Private LLM hosting vs Shadeform
We rent DGX Sparks, so weigh that against everything below, and note upfront that Shadeform isn't quite the same kind of thing GPUwerk is. Shadeform is a GPU marketplace and orchestration layer: it lets you browse and provision GPU instances across a range of underlying cloud providers through one interface and one billing relationship, rather than running its own fleet of data centers. That's fundamentally different from a single company renting out hardware it owns and operates. A DGX Spark from GPUwerk is the latter: one company, one machine specification, one EU location, one accountable operator you're contracting with directly. Comparing the two isn't quite apples to apples, and this page tries to be honest about that rather than pretend otherwise.
Side by side
| Shadeform | Dedicated DGX Spark (GPUwerk) | |
|---|---|---|
| What it is | A marketplace and orchestration layer for provisioning GPU capacity across multiple underlying cloud providers, not Shadeform's own hardware. | A single-purpose GPU host: one dedicated DGX Spark, owned and operated by GPUwerk, rented by the hour. |
| Who operates the hardware you land on | Whichever underlying cloud provider is selected for that instance, which can differ between deployments; Shadeform coordinates the marketplace, it doesn't run the machines itself. | GPUwerk directly. The company you're paying is the company running the machine, full stop. |
| Region | Wherever the underlying provider you select operates; check Shadeform's current provider list for EU availability specifically. | EU-Central (Prague), exclusively. One location, no region selection to get wrong. |
| Who can see your data | Governed by Shadeform's own terms plus whatever access the selected underlying provider's setup allows; worth reading closely given the multi-provider model routes workloads to different operators depending on selection. | Nobody at GPUwerk. Our DPA states GPUwerk "hosts the machine but does not access, read, copy, index, or analyse the content of the controller's instance." |
| Hardware consistency | Varies by which provider and GPU type you select: model, region, network, and reliability history differ across the marketplace. | Every node is the same: 128GB unified memory, NVIDIA's DGX Spark architecture, identical specification across the fleet. |
| Pricing model | Set per underlying provider and GPU type, published on Shadeform's own marketplace pages and changing with availability. | Per hour, billed per minute: $0.79/hour on-demand, $0.59/hour for a customer-requested stop that holds your reservation. One rate regardless of workload, per pricing. |
| Contracts and DPA | Governed by Shadeform's own terms, layered on whatever the selected underlying provider's arrangement covers; check current terms closely given the multi-provider structure. | A standard GDPR Article 28 DPA published free at /legal/dpa, no negotiation required. Sub-processor list at /legal/sub-processors states none are engaged for instance workloads. |
| What you operate yourself | Everything on the provisioned instance: OS, model server, monitoring, backups, same as any bare GPU rental. | The same: OS, model server, any RAG or agent layer, monitoring, backups. GPUwerk keeps only a recovery copy of /workspace for hardware failures, refreshed roughly every six hours; it is not a backup service, so that responsibility is entirely yours. |
The worked cost example
Take a workload that runs inference continuously across a full month. Two ways to serve it:
Shadeform. Priced per underlying provider and GPU type, published on Shadeform's own marketplace pages and varying by which provider and configuration you select. A fair dollar comparison needs a specific provider and its actual rate at the time you'd rent it, so GPUwerk didn't invent a blended figure here. Check Shadeform's current marketplace rates for the GPU class you'd actually need for a real number, and factor in that the underlying operator can differ between the providers on offer.
Dedicated Spark. At $0.79/hour, running continuously for a 730-hour month costs 730 × $0.79 = $576.70, flat, at one known rate from one known, accountable operator. If you hold the reservation instead of running it, the held rate drops that to 730 × $0.59 = $430.70.
A marketplace aggregating multiple providers can undercut a fixed rate on price for a given GPU type, that's part of the appeal of shopping across providers in one place. What it structurally can't offer is a single accountable operator you're contracting with directly, since the entity actually running your hardware can change between the providers listed on the marketplace. Weigh selection and price against provenance for your own risk tolerance, these are genuinely different products.
Migration path
Both ultimately expose a Linux GPU instance, so migration is mostly re-deploying your own stack. Serve a model through vLLM on a Spark and on a Shadeform-provisioned instance, and both expose an OpenAI-compatible endpoint from vLLM itself, not from the underlying infrastructure layer. Put LiteLLM in front of either for request logging and key management. Because a Shadeform instance's underlying provider and GPU model can vary by what you select or what's available, expect more variability re-testing a model's performance there than you would moving between two Sparks, which are identical by design.
When Shadeform is the right choice
- You want to compare and provision across a wide range of GPU types and cloud providers from one interface.
- Price and selection across providers matter more than knowing exactly which company operates the hardware you land on.
- Data residency and a single accountable operator with a standard DPA aren't hard requirements.
When a dedicated Spark is the right choice
- You need to know exactly who operates the hardware, under one contract, with one company accountable.
- A consistent specification across every node matters more than marketplace breadth.
- EU data residency and a standard GDPR DPA are requirements, not nice-to-haves, and a marketplace of varying underlying providers doesn't fit your compliance posture.
FAQ
Is Shadeform a good alternative to GPUwerk for LLM hosting?
Shadeform is a GPU marketplace and orchestration layer: it lets you provision GPU instances across a number of underlying cloud providers through one interface, rather than operating its own fleet of machines. That structure gives access to a wide range of GPU types and providers, but which underlying operator you actually land on, and where that machine sits, depends on what you select and what's available at the time. GPUwerk operates its own DGX Spark fleet directly, so every node is the same known hardware in the same EU location under one company's direct control. These are different models, not a straight price comparison.
What's the difference between a DGX Spark and a Shadeform instance?
A Shadeform instance provisions capacity from one of the underlying cloud providers in its marketplace, so the actual operator, hardware, and location depend on which provider and GPU type you pick at deploy time; check Shadeform's current provider list for what's available. A DGX Spark is a standard machine GPUwerk operates directly: 128GB unified memory, NVIDIA's DGX Spark architecture, the same specification on every node in EU-Central, with GPUwerk as the sole operator you're contracting with.
Is a DGX Spark cheaper than Shadeform?
It depends heavily on which underlying provider and GPU type you'd select through Shadeform's marketplace, since pricing is set per provider and changes with availability. A dedicated Spark costs $0.79/hour flat, a single known number from a single known operator. Check Shadeform's current marketplace rates for the configuration you'd actually need before comparing, and weigh the price against the fact that the underlying operator can vary.
Why choose a DGX Spark over Shadeform?
Mainly provenance and accountability. Every Spark GPUwerk rents out is hardware we operate directly, in a data center we control in Prague, at one published rate, under one standard DPA. A Shadeform instance runs on whichever underlying cloud provider you select or Shadeform's orchestration assigns, which means the entity actually operating your hardware, and its own data handling practices, can differ between deployments. If a specific, known, EU-hosted machine under one accountable operator matters more than a marketplace's breadth of providers and GPU types, the Spark fits better.
GPUwerk did not find a single Shadeform price that fairly represents its multi-provider marketplace; check Shadeform's own site for current rates. GPUwerk's own figures ($0.79/hour, $0.59/hour, $576.70/month) come from our published pricing.