DataCrunch rebranded to Verda in 2026, and datacrunch.io now redirects to verda.com. Same company, same European AI cloud, new name. If you are comparing it to Runpod, the two are closer than most matchups in this category: both are self-serve, both publish every rate, both offer serverless and self-service clusters, and neither makes you talk to sales before you can deploy.
The differences that decide it are not mostly in the rate card, so start with what each one is built around.
What each one is built around
Verda is a European full-stack AI cloud with an in-house research lab. Its lineup starts at the data center tier and goes up to rack-scale GB300, its compliance story is unusually complete for a company this size, and it sells hardware-attested confidential computing as a product rather than a feature. It is built for teams whose workload already justifies a data center GPU.
Runpod is an AI developer cloud. It covers the same data center tier and then keeps going downward, into the cards that most AI work actually runs on. A pod is running in under 30 seconds, and teams move from Pods to Serverless to Clusters without changing vendors. It is built for the stretch between the first experiment and production, which is where most teams spend most of their time.
Both are self-serve. Both publish every rate. Neither makes you sign a contract to start. That makes this a closer comparison than most in this category, and it means the price table below is worth reading carefully rather than skimming for a winner. Neither one wins it outright.
Published on-demand rates
Runpod prices in this comparison are Secure Cloud rates refreshed from the Runpod catalog. Verda prices are single-GPU on-demand, re-verified October 16, 2026 against verda.com/pricing. SXM against SXM throughout. Bundled vCPU and system RAM are shown because the hourly rate alone hides a real difference between the two.
Verda raised every on-demand rate between August and October 2026, by 13% on the A100 SXM up to 37% on the H200. That reversed most of this comparison. Runpod is now cheaper on seven of the nine cards both companies publish. Verda is cheaper on two: the B200 by eight cents an hour, and the RTX PRO 6000 by sixteen.
The widest margin is the L40S, where Runpod is 63 cents an hour cheaper. The B300 is also not quite like-for-like: Runpod’s card carries 288GB of VRAM against Verda’s 268GB, so Runpod is now cheaper both per hour and per gigabyte of memory. Verda’s remaining advantage on the B200 is small enough that bundled resources and billing behavior matter more than the rate.
The bundled resources cut the other way
Runpod ships materially more system RAM on most shared cards: 283GB against 170GB on B200, 276GB against 170GB on H200, 188GB against 90GB on RTX PRO 6000, 167GB against 60GB on RTX 6000 Ada. Verda ships more vCPU on most of them, 44 against 24 on H200 and 30 against 20 on H100.
Which one you want depends on the job. Data loading, preprocessing and multi-stream video work are CPU-bound often enough that Verda’s core count is the better deal. Anything that stages large datasets in host memory, or runs CPU offload during training, gets more out of Runpod’s RAM. This is why comparing bare hourly rates misleads in both directions.
Where the lineups genuinely differ
Runpod goes much further down the stack. Eleven Runpod cards are priced below the L40S: RTX 6000 Ada, RTX 4090, L40, the RTX PRO 6000 24GB MIG partition, A40, L4, RTX A6000, RTX 3090, RTX A5000, RTX A4000 and RTX 2000 Ada. Verda has two under a dollar: the RTX A6000 at $0.74 and a Tesla V100 at $0.22. The V100 is genuinely cheap and it is also a 2017 card with 16GB, which rules out most current work.
If your job fits on a 24GB card, Runpod has four to pick from: the RTX 4090, the RTX 3090, the L4 and the RTX A5000. Verda has effectively none.
Verda is NVIDIA only. Runpod lists AMD MI300X with 192GB at {{gpu:mi300x}}/hr. If your stack is ROCm, that settles it.
Verda goes higher at the top. GB300 NVL72 at $11.48/hr per GPU, and rack-scale NVLink v5 from a single tray up to two or more racks. Runpod does not offer GB300.
Verda’s compliance posture is more built out. SOC 2 Type II, GDPR, EU data residency, and hardware-attested confidential computing on B200, B300 and RTX PRO 6000. The confidential variants carry roughly a 2% premium: B200 CC is $8.07 against $7.91, B300 CC $10.18 against $9.98, RTX PRO 6000 CC $2.37 against $2.33. That is a small price for attestation if a security review is what stands between you and a deployment.
Clusters
Both companies sell a product called Clusters, which makes this section easy to misread. They are different products with the same name.
Verda’s version runs up to 144 GPUs with InfiniBand, self-service through the console, API, Terraform or SkyPilot, priced per GPU: GB300 $11.48, B300 $9.98, B200 $7.91, H200 $5.49.
Runpod Clusters publish H200 SXM at {{cluster:h200-sxm:hour}}/hr and A100 SXM at {{cluster:a100-sxm:hour}}/hr per GPU. L40S, H100 SXM and B200 cluster capacity is contact-sales.
On the one card both publish at cluster scale, Verda is still cheaper: $5.49 against {{cluster:h200-sxm:hour}} on H200. Verda also publishes more of its cluster lineup self-serve. If multi-node is your starting point rather than your eventual destination, price it on both.
Spot and interruptible capacity
Verda publishes a spot rate at exactly 50% of on-demand across every card, which is the most aggressive published discount in this comparison and the single strongest reason to look at them. H200 spot is $2.75/hr. B200 spot is $3.96.
That number does not belong in the table above, because spot capacity is interruptible and on-demand capacity is not. Comparing the two directly would flatter whichever side you put in the spot column. Runpod has no interruptible tier, so there is no like-for-like counterpart to put in that column. Price the interruption risk into Verda’s spot rate before comparing it to anything.
If your training job checkpoints cleanly and can survive being preempted, half price is half price and you should run the numbers properly.
Serverless
Both companies sell scale-to-zero GPU containers, so this is not a differentiator in either direction.
Verda publishes serverless rates above its on-demand ones: B300 $10.98/hr, B200 $8.70, H200 $6.04, H100 SXM $4.63, RTX PRO 6000 $2.56, L40S $1.89. Runpod Serverless is billed per second of active worker time. Pull the current Runpod Serverless rates alongside these before deciding, because the comparison turns on billing granularity and idle behavior rather than the headline number.
Storage
Verda charges $0.20 per GiB per month across NVMe, shared filesystem and container registry alike. Runpod charges $0.10 per GB per month for container disk, and network storage runs $0.07 per GB under 1TB and $0.05 above it.
The units differ, so convert before comparing. A GiB is 1.074 GB, which makes Verda’s rate about $0.186 per GB. Against Runpod’s network storage at $0.07 that is roughly 2.7 times more, and against container disk at $0.10 it is close to double. On a storage-heavy workload that gap compounds faster than any per-hour GPU difference in the table above.
Reserved pricing
Verda publishes its commitment ladder, which is unusual and useful: 2% off at one month, 3% at three, 4% at six, 8% at a year, 25% at two years.
Read it before committing. The first six months of commitment buy almost nothing, and the meaningful discount only arrives at two years. That is a long time to fix a GPU choice in a market where the hardware turns over annually and published rates move month to month on every provider in it.
How to choose
Three questions settle most of it.
Does your work need a data center GPU at all? If no, Runpod’s sub-$1 tier has no real equivalent on the other side, and that gap is far larger than any per-hour difference on an H100.
Are you living on H200 or Blackwell? Check the rate rather than assuming. Verda was cheaper on all of those cards in August and after their October increase is cheaper on the B200 only, by eight cents an hour.
Can your workload tolerate interruption? If yes, Verda’s spot rate is the cheapest thing in this comparison, but it is capacity that can be reclaimed mid-run and Runpod sells nothing equivalent. Price the restart risk before letting that decide.
Runpod is now the cheaper published rate on seven of the nine overlapping cards, and bundles more system RAM on most of them. Verda’s case rests on spot at half price, EU data residency, confidential computing, and hardware at the top end that Runpod does not carry.
FAQ
Is DataCrunch the same as Verda?
Yes. DataCrunch rebranded to Verda in 2026 and datacrunch.io redirects to verda.com. Same company, same infrastructure, same European AI cloud. Older comparisons and reviews using the DataCrunch name still describe the same service.
Is Verda cheaper than Runpod?
Runpod, on most cards, as of October 16, 2026. Of the nine GPUs both companies publish, Runpod is cheaper on seven: B300, H200, H100 SXM, A100 SXM, L40S, RTX 6000 Ada and RTX A6000. Verda is cheaper on two, the B200 by eight cents an hour and the RTX PRO 6000 by sixteen. Runpod also has eleven cards priced below the L40S rate against Verda’s two. This reversed in October, when Verda raised every on-demand rate by between 13% and 37%, so check both pages before relying on either.
What does Verda cost?
On-demand as of October 16, 2026: GB300 $11.48/hr, B300 $9.98/hr, B200 $7.91/hr, H200 $5.49/hr, H100 SXM $4.21/hr, RTX PRO 6000 $2.33/hr, A100 SXM 80GB $2.02/hr, A100 SXM 40GB $1.47/hr, L40S $1.72/hr, RTX 6000 Ada $1.32/hr, RTX A6000 $0.74/hr, Tesla V100 $0.22/hr. Spot is 50% of on-demand on every card. Storage is $0.20 per GiB per month. Reserved discounts run from 2% at one month to 25% at two years.
Does Verda offer AMD GPUs?
No. Verda’s lineup is NVIDIA only. Runpod lists AMD MI300X with 192GB at {{gpu:mi300x}}/hr on Secure Cloud.
Which is better for a small fine-tune or a hobby project?
Runpod, by a wide margin. A 7B fine-tune or a Stable Diffusion workload runs on a 24GB card. Runpod has the RTX A5000, the RTX 3090 and the RTX 4090 in that class. Verda’s only comparable option is a 2017-era V100 with 16GB.
Which is better for a large training run?
Runpod is cheaper per hour on the B300 and H200 since Verda’s October increase; Verda holds the B200 by eight cents. Verda spot at half of on-demand is the cheaper route if your job checkpoints cleanly. Runpod bundles more system RAM on those same cards, 276GB against 170GB on H200, which matters if you stage data in host memory or run CPU offload. Price both against your actual configuration rather than the headline rate.
Does Verda have serverless GPUs?
Yes. Verda sells auto-scaling containers with scale-to-zero, priced above its on-demand instances: H200 $6.04/hr, H100 SXM $4.63/hr, L40S $1.89/hr. Runpod Serverless bills per second of active worker time. Both scale to zero, so compare on billing granularity and idle behavior rather than on whether the feature exists.
