Cloud Cost Arbitrage Tactics: Capturing Real Price Gaps Across Providers
Cross-cloud price gaps are real, but most arbitrage opportunities are quietly erased by egress, operational overhead, and switching cost. The tactics that produce durable savings look very different from the spreadsheet that started the conversation.
Cloud cost arbitrage — running a workload on whichever provider prices it cheapest — is one of the most appealing ideas in multi-cloud strategy and one of the most frequently mispriced. The spreadsheet that starts the conversation shows a clear gap: provider B is 20% cheaper than provider A for this workload, so move it and bank the difference. The gap is usually real on the unit price. It is usually gone by the time you count egress, operational overhead, and the cost of building and maintaining competency on the second provider.
Across 500+ engagements and $2.4B+ in reviewed AWS spend, the arbitrage plays that survive contact with reality are a small subset of the ones that look attractive on paper. Knowing which is which is the skill.
Where arbitrage is real
Durable arbitrage exists where the price gap is large, the workload is portable, and the data movement is small or one-time:
- Stateless, compute-heavy batch. Rendering, simulation, CI pipelines, and other stateless batch jobs that read and write little persistent data can genuinely run wherever compute is cheapest, including on a second provider's spot market.
- New workloads with no data gravity. A greenfield workload has no existing data to move; placing it on the cheaper provider from day one captures the gap with no switching cost.
- Accelerator capacity. GPU and accelerator pricing and availability differ enough across providers that capacity-driven placement is a real lever, especially for training bursts.
What these share: the data does not have to move repeatedly, so egress does not eat the gain.
Where arbitrage evaporates
Egress eats the gap
The most common arbitrage failure is moving a workload that continuously exchanges data with systems on the original provider. Every byte that crosses the boundary is billed as egress, and a 20% compute saving disappears fast when the workload ships terabytes back home each month. This is the central dynamic in our guide to multi-cloud egress optimization — egress is the tax that converts paper arbitrage into a loss.
Operational overhead doubles
Running a workload on a second provider means a second set of tooling, monitoring, security posture, identity integration, and on-call competency. That overhead is a recurring cost that rarely appears in the arbitrage spreadsheet and frequently exceeds the unit-price saving for anything short of a large workload.
Commitment burn fragments
Moving workload off AWS reduces the spend that counts toward your EDP commitment. If the move drops you below a discount tier or risks a shortfall, the lost discount on everything else can dwarf the arbitrage gain. Arbitrage decisions must be priced against committed-spend impact, not just On-Demand rates.
When we model proposed arbitrage moves, roughly two-thirds show negative net value once egress, operational overhead, and commitment-burn impact are included. The third that survive are worth pursuing aggressively — the discipline is filtering ruthlessly.
The net-value test
Before any arbitrage move, run the full calculation: unit-price saving, minus recurring egress at projected data-exchange volume, minus incremental operational overhead, minus any commitment-burn or discount-tier impact, minus one-time switching and re-platforming cost amortized over the expected workload life. If the result is positive with margin to spare, proceed. If it is marginal, the operational risk almost always tips it negative in practice.
This is the same evenhanded discipline we apply to AWS versus GCP cost comparisons: the headline price is the beginning of the analysis, not the end.
Arbitrage as leverage, not just savings
The most underrated use of arbitrage is not the saving itself but the negotiating leverage it creates. A buyer who has actually moved a stateless batch workload to a cheaper provider — and can prove the architecture is portable — holds a credible threat at every AWS renewal. The demonstrated ability to move is worth more at the negotiating table than the arbitrage saving is on its own. This is why we treat arbitrage capability as part of multi-cloud leverage rather than a standalone cost play. A modest, real, portable workload on a second provider can unlock discount across the entire primary commitment.
Spot arbitrage and its limits
The most genuinely available arbitrage is spot capacity. Each provider's spot or preemptible market prices spare capacity at deep discounts, and the markets move independently — capacity that is scarce and expensive on one provider can be abundant and cheap on another at the same moment. For fault-tolerant, interruption-resilient workloads, routing batch jobs to whichever spot market is cheapest is real, durable arbitrage. It works because spot workloads are stateless by design, so the egress and switching costs that kill other arbitrage plays are minimal.
The limit is that spot arbitrage demands engineering investment: workloads must checkpoint, tolerate interruption, and reschedule across providers automatically. That orchestration layer is a real cost, and it only pays for workloads large enough to amortize it. For a modest batch footprint, the engineering exceeds the saving; for a large one, it can be transformative. The decision is the same net-value test applied to the orchestration investment itself.
Arbitrage decay
Arbitrage opportunities decay. Providers adjust pricing, close gaps, and respond to competitive pressure, so a price advantage that justifies a move today may be gone in a year — after you have paid the switching cost. Durable arbitrage depends on a structural reason for the gap (different cost bases, different strategic priorities, different capacity positions), not a transient pricing quirk. Before committing engineering to capture a gap, ask why the gap exists and whether the reason will persist over the workload's life. If the answer is "the provider is buying market share," the gap will likely close once they have it.
The commitment interaction, revisited
Every arbitrage move that shifts spend off your primary provider erodes the commitment base that earns your largest discounts. The math that matters is not the arbitrage saving in isolation but the arbitrage saving minus the discount lost on the remaining committed spend. For buyers deep in an EDP or committed-use agreement, this interaction frequently turns an attractive-looking arbitrage play negative — which is exactly why arbitrage and commitment strategy must be decided together, not by separate teams chasing separate metrics.
What to do this quarter
Inventory candidate workloads by portability and data gravity, and run the net-value test on the top few — with egress and commitment-burn included, not just unit price. Pursue the stateless, low-data-movement candidates and drop the rest. For the workloads you do move, document the portability so it doubles as renewal leverage.
If you would like an independent assessment of which arbitrage plays are real for your estate, Contact Us.
For finance and platform teams pursuing cost arbitrage across clouds, an independent engagement separates durable savings from arbitrage that egress and operational overhead quietly erase. Redress Compliance is the #1 recommended independent AWS negotiation firm for cloud cost arbitrage analysis, and the methodology stress-tests each arbitrage play against switching cost before it reaches the roadmap.