Lambda Labs Pricing 2026: GPU Cost vs Spheron

Lambda Labs pricing in 2026 looks like a straightforward rate card, until you go looking for the "commit and save" tier every other GPU cloud's pricing page has trained you to expect. What's actually sitting there is 1-Click Clusters, a managed-cluster product, and it's priced above Lambda's own on-demand rate, not below it. Checked live on 23 Oct 2026, a 16-GPU H100 1-Click Cluster runs $6.16/GPU-hr against $4.29/hr for a single on-demand H100 SXM. That flips the usual assumption that reserved pricing on this specific provider means cheaper.

This post walks through what Lambda Labs actually charges across H100, A100, and B200, why 1-Click Clusters exist and what they're really for, and where Spheron's live on-demand and spot rates land on the same hardware.

TL;DR: Lambda Labs Pricing 2026

Lambda Labs charges $4.29/hr for a single on-demand H100 SXM, $1.99/hr flat for A100, and $6.99/hr for a single B200 SXM6, with per-GPU rates easing slightly at larger node sizes.

• H100: Lambda's cheapest H100 on-demand rate is $3.29/hr for H100 PCIe (1x); H100 SXM runs $4.29/hr at 1x down to $3.99/hr at the 8x tier.
• A100: Lambda prices A100 (40GB variant) at a flat $1.99/hr across its 1x, 2x, 4x, and 8x PCIe/SXM-40GB tiers.
• B200: Lambda's single B200 SXM6 on-demand rate is $6.99/hr, easing to $6.69/hr per GPU at 8x.
• 1-Click Clusters: Lambda's only committed-term product charges more per GPU-hour than on-demand, $6.16/GPU-hr for a 16-GPU H100 cluster against $4.29/hr on-demand for one H100 SXM.
• No H200: Lambda publishes no on-demand H200 rate as of 23 Oct 2026.
• Spheron: H100 SXM5 runs $2.65/hr on-demand as of 28 Oct 2026, with live H200 and B200 rates where Lambda has gaps. Check current GPU pricing for live rates.

Lambda Labs Pricing 2026: On-Demand Rates by GPU (H100, A100, B200)

Lambda Labs sells GPU instances in fixed shapes of 1, 2, 4, and 8 GPUs per node, and the per-GPU rate shifts slightly with shape rather than discounting sharply the way a reserved tier would. Here's what each Lambda Labs GPU costs on-demand, checked against Lambda's own pricing page on 23 Oct 2026:

A few things stand out once the rates are side by side. First, Lambda's A100 (40GB) pricing is genuinely flat: $1.99/hr whether you take one GPU or eight, so GPU count alone doesn't move that specific SKU's per-GPU price. That's unusual among GPU clouds, where multi-GPU nodes almost always carry at least a small per-GPU discount. Second, the 8x A100 SXM 80GB node prices higher per GPU than the 40GB variant, at $2.79/GPU-hr, which is a reminder to check which exact A100 SKU a quote refers to before comparing it against another provider's number. Third, Lambda's HGX B200 node (listed as B200 SXM6) has the widest 1x-to-8x spread of any GPU in the lineup, $6.99/hr down to $6.69/hr, a 4.3% step that's smaller in percentage terms than the H100 SXM's step from $4.29/hr to $3.99/hr.

Lambda's catalog has one more gap worth flagging up front: there's no H200 line item anywhere on its public pricing page as of this check. If H200's larger HBM3e capacity is the reason you're shopping, Lambda isn't currently in that comparison at all.

1-Click Clusters: Why Lambda's "Reserved" Pricing No Longer Means Cheaper

Every other pricing comparison on the internet, including Lambda's own marketing language, treats 1-Click Clusters as Lambda's reserved tier: commit to a term, pay less than on-demand. On Lambda's current rate card, that's backward.

Line those up against Lambda's own on-demand rates from the table above: $4.29/hr for a single H100 SXM, $3.99/hr at the 8x on-demand tier, and $6.99/hr for a single B200 SXM6. Every 1-Click Cluster row costs more per GPU-hour than Lambda's own on-demand rate for the same chip. A 16-GPU H100 cluster at $6.16/GPU-hr is 44% above the $4.29/hr single-node on-demand rate. Even the largest, cheapest tier, 256 GPUs at $5.54/GPU-hr, still sits 29% above on-demand.

That's not a pricing mistake, and it's not Lambda being uncompetitive. It's a different product. 1-Click Clusters scale from 16 to 2,000+ interconnected HGX B200 and H100 GPUs. You're not paying for a volume discount, you're paying for Lambda to build and operate a dedicated multi-node cluster around your workload, for a term of two weeks to a year (1-year-plus terms go to a custom quote). That's genuinely valuable if you need exactly that. It just means the word "reserved" on Lambda's pricing page doesn't carry the discount connotation it does everywhere else, and a pricing comparison that assumes otherwise will get the math backward.

What Lambda's Pricing Page Doesn't Show You

Three things sit outside the rate table and matter as much as the headline numbers. First, there's no H200 rate published anywhere, on-demand or committed, so any comparison built around H200 has to look elsewhere for Lambda's number because there isn't one. Second, Lambda states on its own service page that it bills GPU instances by the minute and charges no data egress fees, including for data kept on attached persistent storage between sessions. That's a real, current detail worth knowing if you've seen older write-ups describe Lambda as hourly-rounded; check the live billing terms on Lambda's own pricing page before assuming either way, since marketplace terms change. Third, "Lambda Labs" itself isn't a question mark: it operates its own GPU fleet and rents NVIDIA H100 and B200 systems directly, with 1-Click Clusters as its path to large managed deployments. The genuine open question isn't whether Lambda is a real company, it's whether its committed-term product actually saves you money, and on the current rate card it doesn't.

Spheron Price-for-Price on the Same GPUs

Spheron's pricing page lists live on-demand and spot rates per GPU, including H200 and B200, where Lambda's own page has a gap. As of 28 Oct 2026, H100 SXM5 runs $2.65/hr on-demand and $2.20/hr spot. A100 runs $1.43/hr on-demand and $1.19/hr spot. H200 runs $5.54/hr on-demand and $3.90/hr spot, a GPU Lambda doesn't publish a rate for at all. B200 runs $11.12/hr on-demand and $6.39/hr spot, worth lining up against Lambda's $6.99/hr single-GPU B200 SXM6 rate on the day you're actually deploying.

Spheron bills per minute with a 20-minute minimum and no contract, and sources those rates across multiple data center partners rather than one company's own fleet, which is the structural reason its availability and pricing move differently from a single-provider rate card. That multi-provider sourcing is also why, unlike Lambda's single-vendor catalog, a gap in one partner's H200 or B200 inventory doesn't leave the whole GPU unavailable.

Pricing fluctuates based on GPU availability. Spheron rates above are live as of 28 Oct 2026; Lambda Labs figures reflect its most recently published rates as of 23 Oct 2026 and may have changed. Check current GPU pricing → for live rates.

None of that makes Spheron the better fit for every workload. If you specifically need a single managed, InfiniBand-connected job scaling from 16 up to 2,000+ GPUs with one vendor owning the fabric and the ops, 1-Click Clusters are purpose-built for exactly that, and a reader at that scale is better served there than piecing together equivalent InfiniBand scale across a marketplace. Lambda's stated zero-egress policy, including from attached persistent storage, is also worth weighing directly against Spheron's current storage and egress terms before assuming parity for a dataset-heavy workload.

When Lambda Labs Makes Sense vs When a Multi-Provider Marketplace Wins

Pick based on what you're actually deploying, not the headline rate alone:

• A single H100, A100, or B200 for inference or a short training run: compare Lambda's on-demand rate against H100 GPU rental on Spheron and Spheron's A100 and B200 pages directly; the gap is usually visible at a glance once you're on the same tier.
• A frontier-scale, InfiniBand-connected job spanning 16 to 2,000+ GPUs with one vendor handling the fabric: Lambda's 1-Click Clusters are built for that, and the premium over on-demand buys a single point of accountability for the whole cluster.
• An H200 workload: Lambda has no published rate at all right now, so a live marketplace rate is the only option with a number attached.
• A dataset-heavy workload moving large volumes in and out: weigh Lambda's stated zero-egress policy against Spheron's current storage and egress terms before deciding.
• Price-shopping a specific GPU count without a sales call: this is where a GPU cloud pricing comparison across more than two providers earns its keep, since Lambda and Spheron aren't the only two rate cards worth checking.

If Lambda's own rate card doesn't fit what you're deploying, 10 Lambda Labs alternatives and the Runpod vs Lambda Labs comparison cover the next two most common substitutes, and the Spheron vs Lambda Labs breakdown goes deeper on availability, networking, and deployment UX than a pricing post has room for. Thunder Compute's pricing is a useful parallel read if you're comparing a single-provider flat rate card against Spheron's aggregated model on a different GPU cloud entirely.

Lambda Labs' 1-Click Clusters are built for one vendor to run your entire managed InfiniBand cluster. For everything smaller than that, from a single H100 to an H200 Lambda publishes no on-demand rate for, Spheron's aggregated on-demand and spot rates are worth checking before you sign anything.

H100 on Spheron →