io.net requires zero contracts and zero commitments - all GPU instances are 100% on-demand with pay-per-second billing. You can provision GPUs for as little as 1 minute or run them continuously for months, then cancel instantly by stopping instances. Unlike AWS, Azure, or GCP that push 1-3 year reserved instance contracts or Lambda Labs' long-term commitments, io.net gives you complete flexibility to start and stop anytime. For enterprise customers seeking volume discounts, optional 3-12 month reserved capacity plans offer 10-20% additional savings while maintaining monthly billing with no upfront payments and no early termination penalties.

True On-Demand: No Contracts, No Commitments

What "no contracts" actually means:

Zero Minimum Usage:
- Provision a GPU for 5 minutes to test something → Pay $0.015 (for RTX 4090)
- Run inference for 2 hours → Pay $0.36
- Train a model for 3 days → Pay $129.60
- You decide when to start and stop. No minimums.

No Lock-In:
- Cancel at any second by running io stop <instance-id>
- No notice period required
- No cancellation fees
- No questions asked

No Subscription:
- No monthly fees
- No annual plans required
- No "pay for 12 months to get discount" schemes
- Pay only for actual GPU seconds used

No Procurement Complexity:
- No vendor negotiations
- No multi-page contracts
- No legal review needed
- Sign up and deploy in 5 minutes

How This Compares to Competitors

Provider	Minimum Commitment	Cancellation Policy	Contract Complexity
io.net	None (on-demand)	Instant (stop instance anytime)	No contract
AWS	None for on-demand, 1-3 years for reserved	Instant for on-demand, no refunds for reserved	Terms of Service only
Azure	None for pay-as-you-go, 1-3 years for reserved	Instant for PAYG, no refunds for reserved	Terms of Service only
GCP	None for on-demand, 1-3 years for committed use	Instant for on-demand, no refunds for committed	Terms of Service only
Lambda Labs	None officially, but frequent sellouts force planning	Instant	Terms of Service only
CoreWeave	None for on-demand, custom enterprise contracts	Varies by contract	Complex for enterprise
Vast.ai	None (marketplace model)	Instant	Terms of Service only

Key Difference:
While all platforms technically offer on-demand options, io.net's base pricing is 50-70% cheaper than others' on-demand rates, eliminating pressure to commit to long contracts for savings. You get spot/reserved-like pricing without any commitment.

Optional Reserved Capacity (For Additional Savings)

If you want 10-20% extra discount and guaranteed availability:

3-Month Reserved Capacity:
- Discount: 10% off on-demand pricing
- Minimum spend: $5,000/month
- Commitment: 3 months (can cancel after with 30-day notice)
- Early termination: Pay 50% of remaining commitment if canceled early

6-Month Reserved Capacity:
- Discount: 15% off on-demand pricing
- Minimum spend: $10,000/month
- Commitment: 6 months
- Early termination: Pay 50% of remaining commitment

12-Month Reserved Capacity:
- Discount: 20% off on-demand pricing
- Minimum spend: $25,000/month
- Commitment: 12 months
- Early termination: Pay 50% of remaining commitment

Still No Upfront Payment:
Unlike AWS reserved instances requiring 50-100% upfront, io.net reserved capacity bills monthly. You're committing to usage, not prepaying.

Example:
- 12-month commitment: $25,000/month × 12 = $300,000 total value
- Billing: $25,000 per month (not $300,000 upfront)
- Early cancellation after 6 months: Pay 50% × $150,000 remaining = $75,000 penalty
- Total cost if canceled at 6 months: $150,000 paid + $75,000 penalty = $225,000 (vs. $300,000 full contract)

Fair middle ground: Discounts for commitment, but escape clause if needs change.

Why No Contracts Work for Most Teams

Startup Use Case:
- Challenge: Uncertain future compute needs, limited cash flow
- Old model: AWS forces 1-year reserved instance to get decent pricing, tying up $50,000-100,000
- io.net model: Pay on-demand at prices equal to or better than AWS reserved, zero upfront cost
- Benefit: Preserve cash, flexibility to scale up or down based on traction

Research Lab Use Case:
- Challenge: Grant funding is project-specific with defined timelines (6-18 months)
- Old model: Buy servers ($50,000-200,000) or commit to 1-3 year cloud contracts that outlast grant
- io.net model: Pay on-demand for duration of grant, stop when project ends
- Benefit: No capital expenditure, no waste from unused committed capacity

Enterprise Use Case:
- Challenge: Production inference API with unknown scaling trajectory
- Old model: Commit to 100 GPUs for 1 year based on projection, actually need 150 GPUs (insufficient) or 60 GPUs (wasted money)
- io.net model: Start with 50 GPUs on-demand, scale to 150 GPUs as needed, scale back to 60 in off-season
- Benefit: Pay only for actual usage, no over-provisioning waste

Cost Comparison: No Commitment vs. Locked Contracts

Scenario: AI startup training and deploying models over 18 months

AWS Approach:
- Month 1-3: Light usage (development) - 5 GPUs × 8 hrs/day
- Month 4-12: Heavy usage (production launch) - 30 GPUs × 24 hrs/day
- Month 13-18: Moderate usage (steady state) - 15 GPUs × 24 hrs/day

Option 1: AWS On-Demand Throughout
- Months 1-3: 5 × 8 × 30 × $4.10 = $4,920/month = $14,760 total
- Months 4-12: 30 × 24 × 30 × $4.10 = $88,560/month = $797,040 total
- Months 13-18: 15 × 24 × 30 × $4.10 = $44,280/month = $265,680 total
- Total: $1,077,480

Option 2: AWS 1-Year Reserved (locked into 30 GPUs based on month 4-12 projection)
- Months 1-3: 30 GPUs committed, only use 5 (waste 25 GPUs) = $63,504/month = $190,512 total
- Months 4-12: 30 GPUs = $63,504/month = $571,536 total (9 months)
- Months 13-18: Stuck with 30 GPUs, only need 15 (waste 15) = $63,504/month = $381,024 total
- Total: $1,143,072 (more expensive than on-demand due to over-commitment)

Option 3: io.net No Commitment
- Months 1-3: 5 × 8 × 30 × $1.49 = $1,788/month = $5,364 total
- Months 4-12: 30 × 24 × 30 × $1.49 = $32,184/month = $289,656 total
- Months 13-18: 15 × 24 × 30 × $1.49 = $16,092/month = $96,552 total
- Total: $391,572 (64% cheaper than AWS on-demand, 66% cheaper than AWS reserved)

Key Insight: io.net's no-commitment on-demand pricing ($391K) beats both AWS on-demand ($1.08M) and AWS's locked 1-year commitment ($1.14M). You don't need contracts to get best pricing.

Pay-Per-Second Billing Explained

Precision Billing:
- Charges calculated every second
- Minimum charge: 1 second ($0.00006 for RTX 4090 at $0.18/hr)
- Stop instance immediately = billing stops immediately

Example:

# Launch GPU at 10:00:00 AM
io launch --gpu A100-80GB

# Stop GPU at 10:37:42 AM (37 minutes, 42 seconds later)
io stop <instance-id>

# Charge: $1.49/hr × (37min 42sec / 60min) = $0.935
# Not rounded to 1 hour ($1.49) like some providers

Comparison to Competitors:
- AWS: Per-second billing (same as io.net) ✅
- Azure: Per-minute billing (rounds up partial minutes) ⚠️
- GCP: Per-second billing with 1-minute minimum ⚠️
- Lambda Labs: Per-minute billing (rounds up) ⚠️
- io.net: True per-second, no minimum ✅

Annual Savings from Second-Level Precision:
For a team running 100 short jobs per day (avg 7 minutes each):
- Per-minute billing (rounds 7min to 7min): No waste
- Per-minute billing with 1-min minimum (rounds 7min to 7min): No waste
- But for 3.5-minute jobs:
- Per-minute (rounds up to 4min): 14% waste = $12,000/year excess
- Per-second: No waste = Save $12,000/year

Per-second billing benefits workloads with many short GPU sessions (batch inference, rapid experimentation).

What About "Subscription" Models?

Some platforms offer subscriptions. How do they compare?

Subscription Model Example (Competitor X):
- $299/month subscription
- Includes 100 GPU-hours of mid-tier GPUs
- Overage: $3.99/hour
- Commit to 12 months or pay 20% premium

io.net On-Demand Model:
- $0/month base (no subscription)
- RTX 4090: $0.18/hr × 100 hours = $18/month
- A100 80GB: $1.49/hr × 100 hours = $149/month
- No commitment

When Subscriptions Win:
- Never, unless you consistently use exactly the included hours on exactly the included GPU type. Any variance (use more, use less, need different GPU) costs extra.

When On-Demand Wins:
- Always for variable workloads (most real-world scenarios)
- Always for mixed GPU needs (training on A100, inference on RTX 4090)
- Always for uncertain usage (startups, research, seasonal businesses)

Flexibility for Seasonal or Project-Based Work

Seasonal Business Example: Tax Software Company
- Jan-Apr: Heavy usage (tax season) - 50 GPUs
- May-Dec: Light usage (maintenance) - 5 GPUs

With Contracts:
- Commit to 50 GPUs year-round to handle peak
- Pay for 50 GPUs × 8 months = waste 45 GPUs for 8 months

Without Contracts (io.net):
- Jan-Apr: Provision 50 GPUs = $53,784/month (A100 80GB, 24/7) × 4 months = $215,136
- May-Dec: Provision 5 GPUs = $5,378/month × 8 months = $43,024
- Total: $258,160

With AWS Reserved (50 GPUs year-round):
- $147,600/month × 12 months = $1,771,200
- Waste: $1,513,040 (85% of spend is unused capacity)

Savings: $1,513,040 by using io.net's no-commitment model.

Enterprise Contracts: When They Make Sense

For $50,000+/month sustained usage, custom enterprise contracts offer benefits:

What You Get:
- Custom pricing (20-25% off on-demand)
- Guaranteed GPU availability (reserved pool)
- 99.95% uptime SLA
- Dedicated account team
- Custom invoicing and payment terms (NET-30, NET-60)
- Volume commitment discounts

What You Give Up:
- Flexibility (locked into 12-24 month commitment)
- Some pricing variability (fixed rates vs. market-based on-demand)

Who Should Consider:
- Fortune 500 companies with stable, predictable AI infrastructure needs
- Large-scale production inference APIs (100+ GPUs continuously)
- Research institutions with multi-year grants exceeding $500,000
- Companies requiring SOC 2, HIPAA, or custom compliance

Who Shouldn't:
- Startups with <$10,000/month GPU spend
- Teams with variable or experimental workloads
- Companies preferring cash flow flexibility over minor discounts

Related Questions

Can I pause my account without canceling?

Yes. Simply stop all running GPU instances. Your account remains active with zero charges. Data in persistent storage continues to incur $0.05/GB/month storage fees, but you can delete volumes if you want truly zero costs. Restart anytime by provisioning new GPUs - no reactivation fees, no penalties. Many teams go months without using io.net, then spin up GPUs for specific projects, then go quiet again.

What if I need GPUs for just 1 day per month?

Perfect use case for io.net's no-commitment model. Provision GPUs for your 1-day workload (e.g., 8x A100 for 24 hours = $286), then stop. You pay $286 that month, $0 other months. Traditional cloud providers with monthly minimums or subscription models would charge you year-round. io.net's pay-per-second means you only pay for the actual 24 hours used.

How does billing work if I forget to stop an instance?

You continue to be charged per-second until you manually stop the instance. However, io.net offers auto-stop features to prevent runaway costs: io launch --gpu A100 --auto-stop-on-idle 30min stops the GPU if it's idle (low utilization) for 30 minutes. You can also set budget alerts: io alerts create --type budget --threshold 100 --action email sends email when spending reaches $100, preventing surprise bills.

Can I negotiate custom pricing without long-term commitment?

Generally no - volume discounts require some commitment to ensure sustained usage. However, io.net's commitment options are much shorter than competitors (3 months vs. 12+ months elsewhere). For very large enterprise deployments ($100,000+/month), contact [email protected] to discuss flexible options. Some custom arrangements allow month-to-month commitments with 90-day notice for cancellation.

What happens to long-running jobs if I decide to "cancel"?

"Canceling" on io.net means stopping individual GPU instances, not closing your account. If you have a 72-hour training job running and you want to stop, you manually stop the instance - the job terminates and billing stops. Your account remains active for future use. To avoid accidentally terminating long jobs, use persistent storage for checkpoints and resume capabilities. io.net doesn't have forced terminations (no spot-like interruptions), so jobs run to completion unless you manually stop them.

Start with Zero Commitment

Experience true flexibility:
- No contracts - Provision GPUs for 1 minute or 1 year, your choice
- Pay-per-second - Billing stops the instant you stop your GPU
- No minimums - Test with $1, scale to $100,000/month, scale back down
- No subscriptions - Never pay for capacity you don't use

Start provisioning → or view simple pricing →

---

Last updated: April 2026 | No contracts, no commitments, cancel anytime by stopping GPU instances