Make clouds compete
for your workload.
Run compute-heavy Python workloads across multiple clouds through one endpoint. Verlex automatically finds qualified compute based on price, availability, and reliability.
One line of Python. Every cloud's best option.
import verlex result = verlex.cloud(train, data)
Used by engineers who work at
Logos identify the employers of individual Verlex users. They do not imply sponsorship or endorsement by these organizations.
Cheap compute is everywhere.
Using it is the hard part.
Every provider needs its own account, quota request, deploy script, and driver setup. Spot instances need babysitting. Cloud bills need auditing.
Most developers just pick one cloud and overpay for it, rather than navigate the complexity of using several.
Three steps. No infrastructure.
Route
Verlex evaluates your code for the most qualified compute across 14 different cloud providers, priced at live rates before committing to a single job.
Run
The workload executes on the best available option. Code runs unchanged. Verlex meters per second from your prepaid balance.
Recover
If capacity disappears or execution fails, Verlex retries, fails over to the next qualified option, or recovers from checkpoint, automatically.
Let the infrastructure compete.
Before every job runs, Verlex queries all connected providers simultaneously. The cheapest qualified option wins. You see the price before committing.
You don't compare these. Verlex does.
→ routed to Verda · $2.43/hr · H100 x1
Get the optimal model for your code.
Verlex analyzes your code and selects the best model for what it actually needs. Your requirements are met, without overspending on compute you don't need.
The code runs at peak efficiency, on hardware that fits it. Not the biggest machine. The right one.
Use what you already have first.
Before sending a job to the cloud, Verlex checks your available local resources and uses them whenever they can handle the workload. Only when more compute is needed does Verlex route to the best cloud option.
Local means every machine you already pay for: your own backend cloud counts as local hardware too. Make the most out of the hardware you already own, and only pay for cloud compute when you actually need it.
Not another compute marketplace.
Verlex runs the workload. Not just finds you a machine.
One execution layer. Many workloads.
Verlex is not specific to GPUs or AI. Any compute-heavy Python workload can route through one endpoint.
Model training & fine-tuning
Scale training runs across the cheapest available GPUs without managing infrastructure.
LLM inference
Serve inference workloads on cost-effective spot capacity with automatic failover.
Image & video generation
Burst generation workloads to cloud GPUs and pay only for actual compute.
Batch compute
Run large batch jobs across clusters with per-second billing and no idle cost.
Local → cloud overflow
Develop locally. When resources hit threshold, heavy functions move to the cloud automatically.
Lower infrastructure cost without infrastructure hunting.
Spot pricing exists at every major provider. Verlex makes it accessible without the operational burden of managing preemption, checkpoints, and retries yourself.
Built for developers, not cloud consoles.
Built for workloads that matter.
Verlex routes production workloads across infrastructure you'd trust individually. We hold provider quotas so your jobs don't wait.
Pay for compute, not management.
Provider compute at cost. Verlex adds a flat fee. No percentage markup.
- Cost-first routing across 10 clouds
- Automatic failover in price order
- Spot routing with checkpoint recovery
- Serverless lane for small jobs
- 25 GB persistent storage
- Pay in USD, BRL, INR, THB, or CAD
- Everything in Standard
- All 14 clouds unlocked
- Fast mode: fastest qualified hardware
- Clusters: up to 8 GPUs, one line
- Warm capacity prioritized
- 300 GB persistent storage
- Cancel anytime, credits never expire
- Everything in Performance
- 1 TB+ persistent storage
- Team accounts with owner/member roles
- Per-member API keys and spend tracking
- Audit log, GDPR deletion on demand
- Same published fee table, no markup
Common questions.
How does Verlex make money?
The provider's compute price passes through at cost. Verlex adds a flat service fee of $0.02 to $0.75 per GPU-hour depending on hardware tier. No percentage markup, no hidden margin.
How much can I save vs AWS or GCP?
Every job is routed to the cheapest qualified GPU across 14 clouds. Spot discounts typically run 60-80% off list. H100 example: $3.50/hr list → $1.05/hr spot + $0.40 Verlex fee = $1.45/hr all-in, roughly 59% off.
What happens if a spot GPU is reclaimed?
Checkpoints sync continuously to object storage. If a provider reclaims capacity mid-job, Verlex restarts from the last checkpoint on the next best available option. Reclaimed time is not billed.
Can I get a surprise bill?
No. Jobs are prepaid and meter per second. When your balance runs out, the job stops. Auto top-up is opt-in only. Every job has a hard 7-day backstop.
Do I need to change my code?
No. result = verlex.cloud(train, data) runs your existing Python function on cloud infrastructure. Two lines including the import.
Which clouds does Verlex route across?
AWS, GCP, Azure, Verda, RunPod, Vast.ai, Hyperstack, JarvisLabs, TensorDock, Lyceum, Beam, Northflank, Novita, and Cerebrium: 14 clouds priced live per job.
Is my code kept confidential?
Confidentiality is our top priority. Your code is never looked at by a human, and it is deleted immediately after execution: every instance is destroyed or fully wiped before any reuse. We never read or train on your code.
More questions? Write to us at contact@verlex.dev.
Your code shouldn't care
which cloud runs it.
One endpoint. Fourteen clouds. Automatic routing. Free to start.