You're looking at a CoreWeave quote or a Lambda contract, the H100 rate looks fine on the pricing page, and then someone mentions the minimum commitment and the conversation changes. That's the moment most people start searching for CoreWeave competitors or a Lambda alternative — not because the hardware is bad, but because the deal shape doesn't fit a team that wants to rent by the hour and not sign anything.
TL;DR: CoreWeave and Lambda are real enterprise-grade neoclouds, not the expensive option by accident — CoreWeave lists HGX H100 at $6.16/hr and Lambda's 8x H100 SXM instance runs $3.99/hr per GPU (both as of August 2026), and both are built around fleet-scale and contract customers. Cheaper on-demand H100 capacity exists at Crusoe ($3.90/hr), Nebius ($3.85/hr), Together AI ($3.99/hr), and RunPod ($2.99/hr), and Voltage Park starts H100 rentals at $1.99/hr with no contract. The real decision isn't just price — it's whether you want reserved capacity, a marketplace bid, or a serverless rate, and whether your environment is portable if you switch.
Why people search for CoreWeave competitors
CoreWeave and Lambda both do what they say on the label. CoreWeave built one of the largest H100/H200/GB200 fleets anywhere with Kubernetes-native infrastructure for distributed training at scale, and it has the enterprise contracting and compliance apparatus to match. Lambda owns its own datacenters, which buys consistency a broker can't promise, and its 1-Click Clusters give you InfiniBand-connected multi-node training without stitching it together yourself. Neither one is a bad product for the customer it's built for. See our full breakdowns on Aquanode vs CoreWeave and Aquanode vs Lambda if that head-to-head is what you need.
The friction shows up for anyone smaller: a researcher, a small lab, an indie team that wants an H100 for a weekend or a month, not a fleet commitment. This post is a survey for that price-shopper — across neoclouds, GPU marketplaces, and hyperscalers, with real numbers and honest commitment terms.
The alternatives, with real pricing
Every number below comes from the vendor's own pricing page, checked in August 2026. Where a vendor doesn't publish a fixed on-demand rate, that's stated instead of guessed.
| Tier | Provider | GPU types | On-demand price (per GPU/hr) | Commitment terms | Best for |
|---|---|---|---|---|---|
| Enterprise neocloud | CoreWeave | H100, H200, B200, GB200 NVL72 | $6.16 (H100), $6.31 (H200), $8.60 (B200) | On-demand listed; reserved fleet contracts for larger deals | Fleet-scale distributed training with enterprise compliance needs |
| Enterprise neocloud | Lambda | H100, A100 | $3.99 (8x H100 SXM), $2.79 (8x A100) | On-demand instances; separate 1-Click Clusters product for reserved multi-node | Consistent, self-hosted hardware and multi-node InfiniBand training |
| Neocloud | Crusoe | H100, H200 | $3.90 (H100 HGX), $4.29 (H200 HGX) | Hourly on-demand; "contact us" for reserved/guaranteed capacity | Mid-market teams wanting a fixed hourly rate without a contract |
| Neocloud | Nebius | H100, H200, B200 | $3.85 (H100), $4.50 (H200), $7.15 (B200) | On-demand published; reserved multi-month contracts save up to 35% (sales-negotiated) | Teams that want an on-demand rate today with a path to reserved pricing later |
| Neocloud | Together AI | H100, H200, B200 | $3.99 (H100), $5.99 (H200), $8.19 (B200) | On-demand clusters, no long-term contract required | Fast on-demand clusters for training and fine-tuning |
| Serverless/on-demand | Modal | A100, H100, H200, B200 | $2.50 (A100 80GB), $3.95 (H100), $4.54 (H200), $6.25 (B200) | Pay-per-second, no commitment | Bursty inference and batch jobs billed by the second |
| Marketplace | Vast.ai | Varies by host | No fixed rate — host-set bids | None; spot-style marketplace | Cheapest possible price if you can tolerate host variance |
| Marketplace | RunPod | H100, A100 | $2.99 (H100 SXM), $1.49 (A100 SXM) | On-demand and spot-style community pricing | Cheap, fast single-GPU rental with a large template library |
| Hyperscaler (reserved) | AWS EC2 P5 | H100 | $5.19/GPU (Capacity Blocks, reserved) | Reserved capacity block, not on-demand — booked in advance for a fixed window | Teams already on AWS who need a guaranteed reservation window |
| Portable multi-provider | Aquanode | 10+ providers, A100 through B300 | Live rates on /pricing and /marketplace | On-demand per-second billing, no contract | Renting across providers without rebuilding your environment each time |
A few of those rows need the fine print spelled out.
Voltage Park advertises H100 rentals "starting at $1.99 without a contract," with Ethernet and InfiniBand tiers and larger clusters listed as "contact for pricing." Treat $1.99 as a floor for single-GPU access, not a number that holds at cluster scale.
AWS Capacity Blocks are not on-demand pricing — you're reserving a block of time in advance, and $5.19 per GPU on a p5.48xlarge (8x H100) is the reservation rate, not the hourly on-demand EC2 rate, which runs considerably higher.
Vast.ai is a host-bid marketplace with no fixed rate anywhere — that's not a research gap, it's the model. Bids move with live supply.
Commitment terms are the part people skip
The price-per-hour comparison is the easy half. The part that actually decides whether CoreWeave or Lambda is the wrong fit for you is the commitment shape, and it splits into three buckets:
- Pure on-demand, no contract. Crusoe, Nebius (on-demand tier), Together AI, Modal, RunPod, Voltage Park's single-GPU rate, and Aquanode all let you rent by the hour and walk away. This is the bucket to be in if you don't know yet how long you'll need the GPU.
- Reserved / contracted for the discount. CoreWeave and Lambda both publish on-demand rates, but the economics of both companies run on larger reserved deals — CoreWeave's fleet contracts and Lambda's 1-Click Clusters are where the serious volume and the better effective rate live. AWS Capacity Blocks are reserved by definition: you book a window, you don't rent by the hour as it happens.
- Bid-based marketplace. Vast.ai has no commitment concept at all — you take whatever rate a host is currently offering, which can be excellent or can vanish depending on demand.
None of these is universally right. A team running a six-month pretraining job should be talking to CoreWeave or Lambda about a contract — that's the deal shape that gets you fleet-scale hardware and support. A team renting single boxes for a few weeks at a time is paying for infrastructure it doesn't need if it signs the same kind of contract.
What actually changes if you switch off CoreWeave or Lambda
The quiet cost of picking a cheaper on-demand neocloud isn't the GPU rate — it's that your environment doesn't travel with you. Every provider in the table above, CoreWeave and Lambda included, gives you a box. What you install on that box — the CUDA stack, the Python environment, the model weights, the config you spent an evening tuning — stays on that box unless you build your own way to move it.
That's the gap Aquanode is built to close. We rent GPUs across more than ten providers, so if the CoreWeave or Lambda price doesn't fit and you move to a cheaper on-demand tier, you're not starting from a blank machine. If you stop a deployment yourself, we capture its state on the way out and you can restore it later on a different provider — same environment, different hardware underneath. That's claimable because you triggered it. If a provider kills the box out from under you instead, you only get back to whatever automated snapshot you'd already turned on — snapshots are opt-in, not scheduled by default, and there's no free recovery from a provider-side termination with nothing configured. You can turn scheduling on yourself with an interval as short as 15 minutes if you want that safety net.
Pause and resume work the same way across every provider we support: snapshot and terminate, then redeploy and restore. Compare current rates side by side at the live GPU availability index, or start from /compare to line up specs.
Frequently asked questions
What are the best CoreWeave competitors?
For enterprise-scale reserved capacity, Lambda and AWS EC2 (via Capacity Blocks) are the closest comparisons to CoreWeave. For on-demand pricing without a fleet contract, Crusoe ($3.90/hr H100), Nebius ($3.85/hr H100), and Together AI ($3.99/hr H100) all publish lower on-demand H100 rates than CoreWeave's $6.16/hr, as of August 2026. Marketplace options like RunPod ($2.99/hr H100) and Voltage Park (from $1.99/hr) go lower still if you don't need CoreWeave's fleet scale.
Is Lambda more expensive than other GPU clouds?
Yes, on a straight per-GPU-hour basis. Lambda's 8x H100 SXM instance runs $3.99/hr per GPU, which is close to Together AI's on-demand H100 rate but higher than Nebius, Crusoe, or RunPod. What you're paying for is Lambda owning its own datacenters and offering 1-Click Clusters for InfiniBand multi-node training, which several cheaper alternatives don't offer at all.
Is Vast.ai cheaper than CoreWeave?
Usually, but there's no fixed number to compare — Vast.ai is a host-bid marketplace with prices that move with live supply rather than a published rate. It can undercut every neocloud in this list on a good day, and the trade is variable host reliability instead of a guaranteed fleet.
Do CoreWeave or Lambda require a minimum contract?
Not for their on-demand instance tiers, which are what the prices in this post cite. Both companies' larger economics run on reserved and contracted deals — CoreWeave's fleet agreements and Lambda's 1-Click Clusters — which is where enterprise customers typically end up, but neither forces a contract just to rent a single on-demand instance.
What's the cheapest way to rent an H100 right now?
Among vendors with a published, verifiable on-demand rate as of August 2026, Voltage Park's advertised $1.99/hr floor and RunPod's $2.99/hr are the lowest in this survey. Vast.ai's marketplace bids can go lower on any given day but aren't a fixed, checkable number. Check the live GPU availability index for current rates across providers, since these move.
The takeaway
CoreWeave and Lambda earn their price for fleet-scale hardware, Kubernetes-native infrastructure, and datacenters they own outright — that's a real answer for a real customer. If you're not that customer, Crusoe, Nebius, Together AI, RunPod, and Voltage Park all publish lower on-demand rates, and AWS Capacity Blocks are worth checking if you're already on AWS and can commit to a reservation window. Whichever one you land on, the part worth planning for up front is what happens to your environment when you eventually move again — see our deeper piece on renting H100s under $2 for the pricing side, or start comparing current rates at /marketplace.
