Lambda vs Hyperstack GPU pricing
Lambda and Hyperstack both sell AI-focused GPU infrastructure, but their hardware coverage, configurations, billing behavior, and surrounding services differ. Compare the exact GPU first, then price the complete workload.
Live Lambda vs Hyperstack prices
lowest valid tracked listing · checked 2026-10-26| GPU | Lambda low | Hyperstack low | Published-rate gap | |
|---|---|---|---|---|
| H100 SXM5 80GB | $3.99/hr on demand | $3.20/hr on demand | Hyperstack 20% lower | verify ↗ |
| B200 | $6.69/hr on demand | $6.00/hr on demand | Hyperstack 10% lower | verify ↗ |
Directional list-price comparison only. Each cell is the provider's lowest current valid listing for the same canonical GPU. Billing mode, GPU count, CPU, memory, storage, availability, region, and taxes may differ. Open the linked market and source before relying on a rate.
The short answer
there is no provider-wide winnerChoose by exact hardware
Keep H100 SXM, PCIe, and NVL separate, and verify memory size. A lower rate for another form factor is not a price match for the GPU you need.
Choose by deployable capacity
A published price is useful only if the required GPU count, region, and topology are available during the job window. Check stock immediately before deployment.
Choose by workload cost
Add storage, IPs, data movement, idle time, support, taxes, and operational labor. The cheapest GPU-hour and the cheapest completed workload can be different.
Billing and included resources
verified against provider documentation| Factor | Lambda | Hyperstack | Buyer check |
|---|---|---|---|
| On-demand billing | Hourly price, billed in one-minute increments while running | Per-GPU hourly list rate; on-demand pricing states per-minute billing | Launch, stop, hibernate, and delete behavior |
| Configuration | Published instance shapes bundle GPUs with stated vCPU, RAM, and local storage | GPU flavors include allocated CPU, RAM, and local storage | Compare the complete shape, not only GPU count |
| Additional resources | Filesystems are separately billable; taxes may apply | Public IP and attached shared storage can add hourly charges | Persistent storage, IPs, snapshots, and tax |
| Longer commitments | 1-Click Clusters use reservation terms separate from on-demand instances | Reservation pricing and contracts are separate from on-demand listings | Term, capacity guarantee, payment, and exit terms |
| Availability | Self-serve instance availability varies by shape | Stock varies by GPU model, region, and requested count | Live inventory for the target region and quantity |
How to decide
run the same checklist for both providers1. Define the job
Record GPU variant, VRAM, GPU count, interconnect, CPU/RAM floor, storage throughput, region, runtime, and interruption tolerance.
2. Match billing modes
Compare on-demand with on-demand. Evaluate reservations or spot separately and label capacity guarantees and interruption risk.
3. Check usable stock
Confirm that the exact shape is deployable, not merely listed. For multi-GPU training, validate node and network topology as well as count.
4. Model utilization
Estimate setup, data loading, checkpointing, queueing, and idle time. Minute-level billing does not eliminate low utilization.
5. Add external costs
Include persistent storage, retained IPs, snapshots, data movement, support, tax, and duplicated tooling in a multi-cloud setup.
6. Run a benchmark
Use a representative job to compare throughput and reliability. Convert the result into cost per completed training step or inference unit.
Current provider coverage
tracked canonical GPUsLambda · 6 GPUs
Open Lambda's current tracked prices and jump from each GPU to the full cross-provider market.
Hyperstack · 4 GPUs
Open Hyperstack's current tracked prices and verify exact GPU markets and source context.
Compare cloud models
See when a specialized GPU cloud or a broader hyperscaler fits the complete workload.
Turn the GPU-hour into a workload estimate
Use utilization, rental hours, electricity, and ownership inputs to test the economics behind the headline rate.
Lambda vs Hyperstack FAQ
pricing and deployment boundariesIs Lambda or Hyperstack cheaper for GPUs?
It depends on the exact GPU, configuration, billing mode, and current listing. The live table compares the lowest valid tracked rate for models available from both providers; it does not assume one provider is always cheaper.
Are Lambda and Hyperstack prices directly comparable?
Only after checking the exact GPU variant and what the configuration includes. CPU, memory, local storage, persistent storage, IP addresses, GPU count, region, taxes, and contract terms can change total workload cost.
How does Lambda bill on-demand instances?
Lambda documents its On-Demand Cloud instances as hourly-priced and billed in one-minute increments while the instance is running. Cluster reservations use separate terms.
How does Hyperstack bill GPU virtual machines?
Hyperstack publishes per-GPU hourly list rates and says its on-demand platform bills by the minute. Its documentation notes that stopped virtual machines can continue to hold billable hardware; deleting or hibernating has different billing effects.
Which provider should I choose for H100?
Start with the exact H100 form factor, required GPU count, topology, region, and availability. Then compare the complete machine and storage cost, operational tooling, support, and expected utilization—not only the per-GPU headline rate.
Provider sources and methodology
Product and billing context comes from Lambda's official instance pricing and billing documentation, plus Hyperstack's official GPU pricing and resource-pricing documentation. The live table uses gpuprice.fyi's normalized daily collection and preserves provider, billing, source, and quality fields. Read the methodology →