Cloud Egress Cost: The Hidden Line Item Eating Australian Cloud Bills
Cloud egress cost is the line item most Australian CIOs underestimate when they design their cloud architecture and most regret when the bill comes in. In 2026, with AI workloads pulling more data across providers and regions, egress charges are showing up as a meaningful percentage of total cloud spend for some Australian enterprises.
What egress actually costs
Egress in 2026 still costs around 8 to 12 cents per GB for cross-region traffic on the major hyperscalers, depending on volume tier and destination. Most Australian enterprises operating in Sydney or Melbourne regions are paying somewhere in that range. Cross-cloud egress is more expensive. Egress to on-prem can be similar or worse depending on the configuration.
The number that surprises CIOs is how much data their AI workloads move. Training data, vector index sync, agentic systems pulling fresh context, RAG pipelines hitting external data sources. A modest AI workload that nobody flagged as data-intensive can run up tens of thousands of dollars a month in egress.
The patterns making it worse
Three architectural patterns drive disproportionate egress cost. First, cross-region disaster recovery designs that synchronise data more aggressively than the RPO requires. Second, multi-cloud setups where the data plane crosses providers — typically because someone wanted a particular AI service on a different cloud than the data warehouse lives on. Third, lift-and-shift cloud migrations that did not rearchitect the network — every cross-VPC call costs egress.
Egress is also one of the few cloud line items where pricing has not meaningfully dropped in the last five years. The hyperscalers have been quiet about this because egress lock-in is valuable to them. Regulators in Europe have started to push back. In Australia we have not yet seen comparable pressure.
What is working
Australian CIOs who have got their egress cost under control have done some combination of the following. Co-located the data and the workload — moving the AI service to the cloud where the data lives, not the other way around. Implemented private network connectivity where it makes sense — Direct Connect, ExpressRoute, or equivalent, which has its own cost but trades a fixed cost for a volume cost. Reviewed cross-region replication policies and tightened them to the actual business requirement. Used CDN aggressively for outbound user traffic.
The CIOs not paying attention to this in 2026 will see it on a quarterly cost review and not know where the spike came from. Egress cost is a FinOps discipline now, not an architecture concern only.
A practical step
Run a thirty-day egress audit. Pull the egress charges per service, per region, per workload. The number that comes out of that exercise usually contains a surprise. Address the top three line items first. That is where most of the dollars are.