, Sharon AI’s Australian footprint was tiny by comparison, just 432 GPUs and 195 CPUs online, according to public filings. That means the jump to tens of thousands of GPUs represents one of the fastest AI infrastructure build-outs anywhere in the country’s history, not a gradual rollout.

Sharon AI isn’t stopping at Melbourne and Sydney capacity leases either. In June 2026 the company signed a six-year, roughly US$4.9 billion infrastructure collaboration with Nvidia to build a dedicated 72MW data centre in Australia, targeting up to 40,000 Nvidia GB300 GPUs built on Nvidia’s DSX AI factory reference design. Combined with earlier deals, including a reported US$1.25 billion cloud capacity agreement with ESDS Software Solutions and a US$950 million agreement with an unnamed global tech company, Sharon AI has assembled one of the largest single-operator AI compute commitments physically located in Australia.

It would be easy to assume AWS, Azure, and Google Cloud could just build more of their own GPU capacity and squeeze the neoclouds out. The problem is that the constraint isn’t capital, it’s physical: GPUs, power, and grid connections. Synergy Research’s own cloud infrastructure numbers show the “Big Three” still dominate overall cloud spend, AWS at 28%, Azure at 21%, and Google Cloud at 14% of the worldwide cloud infrastructure market in the first quarter of 2026, per Statista’s chart of Synergy’s data. But that dominance is in general-purpose cloud spend, not necessarily in frontier AI accelerator availability, where neoclouds have carved out a defensible niche by moving faster on GPU procurement and datacentre power deals than hyperscalers accustomed to longer planning cycles.

Overall cloud infrastructure spend is also still climbing sharply, giving both hyperscalers and neoclouds room to grow simultaneously rather than fight over a fixed pie. Synergy reported that global enterprise cloud infrastructure spending reached US$143 billion in the second quarter of 2026 alone, an increase of more than US$43 billion year over year and the highest growth rate in eight years, according to Synergy’s Q2 2026 cloud market report. That kind of overall growth is exactly why analysts like CIO.com frame neoclouds not as a threat that will topple hyperscalers, but as a parallel infrastructure tier that soaks up demand hyperscalers can’t fully serve on their own timeline.

The pricing gap is the single most cited reason enterprises look at neoclouds at all. Published on-demand rates in mid-2026 show real, verifiable spreads between neocloud operators and, by extension, against the premium hyperscalers typically charge for equivalent accelerator instances. Lambda publishes the lowest B200 on-demand rate in the market at US$6.69 per GPU-hour. Crusoe’s H200 rate sits at US$4.29 per GPU-hour, with its H100 rate at US$3.90. Nebius’s H100 PCIe on-demand rate has held around US$3.85 per hour. CoreWeave, which publishes the most complete pricing card of the group, charges a premium at US$6.16 per GPU-hour for H100 instances, a rate industry trackers describe as running roughly 60% above the cheapest published Nebius and Lambda equivalents.

Separate industry analysis has estimated that equivalent GPU instances on neoclouds can run 40% to 66% cheaper than comparable AWS or Azure offerings, though exact like-for-like comparisons are difficult because hyperscalers rarely publish raw GPU-hour rates the way neoclouds do. The following table compares the most commonly cited published on-demand rates as of August 2026.

That pricing dynamic doesn’t mean neoclouds are simply “cheap AWS.” Contrary to popular perception, neoclouds operate on distinct economic and technical premises that separate them from general-purpose cloud providers. Their business models rely on heavy upfront capital expenditure for specialized hardware and power infrastructure, and their profitability hinges on high utilization rates and long-term committed contracts. Enterprises must therefore weigh raw compute savings against the broader ecosystem advantages hyperscalers provide, including integrated security frameworks, compliance certifications, and cross-service interoperability. The optimal choice ultimately depends on workload specificity: pure AI training and inference workloads benefit most from neocloud pricing and agility, while applications requiring a comprehensive suite of cloud-native services may still lean toward established hyperscalers.