Cloud provider with the cheapest interruptible compute for batch jobs

For general CPU batch jobs, I’d start with Google Cloud Spot VMs: Google publishes current Spot rates, and some machine types are very inexpensive—for example, its listed c4a-highcpu-4 rate is $0.07944/hour. Spot VMs can be reclaimed at any time, so use checkpointing or make jobs easy to rerun. (cloud.google.com)

There isn’t one universal cheapest provider: the best price depends on the CPU/RAM shape, region, and current capacity. AWS EC2 Spot is also worth comparing, especially if you want a managed batch workflow through AWS Batch; AWS advertises savings of up to 90%, as does Google, but those are maximum discounts—not a guarantee that either is cheapest for your workload. (aws.amazon.com)

If you share whether you mean CPU or GPU, plus roughly how many cores and how much RAM you need, I can narrow it down.

There's no single answer. The cheapest option depends on instance type, region, and the day's capacity, and the sources disagree on the winner.

Google Cloud (Spot VMs) is the most common pick for lowest hourly rate. Spot VMs offer a 60–91 percent discount off on-demand prices. Google's pricing is fixed rather than auction-based: the price is set by Google and the discount is applied automatically. The trade-off is availability, since you either get a Spot VM at the discounted rate, or the request fails because no capacity is available. One comparison of EU batch pricing also found that GCP preemptible pricing often wins the spot bracket.

Azure sometimes comes out cheapest. In one 2026 comparison of m7i/n2/D4s-class instances, Azure spot is cheapest at ~$0.038/hr. That same source warns that Azure spot VMs have a higher eviction rate than AWS spot instances.

AWS is usually not the price leader, but it has the most mature tooling. AWS Spot has the most mature interruption handling and fleet management (Spot Fleet, capacity-optimized allocation). For GPUs, one source says AWS P5 spot exists on paper but is rarely available for H100 at any discount.

Oracle and smaller clouds are worth checking. Oracle Cloud offers preemptible instances at a fixed 50% discount from on-demand pricing. Hetzner, DigitalOcean, Linode, and Vultr have no spot tier, but a comparison found that their on-demand rates ($0.007-0.071/hr) are often cheaper than Big 3 spot prices. These are non-interruptible, so they aren't a direct match for your spot use case, but they can beat spot on cost.

Practical considerations for batch jobs:

• Interruption notice varies. Google's documentation says the preemption notice starts a best-effort shutdown period of up to 30 seconds. AWS gives 2 minutes in one guide, and Azure gives 30 seconds in another.
• Runtime limits differ. Preemptible VMs can only run for up to 24 hours at a time, but Spot VMs don't have a maximum runtime unless you limit the runtime.
• Checkpointing matters more than the provider. One guide recommends checkpointing every 15 to 30 minutes for multi-hour jobs.

Recommendation: Google Cloud Spot VMs are the best default for lowest price with simple, predictable pricing. Run your actual instance type and region through each provider's pricing calculator, and check spot availability. Azure or Oracle may win for your specific workload. Most of the sources above are secondary blog posts with differing numbers, so verify against live pricing before committing.

For batch jobs that can handle interruptions, the "cheapest" provider depends heavily on whether you need CPUs or GPUs.

In 2026, the market for interruptible compute is split between hyperscalers (for standard CPU batch jobs) and specialized GPU cloud providers (for AI/ML and intensive compute).

1. For GPU Batch Jobs (AI/ML & Intensive Compute)

If your batch jobs require GPUs, hyperscalers like AWS, GCP, and Azure are often significantly more expensive than "neocloud" or marketplace providers.

• Vast.ai: Frequently cited as one of the cheapest options for GPU compute. It is a peer-to-peer marketplace where you can find interruptible (spot) instances at prices 70–80% lower than standard cloud retail rates.
• io.net: Another highly competitive provider for GPU clusters, often significantly cheaper (60%+ savings) than AWS or Azure for large-scale GPU needs.
• RunPod / Spheron: Both offer excellent, developer-friendly interfaces with "Community" or "Spot" capacity that is highly cost-effective for batch inference or training jobs that can be checkpointed.
• Key Trade-off: With these marketplaces, you are trading some predictability and support for significantly lower hourly rates. Ensure your code is robust enough to handle instances being reclaimed by the host.

2. For CPU Batch Jobs

If your workload is CPU-based (e.g., data processing, rendering, CI/CD), the major hyperscalers offer "Spot" or "Preemptible" instances that are highly reliable and integrated with standard tooling.

• Google Cloud (GCP) Spot VMs: Often recommended for batch processing due to their simple pricing and the ability to easily integrate with Google Kubernetes Engine (GKE) by simply adding a --provisioning-model=SPOT flag. They offer savings of up to 91%.
• AWS Spot Instances: The industry benchmark. They offer discounts of up to 90% and have no runtime limits (unlike some legacy preemptible models). They are ideal if you are already within the AWS ecosystem.
• Oracle Cloud Infrastructure (OCI): Generally aggressive on pricing for raw compute and often cheaper than AWS/Azure for standard VM instances. If you are running massive, long-running CPU-based batch jobs, OCI's base pricing combined with their generous egress allowances can make them a strong contender.

Summary Comparison Table

Critical Advice for Batch Jobs

• Don't ignore Egress: A "cheap" provider can become expensive if you move large datasets out of their network. Providers like CoreWeave or Lambda Labs often have policies (like no egress fees) that can drastically lower your total cost of ownership.
• Checkpointing is Mandatory: Because interruptible instances can be reclaimed without much notice (often 30 seconds to 2 minutes), your batch jobs must be stateless or support frequent checkpointing to a persistent store (like S3/GCS) to avoid losing work.
• Billing Granularity: Look for providers that offer per-second or per-minute billing. If your jobs are short, rounding up to the nearest hour (common in some older setups) can wipe out any savings you gained from the lower spot price.

Google Cloud Spot VMs are a strong default for low-cost, interruptible batch compute, but there isn’t a universal cheapest provider: prices vary by region, VM type, and capacity.

Google advertises Spot discounts of up to 91% off regular VM prices, and its pricing page lists a c4-standard-2 (2 vCPUs, 7 GiB) at $0.057979/hour in the shown pricing. Spot VMs can be reclaimed, so design jobs to checkpoint and resume. ¹

AWS EC2 Spot is also worth comparing: it advertises discounts of up to 90%, with prices varying by instance type and region. ² Those headline discounts aren’t directly comparable; check the exact machine, region, storage, and data-transfer costs you need.

Practical recommendation: start with Google Cloud Spot for a managed batch setup, then benchmark the equivalent AWS Spot instances in your target region before committing. Google’s Batch service has no additional service charge, according to Google. ¹⁵