Not on its pricing page. Every Pod GPU listed there carries an hourly rate, from $0.27 for an RTX A5000 to $7.89 for a B300. Among the alternatives here, Modal's Starter plan includes $30 a month of free compute.

On price, it's the cheapest of the eight fixed-price GPU clouds checked for an on-demand H100, and it lists more GPU types than any of them. The company says its Pods run across "30+ regions". This ranking doesn't measure uptime, support or capacity, so it can't judge RunPod on those.

RunPod alternatives by H100 price: September 2026

No alternative we checked lists a cheaper on-demand H100 than RunPod. RunPod’s H100 PCIe costs $2.89 per GPU-hour on its own pricing page, checked 28 September 2026.¹ The cheapest alternative is Lambda at $3.29 for a one-GPU H100 PCIe instance.² Nebius at $3.85 and Crusoe at $3.90 come next.^3,4

So a lower H100 list price is a poor reason to leave RunPod. Billing shape, cluster size and a managed runtime are better ones.

TL;DR

RunPod has the lowest listed on-demand H100 price per GPU-hour, checked 28 September 2026, on every H100 variant it sells: $2.89 for PCIe, $3.19 for NVL and $3.49 for SXM.¹
Lambda is the cheapest alternative at $3.29, which is 40 cents (14%) above RunPod for the same PCIe card.²
Four alternatives sit within 14 cents of each other: Nebius $3.85, Crusoe $3.90, Modal $3.95 and Together AI $3.99.^3,4,5,6
CoreWeave prices its on-demand HGX H100 as an eight-GPU node at $49.24 an hour. Per GPU that’s $6.16, more than double RunPod’s rate.⁷
Vast.ai’s marketplace median of $2.16 an hour for an H100 SXM undercuts every list price, but hosts set that price and it changes hourly.⁸
Nebius’s H100 rate rises to $4.50 from 1 October 2026. That drops it from third to seventh.³

How this ranking was built

The ranking uses one field from each provider’s own public pricing page: the lowest listed on-demand H100 price per GPU-hour, for any H100 variant the provider sells. It sits within our model infrastructure coverage. All nine pages were captured on 28 September 2026.

Prices are in US dollars. We treat every figure as a pre-tax list price. Lambda states that its rates are “plus applicable sales tax/VAT/GST”.² DigitalOcean says it applies taxes in some countries.⁹ Each page names the field differently. Nebius calls it On-demand, GPU-hour, Lambda PRICE/GPU/HR and CoreWeave On-Demand Price (Per Hour).^3,2,7

Pages captured: 9. RunPod, Lambda, Nebius, Crusoe, Modal, Together AI, DigitalOcean, CoreWeave and Vast.ai.
Providers with a fixed on-demand H100 price: 8. None of the eight omits the H100.
Excluded from the rank: 1. Vast.ai publishes live prices “set by supply and demand across 40+ data centers”, so it appears as a reference line under the table.⁸
Excluded prices: spot, preemptible, reserved and promotional rates. They appear in the table’s notes column and in the section on why published prices disagree.
Conversions: Modal lists the H100 SXM5 at $0.001097 per second. Multiplied by 3,600, that’s $3.95 an hour.⁵ CoreWeave lists an eight-GPU HGX H100 node at $49.24 an hour. Divided by eight, that’s $6.155, shown as $6.16.⁷
Rows ranked: 8. RunPod plus seven alternatives.

These are prices the companies publish, not prices customers pay. Contract and reserved pricing runs through sales at most of them. Our research methodology sets out how Glamdring dates and checks a figure.

Which RunPod alternative lists the cheapest on-demand H100?

Lambda does, at $3.29 per GPU-hour for a one-GPU H100 PCIe instance.² RunPod stays first overall at $2.89.¹

Rank	Provider	H100 variant priced	On-demand price (USD per GPU-hour, pre-tax)	Other listed H100 prices
1	RunPod	H100 PCIe 80 GB	2.89	H100 NVL $3.19; H100 SXM $3.49¹
2	Lambda	H100 PCIe 80 GB, 1x instance	3.29	H100 SXM $3.99 (8x) to $4.29 (1x)²
3	Nebius	HGX H100	3.85	$4.50 from 1 October 2026; spot from $0.79³
4	Crusoe	HGX H100 80 GB	3.90	Spot through sales⁴
5	Modal	H100 SXM5	3.95	Listed as $0.001097 per second⁵
6	Together AI	HGX H100, GPU cluster	3.99	Preemptible $1.99; reserved $3.69 to $3.19⁶
7	DigitalOcean	HGX H100	4.41	12-month reserved $3.26⁹
8	CoreWeave	HGX H100, 8-GPU node	6.16	Node $49.24 an hour; spot $19.71 per node-hour⁷
Ref.	Vast.ai (marketplace, not ranked)	H100 SXM	2.16 median	Offers from $1.73 an hour⁸

The spread is wide at the ends and tight in the middle. From RunPod’s $2.89 to CoreWeave’s $6.16 is a factor of 2.13. Between third and sixth place, four providers fall within 14 cents: $3.85 to $3.99. For most of the list, the H100 price alone won’t decide the move.

Does any alternative beat RunPod on the same H100 variant?

No. RunPod’s price is lower than every alternative’s on each H100 variant it lists.¹

The providers don’t use one naming scheme, so compare like with like where a page allows it:

H100 PCIe: RunPod $2.89 against Lambda $3.29, the only other provider here that names an H100 PCIe.^1,2
H100 SXM: RunPod $3.49 against Modal’s H100 SXM5 at $3.95 and Lambda’s H100 SXM at $3.99 to $4.29, depending on instance size.^5,2
H100 NVL: RunPod lists it at $3.19 with 94 GB of memory. No other fixed-price provider in this set lists an H100 NVL.¹
HGX H100: Nebius, Crusoe, Together AI, DigitalOcean and CoreWeave price an “HGX H100” and don’t split it into PCIe or SXM on the captured pages. All five sit above RunPod’s SXM price of $3.49, as well as its PCIe price.^3,4,6,9,7

The per-GPU price also buys different amounts of host machine. RunPod’s H100 PCIe pod lists 16 vCPUs and 188 GB of RAM.¹ Lambda’s one-GPU H100 PCIe instance lists 26 vCPUs, 225 GiB of RAM and 1 TiB of SSD.² DigitalOcean’s H100 Droplet lists 20 vCPUs, 240 GiB of memory and a 5 TiB NVMe scratch disk.⁹ A data-loading job that starves on CPU can make the cheaper GPU the dearer run.

Which GPUs does each provider list?

Every provider in the ranking lists an H100, and all eight also list a B200 or B300. The rest of the catalogue differs more than the H100 price does.

Provider	GPUs listed on the captured pricing page	H100 variants
RunPod	B300, B200, H200, H100 NVL, H100 PCIe, H100 SXM, A100 PCIe, A100 SXM, RTX Pro 6000 (plus 48 GB and 24 GB MIG slices), L40S, L40, RTX 6000 Ada, A40, RTX A6000, RTX 5090, RTX 4090, RTX 3090, RTX A5000, L4¹	PCIe, SXM, NVL
Lambda	B200 SXM6, H100 SXM, H100 PCIe, GH200, A100 SXM (80 GB and 40 GB), A100 PCIe, A10, A6000, Quadro RTX 6000, Tesla V100²	PCIe, SXM
Nebius	GB300 NVL72, GB200 NVL72, HGX B300, HGX B200, HGX H200, HGX H100, RTX PRO 6000, L40S³	HGX
Crusoe	GB200 NVL72, B200, H200, H100, A100 SXM, A100 PCIe, L40S, AMD MI355X, AMD MI300X⁴	HGX
Modal	B300, B200, H200 SXM, H100 SXM5, RTX PRO 6000, A100 80 GB, A100 40 GB, L40S, A10, L4, T4⁵	SXM5
Together AI (GPU clusters)	GB200 NVL72, GB300 NVL72, HGX B200, HGX B300, HGX H200, HGX H100⁶	HGX
DigitalOcean	HGX H200, HGX H100, RTX 4000 Ada, RTX 6000 Ada, L40S, AMD MI325X, AMD MI300X on demand; HGX B300 and AMD MI350X and MI355X on reserved or spot plans only⁹	HGX
CoreWeave	GB300 NVL72, GB200 NVL72, HGX B300, HGX B200, HGX H200, HGX H100, GH200, RTX PRO 6000 Blackwell Server Edition, L40S, L40, A100⁷	HGX (8-GPU node)

Three patterns matter for a move off RunPod. RunPod has the deepest bench of consumer and workstation cards, including the RTX 3090, 4090 and 5090.¹ Crusoe and DigitalOcean are the only fixed-price providers here that list AMD accelerators.^4,9 Together AI’s cluster table lists only HGX and NVL72 systems.⁶ Vast.ai’s marketplace lists 68-plus GPU types.⁸

How does each provider bill an H100?

Most quote the H100 per GPU-hour, but three break that pattern: Modal lists a per-second rate, DigitalOcean bills by the second, and CoreWeave prices the whole eight-GPU node.

Provider	Unit the H100 price is quoted in	Billing detail stated on the page
RunPod	Per hour	Page offers per-hour and per-second price views¹
Lambda	Per GPU per hour	Instances come in 1x, 2x, 4x and 8x sizes²
Nebius	Per GPU-hour	Commitments cut up to 35% off on-demand rates³
Crusoe	Per GPU-hour	“Pay by the hour” on demand⁴
Modal	Per second	Pay for compute time used; CPU and memory priced separately⁵
Together AI	Per GPU per hour	GPU clusters billed “on an hourly basis”⁶
DigitalOcean	Per GPU per hour	Billed per second with a 5-minute minimum; 1 or 8 GPUs per Droplet⁹
CoreWeave	Per 8-GPU node per hour	GPU count listed as 8 for the HGX H100; a separate single-GPU inference price of $6.16 an hour⁷
Vast.ai (reference)	Per hour	Per-second billing with no minimum hours and no rounding up⁸

The unit changes the bill more than the rate does for short or bursty work. CoreWeave’s on-demand HGX H100 price covers the whole node, so a job on that node costs $49.24 an hour however many of the eight GPUs it uses.⁷ At DigitalOcean a powered-off GPU Droplet is still billed until it’s destroyed.⁹ Modal argues its case with its own example: 75 GPUs for 24 hours at $3 a GPU-hour costs $5,400, against an average of 50 GPUs at $3.95 for $4,740 on Modal.⁵ That’s the company’s illustration, not a measured result. The saving depends on the lower average: 75 GPUs held for 24 hours at $3.95 would cost $7,110.

Why do other published H100 prices disagree with this table?

A lower H100 figure elsewhere is likely to be one of the prices this ranking excludes: marketplace offers, interruptible instances, reserved terms, promotions or a different product from the same provider.

Marketplace medians. Vast.ai’s H100 SXM median was $2.16 an hour, with offers from $1.73. Its H100 PCIe median was $2.53 and its H100 NVL median $2.76. The page says prices “reflect currently available offers and update hourly”.⁸
Spot and preemptible. Nebius lists preemptible HGX H100 capacity from $0.79 a GPU-hour.³ Together AI lists preemptible HGX H100 at $1.99.⁶ CoreWeave’s spot price is $19.71 per node-hour, which is about $2.46 per GPU.⁷ All three sit below RunPod’s on-demand $2.89.
Reserved terms. Together AI’s reserved HGX H100 runs from $3.69 for 7 to 30 days down to $3.19 for 91 to 180 days.⁶ DigitalOcean’s 12-month reserved H100 is $3.26.⁹
Promotions. Together AI’s dedicated-inference H100 shows a $3.99 promotional rate against a $5.49 list rate, with the promotion valid until 30 September 2026.⁶
Different products. RunPod’s Serverless H100 PRO tier is $4.79 an hour, against $2.89 for the H100 PCIe pod.¹ Lambda’s 16-GPU H100 cluster on a two-week to one-year term is $6.16 per GPU-hour, against $3.29 for a one-GPU instance.²
Dates. RunPod’s page says it was updated on 27 September 2026.¹ DigitalOcean’s current GPU pricing took effect on 1 August 2026.⁹ A list compiled before those dates may quote older rates.

Which H100 prices change next?

Nebius changes first. Its HGX H100 rate moves from $3.85 to $4.50 per GPU-hour on 1 October 2026, a rise of 65 cents or 17%.³ At $4.50, Nebius falls from third to seventh, below DigitalOcean’s $4.41.⁹

The recomputed order from 1 October, with every other page unchanged: RunPod $2.89, Lambda $3.29, Crusoe $3.90, Modal $3.95, Together AI $3.99, DigitalOcean $4.41, Nebius $4.50, CoreWeave $6.16. Lambda stays the cheapest alternative.

Together AI’s dedicated-inference promotion ends on 30 September 2026.⁶ That doesn’t move the ranked Together AI figure, which comes from its GPU cluster table.

Nebius’s rise is the first known change that reorders this table.

What the ranking cannot tell you

The ranking measures one published number. It says nothing about these:

Capacity. A list price doesn’t mean an H100 is free when you ask. Lambda describes its instances as “self-serve, first-come access”.²
Speed per dollar. PCIe, SXM, NVL and HGX listings aren’t one product. RunPod lists 80 GB for its PCIe and SXM cards and 94 GB for the NVL.¹ We didn’t benchmark any of them.
The whole bill. Storage, egress, public IPs and CPU time can sit outside the GPU rate. Modal prices CPU and memory as separate resources.⁵ CoreWeave charges $4.00 a month per public IP.⁷
Negotiated prices. RunPod, Lambda, Crusoe and DigitalOcean all route reserved or contract pricing through sales.^1,2,4,9
Reliability and support. Nothing here measures uptime, support response or how often capacity runs out.

The GPU-hour is one input to what a model costs to run. Our explainer on what controls AI inference cost covers the others. To get the next price ranking when it’s published, subscribe to the Glamdring briefing.

Frequently asked questions

What is better than RunPod?

On listed on-demand H100 price, none of the seven alternatives checked on 28 September 2026. RunPod’s $2.89 H100 PCIe is the lowest, and Lambda’s $3.29 is the nearest.^1,2 Another provider can still suit a job better: Modal lists a per-second rate, CoreWeave sells eight-GPU HGX nodes, and Crusoe and DigitalOcean list AMD accelerators.^5,7,4,9

Is RunPod free?

Not on its pricing page. Every Pod GPU listed there carries an hourly rate, from $0.27 for an RTX A5000 to $7.89 for a B300.¹ Among the alternatives here, Modal’s Starter plan includes $30 a month of free compute.⁵

How good is RunPod?

On price, it’s the cheapest of the eight fixed-price GPU clouds checked for an on-demand H100, and it lists more GPU types than any of them.¹ The company says its Pods run across “30+ regions”.¹ This ranking doesn’t measure uptime, support or capacity, so it can’t judge RunPod on those.

Is Vast.ai cheaper than RunPod for an H100?

At the median, yes. Vast.ai’s H100 SXM median was $2.16 an hour against RunPod’s $3.49, and its H100 PCIe median was $2.53 against $2.89.^8,1 Those are host-set marketplace offers that update hourly, not a fixed list price, so the rate on the day you rent can differ.⁸