When should a startup switch cloud providers?
A startup should switch cloud providers when a specific, measurable problem costs more than the move itself. The usual triggers are startup credits running out on a bill the business cannot carry, egress or compute prices that dominate costs, GPUs or regions the current provider cannot supply, a compliance requirement it cannot meet, and support that fails during incidents. Switching to collect another provider’s credits rarely pays, because the migration takes months and the new credits expire too.
Triggers that justify a move
| Trigger | What to measure | What to check before moving |
|---|---|---|
| Credits run out | Your monthly bill at list prices in the month after the credits expire | Rightsizing, spot capacity and cheaper instance types can cut the same bill without a move |
| Egress fees dominate | Outbound traffic as a share of the bill | Whether a CDN or caching removes most of it |
| Missing GPUs | Weeks waiting for quota or capacity on the GPU type you need | Whether renting GPUs elsewhere for that one workload is enough |
| Missing region | Latency for users or a residency requirement that no region meets | Whether one workload can run on a second provider |
| Compliance | A customer contract or regulation that requires a certification or jurisdiction your provider lacks | Whether the provider offers a compliant variant, such as a sovereign region |
| Support and reliability | Response times during incidents, repeated outages, account or capacity restrictions | Whether a higher support plan fixes it |
Credits running out
Startup programs make the first one or two years cheap. AWS Activate offers up to $200,000 in credits for startups backed by an approved investor before Series B, and up to $5,000 for self-funded startups. Google’s startup program gives up to $200,000 over two years, or up to $350,000 for AI startups. When the credits end, the first full-price bill is often the first time a team sees what its architecture costs. Optimize first: a bill that is too high on one hyperscaler is often too high on the next one for the same reasons.
Egress and compute prices
A product that serves video, downloads, backups or AI model weights can spend more on outbound traffic than on compute, because hyperscalers bill egress per GB while many smaller providers include terabytes per server. What cloud egress fees are compares the pricing models. Steady compute-heavy workloads show a similar gap, illustrated in how Hetzner Cloud compares to AWS.
GPUs and regions
Startups training or serving models often cannot get the GPUs they need on demand from their current provider. GPU-focused providers sell that capacity directly, as explained in what a neocloud is. This trigger rarely requires moving everything: the GPU workload can move while the rest of the product stays.
Compliance
Selling to European enterprises or public-sector customers can bring requirements such as BSI C5, NIS2 obligations or a provider outside US jurisdiction. If the current provider cannot meet them, see what a sovereign cloud is for how to evaluate alternatives.
Reasons that rarely justify a move
- New credits elsewhere. A migration costs months of engineering focus, and the second set of credits also expires.
- A single outage. Every provider has outages. Design for them with multiple zones and backups outside the provider before changing providers.
- A small price difference. If the saving is a few percent of the bill, commitments, rightsizing and architecture changes usually deliver more.
What switching costs
- Running two environments. You pay both providers while workloads move and while you test the new setup.
- Moving data out. Outbound transfer is billed as egress. AWS and Google Cloud waive it for customers leaving entirely, and in the EU the Data Act bans switching charges from 12 January 2027. What cloud egress fees are covers the waivers and what the EU Data Act changes for switching cloud providers covers the legal process.
- Commitments. AWS Savings Plans run for one or three years and can be returned only within 7 days of purchase, in the same calendar month, and for hourly commitments of $100 or less. Google Cloud committed use discounts cannot be cancelled, and you pay for the full term regardless of usage.
- Unused credits. Credits are tied to the provider that issued them and are lost when you leave.
- Rewrites. Applications built on proprietary databases, queues, serverless functions or identity services need equivalents on the new provider.
- Compliance work. Audits, data processing agreements and customer security questionnaires that name your provider have to be updated.
- Cutover risk. Moving databases and DNS carries a risk of downtime or data loss if it is not rehearsed.
Reduce lock-in before you move
The cheapest migration is one prepared before it is needed. Run workloads in containers on conformant Kubernetes, use open-source data services such as PostgreSQL, describe infrastructure as code, use the S3 API for object storage and keep backups outside the provider. How to avoid cloud vendor lock-in explains each control.
A partial move is often the right first step: put the workload that triggered the decision, such as GPU training or a traffic-heavy service, on the provider that fits it and keep the rest where it is. Workload placement covers how to decide what runs where. For a full migration plan, from inventory to cutover, see how to migrate from a US hyperscaler to a European cloud.
Switching providers with Cloudfleet
Cloudfleet Kubernetes Engine (CFKE) runs one cluster with nodes in your own AWS and Google Cloud accounts, on Hetzner Cloud and on your own servers, so you can keep using existing credits and discounts while moving workloads. Moving a workload between providers changes its scheduling constraints, not its manifests. See how Fleets work.