GPU cloud in Sydney: who actually has the capacity right now

Try to order H100s in Sydney right now, and there is a good chance you will bump into either a capacity error on the hyperscaler’s console or a join the waitlist button at Neocloud. This is not a lack of information. This is a lack of capacity, which the majority of GPU cloud comparison material glosses over by assuming that “available” means “announced.”

They don’t. Here’s what is currently available in Sydney, in mid-2026, versus what is still under development.

Key takeaways

• There is capacity for H100 hyperscalers available in Sydney through AWS, Azure, and GCP, but it’s unreliable, and the latest versions are frequently available only under quota.
• There is a brand-new tier of sovereign neoclouds that have capacity in Sydney right now. SCX.ai and Sharon AI both have current deployments.
• Regulation of sovereignty is becoming more critical than latency or cost for those buying regulated data in Australia. This is what drives them to such local service providers.
• Macquarie’s IC3 Super West, the largest new physical capacity coming to Sydney, finished construction in August 2026. Its first tranche was expected to be commissioned this month, but isn’t confirmed live yet.
• The most common error that customers make in their planning concerning Sydney capacity is confusing “announced” and “available today.”

Who actually has GPU capacity in Sydney right now

There are three types of solutions in this market, and their maturity level is very different.

The hyperscaler claims to offer massive scale. However, their capacity varies widely, and you can just forget about getting access to the next-gen GPUs right away. The sovereign neoclouds make for a more consistent and compliant option. It provides an existing infrastructure that is currently available. It has been tailored to the regulations in Australia, and it went live in Sydney in 2026. The third tier provides a completely new physical data center capability. It is under construction and will go live shortly.

Hyperscalers in Sydney: real but constrained capacity

For years, Sydney has had GPU instances from AWS, Azure, and GCP. However, what has changed now is the extent of capacity strain, as growth outpaces the physical capacity to support it.

AWS operates P5 instances (H100 SXM) in ap-southeast-2: the Sydney region. These are available through EC2 Capacity Blocks for better availability. Availability is not guaranteed by AWS. 

One user reported on AWS’s own developer forum, re:Post, that they couldn’t launch G4 or G5 GPU instances in Sydney for several consecutive days, despite already holding an approved quota increase. That’s the practical reality behind the word “available” on a pricing page: a quota approval isn’t the same as hardware sitting ready to allocate.

The main APAC H100 option from Azure is the ND H100 v4 in australiaeast, and it seems to be the most reliable of the three hyperscalers in Sydney specifically. GCP offers H100 SXM through its A3 High instances in australia-southeast1. However, A3 Ultra (H200) instances are only partially available across APAC as of mid-2026 and are not guaranteed in Sydney specifically.

All of this does not make the hyperscalers a poor option. Those who have been using the infrastructure on AWS, Azure, or GCP would definitely benefit from sticking around: mature tools, established IAM and networking, and enterprise support contracts. The downside is that the latest hardware generations are available later and in smaller quantities in Sydney as opposed to US regions, and even quota approval will not ensure that you would be able to provision an instance right away.

The sovereign neocloud layer: who’s actually operational

This is the part of the Sydney market that’s changed the most in the last year, and it’s the part most existing coverage still treats as speculative rather than real.

SCX.ai brought Australia’s first sovereign AI infrastructure node online at Equinix’s SY5 data center in Sydney in January 2026. Worth noting: SCX.ai doesn’t run GPUs at all.

Its platform uses SambaNova SN40L ASIC processors, purpose-built for AI inference rather than general-purpose GPU compute. The company reports its infrastructure delivers roughly 2.5 to 5.6 times the performance per watt of GPU-based systems on selected inference workloads, though these are vendor-disclosed figures rather than independently verified benchmarks.

SCX.ai targets the end of 2026 for a second node. If your workload specifically needs NVIDIA GPUs, this isn’t your provider. If you need sovereign inference capacity and can work with an ASIC-based platform, it’s live today.

Sharon AI, in partnership with Cisco and NVIDIA, launched what they call Australia’s first Cisco Secure AI Factory in February 2026, hosted at NEXTDC’s S3 data center in Sydney. This one does run GPUs: 1,024 NVIDIA Blackwell Ultra GPUs, paired with Cisco UCS servers and Nexus Hyperfabric networking, with storage from VAST Data.

This is genuinely current-generation NVIDIA hardware, live in Sydney, not a roadmap item. Sharon AI has stated plans for additional clusters through 2026 and beyond.

ResetData, backed by ASX-listed Centuria Capital, is worth knowing about, but its Sydney story is less settled than SCX.ai’s or Sharon AI’s. ResetData’s confirmed operational AI factory runs in Melbourne, at Centuria’s 818 Bourke Street facility.

In August 2026, ResetData and Centuria secured A$165 million in GPU bridge financing from Macquarie Bank, along with a colocation agreement with CDC Data Centres and 72MW of dedicated power, pulling their broader deployment timeline forward by roughly two years. CDC operates campuses in both Sydney and Melbourne, but the initial 7MW allocation under this specific deal hasn’t been publicly tied to a Sydney site. Worth tracking, not yet a confirmed Sydney option.

Why sovereignty, not just latency, is driving this shift

Understanding why this neocloud tier exists at all matters as much as knowing it exists.

Australia’s Privacy Act, specifically Australian Privacy Principle 8, restricts how personal information can move offshore. On top of that, the Australia-US CLOUD Act Agreement took effect in January 2026, changing how US law enforcement can access data held by US-headquartered cloud providers, even when that data physically sits in an Australian data center.

That second point is the one buyers frequently miss. Running your workload in AWS’s or Azure’s Sydney region keeps your data physically in Australia, but it doesn’t fully remove the provider from US jurisdiction, since AWS and Microsoft are both US companies.

For financial services, healthcare, government, and defence workloads specifically, that distinction can matter more than latency or price. It’s the core reason a company like SCX.ai can build a real, well-funded business around infrastructure that’s meaningfully more expensive per unit of compute in some cases: it’s selling jurisdiction, not just hardware.

If your workload doesn’t carry that kind of regulatory sensitivity, this distinction matters less, and a hyperscaler’s Sydney region solves your latency needs perfectly well.

What’s coming but isn’t usable yet

The Macquarie IC3 Super West is the biggest capacity project in Sydney at the moment, making its deployment status crucial to track. This 47 MW data center, which is part of the 200 MW Sydney capacity pipeline of Macquarie Data Centres, topped out in December 2025, and construction finished on schedule and on budget in August 2026, with cabling and electrical fit-out now underway. 

The first 6MW tranche of IT load is launching this month, September 2026, according to the company’s own construction updates. This is the only data center that provides new AI-ready capacity in the northern zone of Sydney, as it has a hybrid air and liquid cooling system for GPU clusters.

This project signals how the market will evolve, but you cannot book this capacity yet. If your timeline precedes September 2026, this data center is not available at the moment.

Besides Macquarie, NEXTDC, AirTrunk, and CDC are extending their presence in Sydney; however, with an extended building timeline till 2027.

How to actually choose, given this landscape

The correct solution doesn’t depend as much on pricing and more on answering three main questions: what exactly you are deploying, the sensitivity of the data you process, and the timeframe for availability of resources.

For training purposes and workloads that do not have data residency requirements, the hyperscaler Sydney region is still the most feasible option. You will have access to robust tools and infrastructure, and even though capacity limitations are real, you can plan ahead using Capacity Blocks or the committed use model. The GPU shortage has increased rental prices in all regions during the last year, and this includes Sydney too.

For regulated workloads such as government workloads, healthcare, or financial services, as well as any other workloads that require compliance with APP 8 or any similar sovereignty requirement, Sharon AI’s Blackwell Ultra capacity is currently the best available solution if you specifically need NVIDIA GPUs. SCX.ai offers an ideal alternative if your workload focuses on inference and runs efficiently on ASICs.

Macquarie’s IC3 Super West is close enough now that it’s worth checking directly rather than ruling it out, especially if your timeline has any flexibility into September or October. Don’t build a deployment plan around it until you’ve confirmed the first tranche is actually live. 

Frequently asked questions

Does AWS have H100 GPUs in Sydney?

Yes. P5 instance type is available in the AWS ap-southeast-2 region (Sydney) based on H100 SXM. AWS cannot guarantee on-demand availability, with multiple days of shortages experienced by customers with an approved quota despite this.

What is a sovereign AI cloud, and why does it matter for Australian companies?

A sovereign AI cloud ensures that not only the storage of data but also the provider of the cloud infrastructure is outside the jurisdiction of foreign authorities. This is important considering the Australia-US CLOUD Act Agreement, which took effect in January 2026 and grants foreign authorities access to data stored by US-headquartered companies, even if they host that data in Australian data centers.

Is GPU capacity in Sydney more limited than in the US or Europe?

Yes, usually. Newer generations of NVIDIA hardware generally arrive in US territories first, followed by smaller shipments in APAC territories, such as Sydney. Azure’s ND H100 v4 seems to be the most consistently available hyperscaler offering in Sydney.

What’s the difference between a neocloud and a hyperscaler?

A hyperscaler (AWS, Azure, GCP) operates global-scale infrastructure in multiple regions with comprehensive ecosystems of services. A neocloud is usually a more specialized provider that focuses on GPU/AI infrastructure from the technical or regulatory angle that the hyperscalers do not cover.

When will Macquarie’s new Sydney data center be operational?

Construction finished on schedule in August 2026, and Macquarie’s deployment of its first 6MW of IT load is slated to commission this month, September 2026, the company’s latest updates confirm. Check current status directly before planning a deployment around it, as exact commissioning timelines frequently shift by days or weeks this close to launch. 

Sydney’s GPU capacity is shifting faster than most coverage reflects

As for one year ago, Sydney’s GPU cloud market was almost exclusively a hyperscaler story, and the market dismissed sovereignty as something for the future but not the present moment. But this is no longer the case. Both SCX.ai and Sharon AI now deliver active capacity, not just roadmaps, and providers are aggressively building additional capacity for later this year.

The key practical point for anyone planning around Sydney’s capacity: ask yourself if a provider’s infrastructure works right now and not just on paper. This is what makes all the difference in this market.

When it comes time to compare current GPU capacity and pricing between different providers, browse live options across ComputeStacker’s verified index.