Compute comparer

Where to rent GPUs. Three lanes: Australian-owned and operated (sovereign), Australian data region (foreign-owned but data stays here), and global marketplaces (cheapest, least sovereign). Filter by Location to keep it local. Prices in AUD (converted at AUD/USD 0.70, 2026-07-21); the Australian-owned clouds are enterprise-scale and priced by quote, so a solo creator's practical self-serve local path is a hyperscaler Australian region or a global marketplace filtered to Sydney. Rates drift, so re-verify before a big batch.

Not sure which box? Tell me two things and I'll point at the best one and why, so you don't have to weigh all of them.
Pick two, then press Recommend.
Location
Best for
Spin-up
Provider
Sort
AU ownedSustainable Metal Cloud (Firmus)batchcontact sales

H100 / A100 clusters

GPU RAM80 GB/GPU (H100/A100 cluster)
A$3.0/hr

Sovereign and about half the carbon of air-cooled clouds; ranked by SemiAnalysis alongside the hyperscalers. Built for clusters, so contact sales; H100 around A$3/hr at commitment.

Offering: immersion-cooled bare metal · Billing: committed
Spin-up: slow · Region: Sydney (also Singapore, India)
visit provider →
AU ownedSharon AIimmediatecontact sales

H100 / H200 / B200

GPU RAM80-192 GB/GPU (H100/H200/B200)
quote

Australia's largest sovereign AI cloud buildout, NVIDIA and Cisco partnered. Enterprise scale; talk to them for access and pricing.

Offering: sovereign neocloud · Billing: contract
Spin-up: medium · Region: Sydney and Melbourne (NEXTDC)
visit provider →
AU ownedMicron21 (mCloud)immediatecontact sales

A100 / H100 / H200

GPU RAM40-141 GB (A100/H100/H200)
quote

100% Australian owned, Tier IV data centre, IRAP-assessed and ISO/PCI certified. GPU passthrough or time-sliced. Contact for a quote.

Offering: sovereign GPU cloud · Billing: quote
Spin-up: medium · Region: Melbourne (Tier IV)
visit provider →
AU ownedDUG Technologybatchcontact sales

HPC GPU nodes

GPU RAMHPC nodes (varies)
quote

Perth-based, immersion-cooled (DUG Cool), scientific HPC heritage. Best when you have a big batch and want an Australian HPC partner.

Offering: HPC-as-a-service · Billing: quote
Spin-up: slow · Region: Perth
visit provider →
AU dataAWS (ap-southeast-2)immediateself-serve

H100 (p5)

GPU RAM80 GB (H100)
A$17.6/hr

Data stays in Sydney, but AWS is US-owned. On-demand list is dear; spot and savings plans cut it a lot. Self-serve if you know AWS.

Offering: on-demand / spot · Billing: per-second
Spin-up: medium · Region: Sydney
visit provider →
AU dataGoogle Cloud (australia-southeast1)immediateself-serve

H100

GPU RAM80 GB (H100)
A$17.8/hr

Sydney region, US-owned. Similar list price to AWS; spot is much cheaper for restartable work.

Offering: on-demand / spot · Billing: per-second
Spin-up: medium · Region: Sydney
visit provider →
AU dataAzure (Australia East)immediateself-serve

ND H100 v4

GPU RAM80 GB (H100)
A$20.0/hr

Sydney region, US-owned; deepest government and finance adoption. Priciest list of the three, strong compliance story.

Offering: on-demand / spot · Billing: per-second
Spin-up: medium · Region: Sydney
visit provider →
globalVast.aibatchself-serve

RTX 4090

GPU RAM24 GB (RTX 4090)
A$0.45/hr

Best $/VRAM-hour for restartable draft batches. You can filter to Australian hosts for a semi-local, cheap, self-serve option.

Offering: interruptible · Billing: per-second
Spin-up: medium · Region: marketplace (filter to Australian hosts)
visit provider →
globalRunPodbatchself-serve

RTX 4090

GPU RAM24 GB (RTX 4090)
A$0.5/hr

Reliability step up from Vast, real network volumes so weights persist. Has a Sydney region; pick it for closer, self-serve compute.

Offering: community cloud · Billing: per-second
Spin-up: medium · Region: many (Sydney region exists)
visit provider →
globalRunPodimmediateself-serve

A100 80GB

GPU RAM80 GB (A100)
A$2.0/hr

All-round hero tier; network volumes for 40GB+ weights.

Offering: secure cloud · Billing: per-second
Spin-up: medium · Region: many (Sydney region exists)
visit provider →
globalRunPodimmediateself-serve

H100

GPU RAM80 GB (H100)
A$4.3/hr

Top self-hosted tier for the heaviest video and control stacks.

Offering: secure cloud · Billing: per-second
Spin-up: medium · Region: many
visit provider →
globalSaladbatchself-serve

RTX 5090

GPU RAM32 GB (RTX 5090)
A$0.4/hr

Cheapest anywhere, but no persistent volumes; needs external checkpointing. Nodes are strangers' PCs, so least sovereign of all.

Offering: batch priority · Billing: per-second
Spin-up: slow · Region: distributed consumer nodes
visit provider →
globalModalimmediateself-serve

H100

GPU RAM80 GB (H100)
A$5.65/hr

Excellent cold starts (container snapshots); free monthly credits. Best for bursty runs.

Offering: serverless · Billing: per-second
Spin-up: fast · Region: US/EU
visit provider →
globalModalimmediateself-serve

A100 80GB

GPU RAM80 GB (A100)
A$3.6/hr

Scale-to-zero; great for the embeddings/TTS/image side jobs.

Offering: serverless · Billing: per-second
Spin-up: fast · Region: US/EU
visit provider →
globalfal.aiimmediateself-serve

hosted (Wan/Kling/Veo)

GPU RAMhosted (managed)
≈A$0.07 per second of video

Zero-ops hero shots where a hosted model beats self-hosting.

Offering: per-output API · Billing: per-output
Spin-up: fast · Region: US/EU
visit provider →
globalLambdaimmediateself-serve

A100 80GB

GPU RAM80 GB (A100)
A$3.99/hr

Simple VMs, no spot; steady rather than cheap.

Offering: on-demand · Billing: per-minute
Spin-up: medium · Region: US
visit provider →