How do VESSL AI credits work (1 credit = $1) and how do I buy more credits?

VESSL AI credits are the simplest unit for paying for GPUs and storage on VESSL Cloud: 1 credit = $1 USD. You load credits into your workspace, then VESSL deducts usage in real time as your jobs run. No custom math, no hidden multipliers.

This guide walks through how credits work, how they’re consumed, and how to buy more when your balance runs low.

How VESSL AI credits work

1 credit = $1: simple, dollar‑linked balance

Billing unit:
- 1 VESSL AI credit = $1 USD.
- Published hourly GPU prices (e.g., for A100, H100, H200, B200, GB200, B300) map directly to credits.
Shared at the workspace level:
- Credits are tied to your VESSL workspace, not to an individual user.
- Anyone running jobs in that workspace draws from the same credit pool.
Real-time deduction:
- As jobs run, VESSL calculates per-second usage from hourly rates and deducts credits continuously.
- When a job ends, total credits consumed = sum of all resources used over time.

You can think of your balance as a prepaid wallet: load $X into credits, run workloads, and watch the remaining balance drop as compute and storage are consumed.

What consumes VESSL AI credits?

Credits are spent on any metered resource your workloads touch.

1. GPU compute

Credits are primarily consumed by GPU time, based on:

GPU model: A100, H100, H200, B200, GB200, B300, etc.
Reliability tier:
- Spot: Lowest cost, can be preempted.
- On-Demand: Reliable capacity with automatic failover across providers.
- Reserved: Guaranteed capacity with deeper discounts for commitments.
Runtime: How long your job or service runs.
GPU count: Number of GPUs attached to each workload.

Example (illustrative, not actual pricing):

H100 On-Demand: $4.00/hour per GPU
You run an 8x H100 training job for 2 hours.
Cost = 8 GPUs × 2 hours × $4.00 = $64.00, so 64 credits are deducted.

2. CPU, RAM, and other cluster resources

Depending on the configuration, you may also spend credits on:

CPU vCores and system RAM attached to your GPU nodes
Any non-GPU nodes or auxiliary compute used by your workloads

These rates are set in the same dollar-denominated model, so again, 1 credit = $1.

3. Storage and data

If you use VESSL’s storage primitives, credits are consumed based on:

Cluster Storage:
- High-performance, shared file system used across jobs.
- Billed on provisioned capacity (e.g., per GB per month) and sometimes I/O.
Object Storage:
- Lower-cost storage for datasets, model artifacts, and logs.
- Billed per GB stored, plus any applicable egress/operation fees.

Storage usage is translated directly into credits over time (e.g., daily or monthly proration).

How to track and manage your credit balance

You don’t want a long training run to stop because you misjudged spend. Use the console and monitoring tools to stay ahead of your balance.

Check your remaining credits

In the VESSL Web Console:

Go to your workspace.
Open the Billing or Credits section (naming may vary slightly).
View:
- Current credit balance
- Recent usage history
- Any upcoming reservations or committed spend

This gives you a quick sense of how many GPU hours you can still run at your current usage pattern.

Monitor job-level spending

For each job or service:

See the GPU type, count, and tier (Spot/On-Demand/Reserved).
Estimate cost using the published hourly price per GPU.
Use real-time monitoring to see how long a run has been active and adjust:
- Stop runs early if they’re not converging.
- Scale down GPU count if you’re overshooting your budget.
- Move from On-Demand to Spot for non-critical experiments.

This reduces “job wrangling” around cost management and lets you run more “fire-and-forget” experiments without surprise bills.

How to buy more VESSL AI credits

When your credit balance drops, you have two ways to top up: self-serve in the console or talk to sales for larger or reserved agreements.

Option 1: Buy credits in the Web Console (self-serve)

Best when you want to start fast or add more capacity for upcoming experiments.

Typical flow:

Sign in to VESSL Cloud.
Select the workspace where you want to add credits.
Navigate to Billing → Add Credits or similar.
Choose your top-up amount (e.g., $100, $1,000, $10,000 in credits).
Enter payment details (commonly credit card or available local methods).
Confirm the purchase.

Once payment is processed:

Credits are added to your workspace balance.
You can immediately launch more jobs with A100/H100/H200/B200/GB200/B300 GPUs on Spot, On-Demand, or Reserved as available.

Option 2: Contact sales for Reserved capacity and larger purchases

Use this path if:

You’re running mission-critical workloads (production LLM inference, 24/7 services).
You need guaranteed capacity on specific GPU SKUs.
You want discounts via committed spend or longer-term contracts.
You need procurement-ready documents: SOC 2 Type II, ISO 27001, SLAs, or custom terms.

How it usually works:

Go to vessl.ai and click Get Started or Talk to Sales.
Share:
- Target GPUs (e.g., H100, B200, GB200)
- Expected GPU count and duration
- Workload types (LLM post-training, Physical AI, AI for Science, etc.)
VESSL proposes:
- A Reserved plan with capacity guarantees.
- Potential discounts for committed or academic usage.
- Any integration/onboarding support you need.

Approved contracts typically translate into:

A committed credit allocation.
Possibly reserved-rate pricing per GPU hour, reflected in your credits and usage.

What happens if I run out of credits?

When your credit balance approaches zero, VESSL protects you from uncontrolled spend.

Typical behavior (exact behavior may depend on your workspace settings):

New jobs may be blocked from starting once you hit a threshold.
Running jobs may:
- Continue until balance is exhausted, then be stopped, or
- Be prevented from starting if the system predicts they would exceed the remaining balance.

Best practice:

Set up internal alerts based on credit thresholds (e.g., at 70%, 40%, 10% remaining).
Top up credits before starting long LLM training runs or large-scale experiments.
For production services, consider Reserved or On-Demand with failover plus a steady credit buffer to avoid interruptions.

How credits interact with Spot, On-Demand, and Reserved

Your credit-consumption strategy should match how critical each workload is.

Spot: Experiment hard, spend less

Best for:
- Early-stage training runs, hyperparameter sweeps, and non-critical batch jobs.
Behavior:
- Cheaper hourly rates → slower credit burn.
- Can be preempted, so expect occasional interruptions.
Credit impact:
- Great for maximizing experiments per dollar on your credit balance.

On-Demand: Production runs with failover

Best for:
- Production-adjacent runs, evaluation pipelines, and services that must stay up but can tolerate cloud-provider shifting.
Behavior:
- Automatic failover across providers and regions when capacity is disrupted.
Credit impact:
- Higher rate than Spot, but your runs survive provider outages and region failures.

Reserved: Mission-critical with guarantees

Best for:
- Long-running LLM post-training, production inference with strict SLOs, and government/enterprise workloads.
Behavior:
- Capacity is reserved for you.
- Often paired with SLAs, onboarding help, and dedicated support.
Credit impact:
- You typically commit to a certain spend; in exchange, you get discounted rates and guaranteed GPU access.

FAQs about VESSL AI credits

Do credits ever expire?

This can depend on your plan or contract:

Self-serve credits may have a different expiration policy than enterprise/Reserved contracts.
For exact terms, check your workspace billing page or your agreement with VESSL AI.

If you’re unsure, contact support or your VESSL account manager before making a large top-up.

Can I move credits between workspaces?

In many setups, credits are allocated per workspace:

Moving credits between workspaces may require admin or support involvement.
For organizations with multiple teams, it’s common to:
- Keep a central billing workspace, or
- Allocate credits per team and consolidate billing via invoices.

Check with VESSL support if you need cross-workspace credit management.

Can I see detailed usage by job or user?

Yes. Use:

Web Console: Job/service detail views plus billing summaries.
CLI (vessl run): Integrate usage tracking into your own reporting or dashboards.

This is how teams justify spend across research groups, products, or customers while keeping a clean audit trail.

Summary: Use credits like a GPU control budget

1 credit = $1. No conversion games.
Credits are workspace-level, spent on GPU/CPU, storage, and related infrastructure.
You can:
- Buy credits instantly in the Web Console for on-the-fly experiments.
- Work with sales for Reserved capacity, discounts, and SLAs.
Match your credit strategy to workload criticality:
- Spot for cheap experiments.
- On-Demand for reliable runs with failover.
- Reserved for guaranteed capacity and predictable spend.

When you’re ready to add more capacity or secure Reserved GPUs for upcoming LLM post-training or Physical AI projects, you can Get Started and align your credit plan with how your teams actually run workloads.