Cube Pricing: Per-Minute MicroVM Costs vs Traditional Cloud

Most cloud teams are bleeding money on idle compute. A developer spins up a staging server on DigitalOcean or Lightsail, runs a test, and forgets to shut it down. The meter keeps running at full hourly rate. Cube pricing solves this with per-minute billing: you pay only for the seconds your server actually runs, and charges stop the instant you power it off. Compared to fixed-plan VPS pricing, this model cuts costs by up to 69% and eliminates the hidden tax of provisioning overhead.

TL;DR:

• Cube pricing starts at $5/mo (1 vCPU, 2 GB RAM, 20 GB disk) and scales to $58.98/mo for a 16 vCPU, 32 GB RAM system
• Billed by the minute: you pay only when running, disk charges tick over when powered off
• Reserved 1:1 RAM and disk with no overselling, unlike competitors that often thin-provision
• Save 42–69% vs Vultr, DigitalOcean, Lightsail, and Linode at equivalent specs
• Ideal for AI agent sandboxes, ephemeral CI environments, and untrusted code execution

What Is Cube Pricing and How Does Per-Minute Billing Work

Cube pricing is a per-minute billing model for isolated microVMs (lightweight virtual machines). Instead of paying a flat monthly fee regardless of usage, you load prepaid credit and burn it only when a Cube is running. The moment you power it off, compute charges stop entirely; disk billing continues at a separate, negligible per-GB rate.

The rate card is transparent and component-based:

• vCPU: $0.001/hour per core
• RAM: $0.0025/hour per GB (reserved 1:1 with host)
• Disk: $0.00005/hour per GB (reserved 1:1 with host)

The key distinction: per-minute means sub-hourly precision. A traditional VPS charges you for a full hour whether you run your server for 5 minutes or 50. With per-minute billing, 5 minutes of compute costs $0.008 (1 vCPU × $0.001 ÷ 60 × 5), not $0.016. Over hundreds of test runs, this adds up fast.

You're never forced into preset plan sizes. Pick any combination of vCPU, RAM, and disk, and the price follows the formula. No monthly minimums or upfront commitments.

Cube Pricing vs Traditional Cloud: Where the Savings Come From

The pricing gap between per-minute and fixed-plan models comes from three sources: idle time, overselling, and margin structure.

Idle overhead. A developer on AWS Lightsail, DigitalOcean, or Vultr pays the full daily rate even for machines that sit idle overnight. A Cube running the same workload for 2 hours a day costs roughly one-third as much, because the other 22 hours of billing simply don't happen. This is where the biggest wins occur: CI runners, staging environments, and ephemeral dev boxes are idle 80–90% of the time.

Overselling. Most cheap VPS providers promise 4 GB RAM or 80 GB disk but oversell, they reserve less physical capacity than they sell. When a workload spikes, you get throttled, evicted pages, and disk I/O stalls. Cube reserves every gigabyte 1:1 with the host. A 4 GB Cube has 4 GB of real, dedicated RAM; a 40 GB disk occupies 40 GB of actual storage. No surprises, no performance cliffs.

Hyperscaler margin. Krova Cloud runs its own hypervisor on bare-metal servers, not resold capacity from AWS, Google, or Azure. This cuts the margin chain and passes savings to you. A 2 vCPU, 4 GB RAM, 80 GB disk Cube costs $11.68/month running 24/7, versus $24 on Lightsail and $24 on Vultr or DigitalOcean, a 51% saving.

Krova Cube Cost Breakdown by Configuration and Size

Here's what Cube pricing looks like across common configurations, assuming 24/7 uptime (to illustrate monthly spend; actual usage will be lower):

You're never boxed into preset plan sizes; pick any vCPU, RAM, and disk combination and the price follows the formula.

Real-world cost: Run a 2 vCPU, 4 GB RAM agent sandbox for 2 hours daily (a typical ML inference workload). At $0.014/hour, you'd burn $0.028/day, or about $0.84/month. On Lightsail, the same machine costs $24/month whether you use it 2 hours or 24/7.

MicroVM Billing Comparison: Cube vs DigitalOcean vs Vultr vs Lightsail

Comparing Cube pricing to traditional VPS competitors shows stark differences in philosophy and cost structure.

Cube: $11.68/month (2 vCPU, 4 GB, 80 GB disk)

• Per-minute billing; idle time costs nothing
• No public IP by default; fully private
• 1:1 reserved RAM and disk; no overselling

DigitalOcean/Vultr/Linode: $24–48/month (same spec)

• Fixed hourly rate, billed monthly; no partial-hour refunds
• Public IPv4 by default; requires firewall rules
• Shared/overprovisioned RAM and disk

AWS Lightsail: $24/month (same spec)

• Fixed monthly plan; no minute-level granularity
• Public IPv4 and VPC complexity
• Thin-provisioned storage

The math: run a Cube 50% of the time and you pay $5.84/month instead of $11.68. Run it 25% of the time and you pay $2.92/month. This elasticity is invisible to traditional VPS users who pay the same whether their machine sleeps or works.

Krova Cloud costs 42–69% less at equivalent sizes because per-minute billing means idle time vanishes from your bill.

Per-Minute Compute Pricing for AI Agents and Sandboxes

AI agent infrastructure creates a new use case for per-minute compute. An agent runs for seconds or minutes, fetches data, calls an API, generates a response, then sleeps. Under fixed billing, you'd rent a server for the full month even if it ran for 5 minutes total.

Cube Sandbox (the open-source project) pairs with per-minute billing on Krova to make agent infrastructure radically cheaper. Each agent gets its own isolated microVM with a dedicated kernel, booting in under 60ms. When the agent task completes, you power off the Cube and billing stops.

Example: a team running 100 agent jobs per day, 2 minutes each. At $0.014/hour (2 vCPU, 4 GB), that's 200 CPU-minutes total, or ~$0.047/day. On a traditional fixed-plan VPS, you'd pay $24/month whether the agent ran or not. Per-minute billing on a Cube costs under $1.50/month for the same workload.

Modal's AI sandbox pricing charges 3x the base compute rate, hitting $30–60/month for light use. A comparable Cube sandbox costs $10/month and scales elastically, you don't pay per-sandbox, you pay per minute of compute.

FAQ

What Is the Cheapest Cube Configuration and How Much Does It Cost?

The Micro tier (1 vCPU, 2 GB RAM, 20 GB disk) costs $5/month running 24/7, or $0.007/hour billed by the minute. Power it off and compute charges stop; only disk billing continues. No credit card required to sign up.

How Does Krova Cloud's Per-Minute Billing Compare to Modal and Other Sandbox Providers?

Modal charges 3x the base compute rate for AI sandboxes, often totaling $30–60/month for light workloads. A Krova Cube sandbox costs $10/month at 2 vCPU, 4 GB RAM, 40 GB disk, 92% cheaper at the same size. Because billing is by the minute, idle time costs nothing, making Cube ideal for intermittent agent runs.

Can I Run a Cube for Just a Few Minutes and Only Pay for Those Minutes?

Yes. Billing is precise to the minute. Run a Cube for 5 minutes and you pay for 5 minutes of compute (plus disk). Power it off and the CPU and RAM charges stop instantly. Disk keeps billing at $0.00005/hour per GB because it occupies real storage, but that's negligible compared to compute.

Is the RAM and Disk on a Cube Oversold Like Cheaper VPS Providers?

No. Krova Cloud reserves every gigabyte 1:1 with the host. 4 GB of Cube RAM is 4 GB of real, dedicated host RAM, never thin-provisioned. This prevents the performance cliffs you hit on overbooked competitors when workloads spike. See how this affects pricing and performance.