Amazon Web Services is expanding GPU capacity in its Sydney region with new clusters housing NVIDIA H100 and Blackwell processors, reducing lead times for Australian enterprises requiring high-performance AI compute.
Tech & Ideas Desk is a contributing writer covering tech and public affairs for The Sydney Times.
Amazon Web Services is expanding GPU capacity in its Sydney region with new clusters housing NVIDIA H100 and Blackwell processors, reducing lead times for Australian enterprises requiring high-performance AI compute. The expansion adds roughly 20,000 additional GPUs to the Sydney region, increasing total AI-optimised compute capacity by an estimated 60 percent. The new clusters are housed in AWS's existing data centre facilities in Sydney, with power and cooling infrastructure upgraded to support the higher density of GPU server racks.
The timing aligns with the easing global AI chip shortage and AWS's broader strategy to reduce regional disparities in AI compute availability. Australian enterprises have historically faced longer wait times and higher spot prices for GPU instances than counterparts in the United States or Europe, a gap that widened during the 2024 to 2025 chip shortage. The Sydney region expansion narrows that gap, though AWS Australia acknowledges that capacity constraints will still occur during peak demand periods.
Alongside the NVIDIA GPU expansion, AWS is increasing availability of its custom AI chips, Inferentia2 and Trainium2, in the Sydney region. The custom silicon is designed for inference and training workloads respectively, and it offers better price-performance than NVIDIA GPUs for workloads that can be optimised for AWS's own hardware architecture. Trainium2 is capable of training large language models at roughly half the cost of equivalent GPU clusters, while Inferentia2 delivers higher throughput per dollar on inference tasks for models that have been compiled for the AWS Neuron SDK.
The custom silicon option is particularly relevant for Australian enterprises running high-volume inference workloads such as customer service chatbots, document classification, and fraud detection. Those tasks involve processing large numbers of relatively simple requests, which is where Inferentia2's throughput advantage shows up most clearly. AWS is offering migration support for enterprises that want to move existing inference workloads from GPU instances to Inferentia2, with the company claiming that most models can be adapted with less than a week of engineering work.
AWS has been the primary cloud provider for Australian Government agencies moving to AI-enabled services, and the Sydney region expansion strengthens its position in that segment. The Department of Home Affairs, Services Australia, and the Australian Taxation Office are all AWS enterprise customers with data residency requirements that restrict their AI workloads to Australian infrastructure. The new GPU clusters in Sydney will support those agencies' plans to deploy frontier models for document processing, citizen services, and fraud detection without sending data offshore.
The expansion also supports AWS's growing presence in the Australian financial services sector, where data residency requirements under the Privacy Act 1988 and the forthcoming AI regulation framework require sensitive customer data to remain within Australian borders. Westpac, Commonwealth Bank, and Macquarie Bank are all AWS enterprise customers with significant AI workloads, and the Sydney region GPU expansion gives them additional capacity to scale those workloads without moving data to offshore regions.
Google Cloud and Microsoft Azure are both expanding their AI infrastructure in Australia, with Google committing to additional TPU capacity in Sydney and Microsoft planning new data centre regions in Melbourne and Brisbane that will include AI-optimised compute. The competitive pressure is reducing AWS's pricing advantage in the Australian market, with spot prices for GPU instances falling by roughly 20 percent over the past year. The price competition benefits Australian enterprises by reducing the cost of AI experimentation and production deployment.
The competition is also driving service differentiation. Google Cloud is emphasising its TPU v5 infrastructure and Vertex AI platform for enterprises that want to build on Google DeepMind's Gemini models, while Microsoft is bundling Azure OpenAI Service with its enterprise productivity suite and offering integrated security and compliance controls that appeal to regulated industries. AWS is responding with expanded AI service offerings including Bedrock, SageMaker, and new security tools designed specifically for AI workload protection. Explore more cloud infrastructure analysis at the Tech & Ideas hub
For AWS Sydney region and AI service details, see AWS Australia. NVIDIA H100 and Blackwell product information is at NVIDIA data centre. AWS Inferentia2 and Trainium2 documentation is published at AWS Neuron.
Direct inquiries, corrections, or documentation concerning this dispatch to our editorial newsroom desk.
Australian AI startups are facing a funding contraction in 2026 as venture capital firms tighten due diligence requirements, with pre-revenue companies facing the steepest reduction in available capital.
The global AI chip shortage is easing in 2026 as NVIDIA H100 and Blackwell allocations stabilise, but export controls on advanced GPUs to Southeast Asia are redirecting demand toward alternative suppliers and custom silicon.
AI-generated code ownership disputes are emerging in Australian software contracts, as enterprises and vendors negotiate intellectual property rights for code produced by frontier models during outsourced development projects.
AI regulation compliance costs are rising as enterprise legal budgets expand to cover impact assessments, conformity documentation, and ongoing monitoring requirements for high-risk AI deployments.