GPU data health.
Every 3 days a GitHub Action reads 10 providers without a single credential: public keyless APIs, one catalogue a provider publishes inside its own page, and server-rendered price tables. Each result is checked against the one before it. A fetch that fails or looks broken is refused, the last good reading keeps the section up, and the provider is named here with its own date.
The last refresh read 10 of 10 providers live on 1 Nov 2026, covering 178 machines. Every provider's own reading is from today.
Last refresh
1 Nov 2026
12:16 UTC
Providers read
10/10
all read live
Machines
178
over 31 GPU models
Flagged unavailable
0
every machine is orderable
Verda (DataCrunch)
datacrunch
57 machines
read
1 Nov 2026
RunPod
runpod
38 machines
read
1 Nov 2026
Lambda
lambda
22 machines
read
1 Nov 2026
Akamai Cloud (Linode)
linode
13 machines
read
1 Nov 2026
Vultr
vultr
12 machines
read
1 Nov 2026
Modal
modal
11 machines
read
1 Nov 2026
CoreWeave
coreweave
9 machines
read
1 Nov 2026
Crusoe
crusoe
7 machines
read
1 Nov 2026
Baseten
baseten
5 machines
read
1 Nov 2026
Together AI
together
4 machines
read
1 Nov 2026
Prices being checked
1 of 11 audited prices matched the provider's page, week of 19 Oct; 5 could not be read.
A price that moves 30% or more, or would reach a top three, waits for a second reading at least 12 hours later and is checked against the provider's own page, and so is one that fails a check or that a reader reports. Meanwhile the machine shows its last confirmed price and leads no list. Today 5 differ from the provider's page.
2H100.80S.80V · 2x H100 SXM5 80GB at Verda (DataCrunch)
page disagrees
shows $4.56 per GPU-hour from 25 Oct · read $4.79 per GPU-hour on 31 Oct · the second reader read 10.54 USD per hour for all 2 GPUs on the provider's page: "{"@id":"https://verda.com/pricing#offer-169","@type":"Offer","name":"2x H100 SXM5 80GB serverless continuous","description":"Hourly serverless container price","url":"https://verda.com/pricing","price":10.54,"priceCurrency":"USD","priceSpec"
2H200.141S.88V · 2x H200 SXM5 141GB at Verda (DataCrunch)
page disagrees
shows $5.60 per GPU-hour from 18 Oct · read $6.19 per GPU-hour on 30 Oct · the second reader read 13.75 USD per hour for all 2 GPUs on the provider's page: "{"@id":"https://verda.com/pricing#offer-161","@type":"Offer","name":"2x H200 SXM5 141GB serverless continuous","description":"Hourly serverless container price","url":"https://verda.com/pricing","price":13.75,"priceCurrency":"USD","priceSpe"
8B200.240V · 8x B200 SXM6 180GB at Verda (DataCrunch)
page disagrees
shows $7.56 per GPU-hour from 11 Oct · read $8.78 per GPU-hour on 30 Oct · the second reader read 78.05 USD per hour for all 8 GPUs on the provider's page: ""@id":"https://verda.com/pricing#offer-157","@type":"Offer","name":"8x B200 SXM6 180GB serverless continuous","description":"Hourly serverless container price","url":"https://verda.com/pricing","price":78.05,"priceCurrency":"USD","priceSpec"
8H100.80S.176V · 8x H100 SXM5 80GB at Verda (DataCrunch)
page disagrees
shows $4.56 per GPU-hour from 25 Oct · read $4.75 per GPU-hour on 30 Oct · the second reader read 42.18 USD per hour for all 8 GPUs on the provider's page: ""@id":"https://verda.com/pricing#offer-173","@type":"Offer","name":"8x H100 SXM5 80GB serverless continuous","description":"Hourly serverless container price","url":"https://verda.com/pricing","price":42.18,"priceCurrency":"USD","priceSpeci"
2x NVIDIA H100 SXM at Lambda
page disagrees
shows $4.19 per GPU-hour from 18 Oct · read $4.19 per GPU-hour on 18 Oct · the second reader read 3.99 USD per GPU per hour on the provider's page: "NVIDIA H100 SXM 80 GB 208 1800 GiB 22 TiB SSD $3.99"
How the numbers are protected
- Only fetched numbers. An adapter throws on anything it does not recognise, and nothing is typed in by hand.
- A broken read is refused. A fetch whose machine count falls below half the last one, or whose median rate per GPU-hour moves more than 2.5 times either way, is rejected and the previous reading kept, so a half-broken parser cannot publish nonsense.
- A changed price waits. A move of 30% or more, or into a top three, shows once a second reading agrees.
- The board is not the provider's to describe. Memory, bandwidth, architecture and launch year come from a registry a person writes. An adapter meeting a board with no entry throws, so a new GPU arrives as a visible failure rather than as a page printing a chip with no specification.
- One number, whatever the machine. Every rate is divided down to one GPU for one hour, which is the only way a single card and a rack compare. The market median today is $2.27/GPU-hr.
- History on the published number. A currency moving is not a price change.