Lambda is a research-grade GPU cloud with a short, intentional catalogue: H100, A100, B200 and a few others, mostly sold in 8-GPU HGX nodes with InfiniBand. Its 1-Click Clusters, reserved terms and Quantum-2 fabric are designed for teams running sustained large-model training. It is not designed for renting one GPU at a time.
packet.ai is. This comparison is about what that difference costs and who it matters to.
Platform overview: packet.ai vs Lambda
What Lambda offers
Lambda's on-demand list is short and premium: A100 80GB from $1.99/hr (1×), A100 SXM in 8× nodes at $2.79/hr, H100 PCIe at $3.29/hr, H100 SXM at $3.99/hr per GPU in 8× nodes ($4.29/hr for a single card), and B200 at $6.69/hr per GPU in 8× nodes or $6.99/hr as a single instance. 1-Click Clusters run HGX B200 and H100 with Quantum-2 InfiniBand.
- Per-minute billing, no egress fees, persistent filesystem around $0.20/GiB-month that keeps billing while instances are stopped
- Reserved 1-year and 3-year terms take H100 to roughly $2.63 and $2.43/hr
- No consumer or workstation GPUs: nothing below the A100 except a legacy V100/A10 line
- Single-GPU capacity is frequently out of stock; 8× nodes and reserved clusters are the core product
What packet.ai delivers
Lambda is built around 8-GPU nodes and reserved cluster capacity. packet.ai is built around one card, available now, at a published rate. That difference shows up before you even get to price: Lambda's single-GPU on-demand inventory is frequently out of stock, and its catalogue stops at the A100 on the low end. packet.ai starts below the A100.
On the two cards both list at published on-demand rates, packet.ai is cheaper: A100 80GB at $1.43/hr Dedicated versus Lambda's $1.99/hr (28% less), and B200 at $3.75/hr Dynamic versus Lambda's $6.99/hr single-instance rate (46% less). Below the A100, Lambda has nothing; packet.ai has RTX 4090 at $0.39/hr, L40S at $0.92/hr and RTX 6000 Pro at $0.66/hr Dynamic.
- Dynamic: scheduler-enforced multi-tenant on first-party hardware. Same peak performance and VRAM as Dedicated. RTX 6000 Pro at $0.66/hr, B200 at $3.75/hr. Available now, billed hourly, no minimum.
- Dedicated: a whole card, single-tenant, 99% SLA. RTX 4090 at $0.39/hr, L40S at $0.92/hr, A100 80GB at $1.43/hr, B200 at $6.99/hr.
- Clusters: InfiniBand-connected multi-node fabric up to 1,024+ GPUs for training runs that genuinely need scale, priced around 30% below retail on committed terms.
What Lambda has and packet.ai does not: H100 on-demand at published rates, Quantum-2 InfiniBand clusters with 1-Click provisioning, and reserved 1- and 3-year terms. If your workload is a sustained multi-node training run and you can commit, Lambda's reserved cluster pricing is worth evaluating. If you need one card today at a published rate, packet.ai is available and cheaper on the shared cards.
packet.ai vs Lambda at a glance
Published on-demand rates in USD per GPU-hour. Lambda 8× node rates are per-GPU.
| GPU | packet.aiYOU | Lambda |
|---|---|---|
| A100 80GB PCIe | $1.43/hr Dedicated−28% | $1.99/hr (1× on-demand) |
| A100 SXM 80GB | $1.43/hr Dedicated | $2.79/hr (8× node only) |
| B200 | $3.75/hr Dynamic · $6.99/hr Dedicated−46% Dynamic | $6.99/hr (1×) · $6.69/hr (8× node) |
| H100 PCIe 80GB | Notify list | $3.29/hr |
| H100 SXM 80GB | Notify list | $3.99/hr (8×) · $4.29/hr (1×) |
| RTX 6000 Pro (96GB) | $0.66/hr Dynamic | Not offered |
| RTX 4090 (24GB) | $0.39/hr Dedicated | Not offered |
| L40S (48GB) | $0.92/hr Dedicated | Not offered |
| Single-GPU availability | Available on-demand | Frequently out of stock |
| Minimum buy | 1 GPU, 1 hour | 1 GPU on-demand; 8× node for SXM and clusters |
| Billing | Hourly; monthly commits up to 20% off | Per-minute; 1- and 3-year reserved terms |
| Egress | $0.04/GB | No egress fees |
| InfiniBand clusters | Yes (committed terms) | Yes, Quantum-2, 1-Click provisioning |
Detailed comparison
Availability
Lambda's single-GPU on-demand capacity is a known constraint. The A100 1× instance is listed but frequently shows no availability; the H100 SXM single card is $4.29/hr when it exists. The core Lambda product is the 8× node and the reserved cluster, not the single card. For teams that want to spin up one GPU today and get to work, this is a real friction point.
packet.ai's Dynamic and Dedicated inventory is available on-demand across all listed cards. No waitlist, no availability check, no node minimum.
Price on shared cards
Where Lambda and packet.ai list the same card, packet.ai is cheaper. A100 80GB is $1.43/hr versus Lambda's $1.99/hr on-demand. B200 Dynamic at $3.75/hr is 46% below Lambda's $6.99/hr single-instance rate, and even Dedicated at $6.99/hr matches Lambda rather than exceeding it.
Below the A100, the comparison does not exist: Lambda has no RTX 4090, no L40S, no RTX 6000 Pro. For fine-tuning, inference development or any workload that does not need datacenter-class silicon, Lambda is simply not an option. packet.ai has four cards in that tier.
Where Lambda wins
Reserved cluster pricing is Lambda's real advantage. A 1-year H100 SXM reservation at roughly $2.63/hr per GPU is competitive for a team that knows exactly what it needs for the next twelve months. The Quantum-2 InfiniBand fabric and 1-Click Clusters make large distributed training simpler to provision than assembling it yourself. For frontier model training at scale on a committed budget, Lambda is purpose-built.
Who should choose which
Choose packet.ai if
- You need one GPU available now without joining a waitlist
- Your workload fits RTX 4090, L40S or RTX 6000 Pro - cards Lambda does not offer
- You want A100 or B200 at a lower on-demand rate and hourly billing without a long commit
- You are in the US or EU and want to grow into InfiniBand clusters at 30% below retail
Choose Lambda if
- You are running a sustained multi-node H100 or B200 training job
- You can commit to a 1- or 3-year reserved term and want the lower rate
- You need Quantum-2 InfiniBand with 1-Click cluster provisioning
- Per-minute billing granularity matters for short burst jobs
Conclusion
Lambda and packet.ai serve different ends of the GPU demand curve. Lambda is for teams that know they need 64 H100s for six months and want enterprise InfiniBand provisioning. packet.ai is for teams that need one card today, want it cheaper than Lambda's on-demand rate, and are not ready to commit to a reserved cluster. On the cards both list, packet.ai is cheaper. On the cluster and reserved use case, Lambda is purpose-built.
Same silicon. Smarter economics.
Every rate on this page is a published starting price you can deploy against today. Dynamic launches in under five minutes; no credit card to start.
Additional resources
- packet.ai GPU pricing: the full Dynamic, Dedicated and Clusters rate card.
- packet.ai documentation: API, CLI and platform guides.
- How intelligent scheduling works: why Dynamic keeps peak performance at a lower price.
- Clusters and wholesale quotes: multi-node InfiniBand capacity at ~30% below retail.
- Rent GPUs on packet.ai: specs, pricing and workload fit for every card.