Lambda Cloud is one of the more established GPU cloud providers for H100 access. Their pricing has gotten more competitive over the past two years, but the structure, per-hour billing, reserved contracts for the best rates, and no spot options, creates a cost profile that may or may not match your workload. This post covers the actual per-hour rates, breaks down the 1x vs 8x node pricing, and compares Lambda against Spheron, RunPod, and CoreWeave so you can make a direct comparison.
Lambda Cloud H100 Pricing (2026)
Lambda Cloud offers two H100 configurations: single H100 PCIe instances and 8-GPU H100 SXM nodes. The per-GPU price differs significantly between them.
| Configuration | On-Demand $/hr | Reserved 1-Year | Reserved 3-Year |
|---|---|---|---|
| H100 PCIe (1x) | $3.29 | ~$2.63 | ~$2.43 |
| H100 SXM (8x node, per GPU) | $3.99 | ~$3.19 | ~$2.95 |
Lambda no longer publicly lists all reserved contract rates on their main pricing page. The figures above for reserved pricing are estimates based on their historical ~20% discount for 1-year and ~26% for 3-year contracts. Contact Lambda for exact contract quotes.
For context on what these H100 configurations actually deliver, see the H100 SXM5 specs for the full technical breakdown on memory bandwidth, NVLink topology, and FP8 throughput.
Pricing fluctuates based on GPU availability. The prices above are based on 22 May 2026 and may have changed. Check current GPU pricing → for live rates.
1x H100 vs 8x H100 Node Pricing on Lambda
Lambda's inventory split matters here. Single H100 PCIe instances are available individually. H100 SXM instances are only sold as 8-GPU nodes, meaning you pay for the entire bundle even if you need fewer GPUs.
| Configuration | Lambda On-Demand | Total $/hr | Monthly (720 hrs) |
|---|---|---|---|
| 1x H100 PCIe | $3.29/GPU | $3.29 | $2,369 |
| 8x H100 SXM (per GPU) | $3.99/GPU | $31.92 (8 GPUs) | $22,982 |
For teams who need only 2-4 GPUs for training runs, Lambda's 8-node-only SXM offering is a poor fit. You end up paying for 8 GPUs when you need 4, which doubles your effective per-job cost.
For a detailed comparison of H100 PCIe vs SXM form factor differences, including when the NVSwitch interconnect of SXM actually matters for your workload, that guide covers the tradeoffs in depth.
Pricing fluctuates based on GPU availability. The prices above are based on 22 May 2026 and may have changed. Check current GPU pricing → for live rates.
Lambda Reserved Capacity and Commitment Discounts
Lambda's reserved pricing works like most cloud reserved contracts: you commit to 1 or 3 years of capacity in exchange for a discount off on-demand rates.
For H100 PCIe:
- On-demand: $3.29/hr
- 1-year reserved: ~$2.63/hr (20% off)
- 3-year reserved: ~$2.43/hr (26% off)
Break-even calculation: if you run H100 instances more than roughly 80% of the hours in a month, reserved starts to save money versus on-demand. At 100% utilization, the 3-year reserved saves $0.86/hr per GPU, or about $619/month per H100.
The catch is the commitment. Three years is a long window in GPU hardware. H200 and B200 instances are now available at competitive prices. A team that locked into 3-year H100 contracts in 2023 is now paying above-market rates compared to newer hardware options.
Spheron takes the opposite approach: no reserved contracts, per-minute billing, and pricing that reflects current market rates without locking you into anything. For a detailed look at the math behind on-demand vs reserved GPU pricing tradeoffs, that post covers break-even analysis at various utilization rates.
Lambda H100 Hidden Costs
Egress: Lambda offers free data egress. This is a genuine advantage over AWS, Azure, and GCP, which charge $0.08-$0.12/GB. For teams moving large model checkpoints or training datasets between providers, free egress saves real money.
Storage: Lambda persistent storage volumes are priced separately from compute. If you use significant persistent storage (model weights, datasets), factor this into your total cost.
Per-hour billing: Lambda bills in full-hour increments. A 47-minute training job costs one full hour. Run 10 such jobs daily and you pay for 10 hours but only use 7.8. At $3.29/hr, that's $7.13/day in rounding waste, or about $214/month per GPU for intermittent usage patterns.
No spot instances: Lambda has no spot or preemptible offering. You either pay on-demand rates or commit to a contract. For fault-tolerant training workloads that can checkpoint and resume, the absence of spot pricing means you're leaving 30-50% potential savings on the table compared to providers that offer it.
Availability gaps: Lambda H100 inventory goes out of stock. This isn't a hidden cost exactly, but it's a real operational risk. If your H100 capacity plan depends on Lambda availability and they're sold out, you either wait or pay on-demand rates elsewhere.
Lambda H100 vs Spheron H100: Direct Comparison
This is where the numbers matter most. Spheron pricing below is fetched live from the Spheron API as of 22 May 2026.
Teams looking for H100 instances on Spheron can access both on-demand and spot instances without any commitment, billed per minute.
| Metric | Lambda Cloud | Spheron |
|---|---|---|
| H100 PCIe on-demand | $3.29/hr | $2.09/hr |
| H100 SXM on-demand | $3.99/hr (8x node only) | $3.90/hr |
| H100 SXM spot | Not available | $1.66/hr |
| Billing granularity | Per hour | Per minute |
| Min commitment | None (on-demand) | None |
| Egress fees | Free | Included |
| Single-GPU SXM | No (8x node only) | Yes |
Per-million-tokens cost (assuming 1,500 tokens/sec throughput on H100 SXM5):
- 1M tokens = 666 seconds = 0.185 hours
- Lambda SXM on-demand: $3.99 × 0.185 = $0.74/1M tokens
- Spheron SXM5 on-demand: $3.90 × 0.185 = $0.72/1M tokens
- Spheron SXM5 spot: $1.66 × 0.185 = $0.31/1M tokens
Monthly cost for 8x H100 SXM running 24/7 (720 hrs):
- Lambda on-demand: $3.99 × 8 × 720 = $22,982
- Spheron on-demand: $3.90 × 8 × 720 = $22,464
- Spheron spot: $1.66 × 8 × 720 = $9,562
Spheron's H100 SXM5 spot comes in at 58% below Lambda's on-demand rate. Even on-demand, Spheron's $3.90/hr undercuts Lambda's $3.99/hr. The spot gap widens because Lambda has no preemptible option, while Spheron aggregates supply from 5+ providers and offers single-GPU SXM5 instances at competitive rates.
Pricing fluctuates based on GPU availability. The prices above are based on 22 May 2026 and may have changed. Check current GPU pricing → for live rates.
Lambda H100 vs RunPod, CoreWeave, and Other Neoclouds
| Provider | H100 On-Demand $/hr | Spot Available | Billing | Contract Required |
|---|---|---|---|---|
| Spheron (SXM5) | $3.90 | Yes ($1.66) | Per minute | No |
| Spheron (PCIe) | $2.09 | No | Per minute | No |
| Lambda Labs | $3.29-$3.99 | No | Per hour | Optional (reserved) |
| RunPod | $2.69 | Yes | Per hour | No |
| Vast.ai | ~$1.53-$2.27 | Yes | Per hour | No |
| CoreWeave | ~$6.16 | No | Per hour | Volume |
| Nebius | $2.95 | No | Per hour | No |
| Hyperstack | ~$1.90 | No | Per hour | No |
For a complete breakdown across all providers, see the GPU cloud pricing benchmark across 15+ providers with spot, on-demand, and reserved rates for H100, H200, B200, and A100.
If you're specifically shopping for Lambda alternatives beyond just H100 pricing, the full Lambda Labs alternatives comparison covers support quality, availability, tooling, and use-case fit across 10 providers.
When Lambda Makes Sense vs When to Switch
Lambda is a reasonable choice if:
- You have a long-term committed team that values a single-vendor relationship with established support
- Egress costs are a major factor (large dataset transfers make Lambda's free egress worth more)
- You're already on a Lambda reserved contract that hasn't hit break-even yet, so switching incurs switching costs
- You specifically need Lambda's academic/research relationships for institutional work
Consider switching when:
- Your workloads are bursty or short-duration. Per-hour rounding on Lambda hurts teams running many short jobs. Spheron's per-minute billing eliminates this overhead.
- You need spot instances for fault-tolerant training. Lambda has no spot offering, which means you can't access the 30-50% savings that come with interruptible workloads.
- You need single H100 SXM5 instances. Lambda only sells SXM in 8-node bundles at $3.99/GPU. Spheron offers single-GPU SXM5 access.
- You're evaluating cost vs newer hardware like H200 or B200. Lambda's reserved contracts can become a cost anchor when newer, cheaper alternatives become available.
Spheron H100 Pricing: Live Rates (May 2026)
Current Spheron H100 on-demand rates fetched from the Spheron pricing API on 22 May 2026:
| GPU | On-Demand $/hr | Spot $/hr | Notes |
|---|---|---|---|
| H100 SXM5 | $3.90 | $1.66 | Single-GPU access available |
| H100 NVL | $2.15 | N/A | 8-GPU bundle |
| H100 PCIe | $2.09 | N/A | Lowest H100 entry tier |
Spheron bills per minute with no minimum commitment. No reserved contracts are required to access these rates. You get SSH root access on bare-metal hardware within minutes of provisioning.
Deployment guides and SSH setup docs are available at docs.spheron.ai for connecting to your instance and configuring distributed training or inference workloads.
Check the Spheron GPU catalog for H100 availability and one-click provisioning.
Pricing fluctuates based on GPU availability. The prices above are based on 22 May 2026 and may have changed. Check current GPU pricing → for live rates.
Lambda Cloud H100 on-demand runs $3.29-$3.99/hr with per-hour billing and no spot option. Spheron's H100 SXM5 is $3.90/hr on-demand and $1.66/hr spot, billed per minute with no commitment required.
Rent H100 on Spheron → | View live pricing → | Get started →
Quick Setup Guide
Visit cloud.lambdalabs.com and navigate to the GPU instances page. Select H100 from the instance type filter. Note whether 1x H100 PCIe, 8x H100 SXM, or both are available in your preferred region.
Multiply your expected daily GPU-hours by 30 and by Lambda's on-demand rate ($3.29-$3.99/hr depending on config). Factor in storage if you'll use persistent Lambda storage volumes. Compare this against Spheron's per-minute billing to find the break-even point for your workload length.
Go to spheron.network/pricing/ and filter by H100. Check both on-demand and spot rates. Spheron H100 SXM5 spot at $1.66/hr is 50% below Lambda's H100 PCIe on-demand rate, making it cost-effective for fault-tolerant training jobs with checkpointing.
Sign in at app.spheron.ai, select H100 from the GPU catalog, choose on-demand or spot, and deploy. You get SSH root access within minutes. Spheron bills per minute with no minimum commitment.
Frequently Asked Questions
Lambda Cloud charges $3.29/hr for a single H100 PCIe on-demand and $3.99/hr for a single H100 SXM on-demand (part of an 8x node). With a 3-year reserved contract, rates drop to roughly $2.43/hr for H100 PCIe and around $2.95/hr for SXM. Spheron lists H100 PCIe at $2.09/hr on-demand and H100 SXM5 at $3.90/hr on-demand and $1.66/hr spot, with no commitment required.
As of 2026, Lambda Cloud does not offer spot or preemptible H100 instances. Their pricing model is on-demand or committed reserved. Alternatives like Spheron offer H100 SXM5 spot instances at $1.66/hr, billed per minute with no minimum commitment.
Lambda Cloud offers free data egress, which is a genuine advantage over hyperscalers. However, watch for: persistent storage fees charged separately per GB/month, per-hour billing that rounds up partial hours versus per-minute billing on Spheron, and the opportunity cost of reserved contracts that lock you into 1-3 years without flexibility. For teams with bursty workloads, per-hour minimums add up.
Yes, significantly. AWS H100 on-demand runs around $6.88/hr per GPU, and Azure's ND H100 v5 instances price per GPU at roughly $12.29/hr. Lambda Cloud H100 on-demand at $3.29-$3.99/hr is 2-3x cheaper than hyperscalers. Spheron's H100 at $2.09/hr PCIe on-demand or $1.66/hr SXM5 spot is cheaper still.
Lambda Cloud's primary discount mechanism is reserved capacity contracts. 1-year contracts offer roughly 20% off on-demand rates; 3-year contracts can reach 26-35% off. There is no per-seat volume discount for on-demand usage. For teams needing 8+ GPUs without a multi-year lock-in, providers like Spheron offer comparable hardware at lower base rates without commitment requirements.