CoreWeave Competitors
and Alternatives: Lambda, Nebius, Runpod and the GPU Clouds Compared (2026)
Published prices for every major GPU cloud, and what each is best for. When you’re ready, we compare them against your requirement and negotiate the terms. The provider pays us, so it costs you nothing.
Price My GPU RequirementUpdated September 2026
The main CoreWeave competitors are the other specialist GPU clouds, often called neoclouds: Lambda, Nebius, Crusoe, Together AI, Fluidstack and Voltage Park. Buyers also compare the hyperscalers (AWS, Microsoft Azure and Google Cloud), dedicated GPU server providers such as Latitude.sh, and lower-cost marketplaces such as Runpod and Vast.ai. The right alternative depends on the size of the deployment, how long you need it, which GPU, and where.
GPU cloud prices are rising
Sources: CoreWeave's second-quarter 2026 earnings call and Inc.; Lambda via UsagePricing; Nebius via its customer notice, as reported.
In that market, the rate card is the worst price you can pay. Where you buy and how the contract is negotiated matter more than they did a year ago. That is the work we do: comparing the providers that can deliver your requirement, negotiating price and terms, and staying on the account after you sign.
What an H100 Costs at Each GPU Cloud
Published on-demand price per GPU-hour, September 2026. Reserved and negotiated pricing is typically lower.
Source: Metro Colo Advisory compilation of published rates, September 2026, from the sources listed under the table below. Free to cite with a link to this page.
CoreWeave Competitors Compared
| Provider | Best for | Published pricing | Watch for |
|---|---|---|---|
| CoreWeaveNeocloud | The largest training clusters and enterprise AI | About $6.16 per GPU for an eight-GPU H100 node on demand; raised about 25% in July 2026 | Near-term capacity described as sold out; best pricing needs large commitments |
| LambdaNeocloud | Clusters from 16 to 2,000+ GPUs and self-serve instances | H100 $3.99 on demand, up from $2.99; B200 from $6.69 | On-demand stock is first come; cluster reservations need approval |
| NebiusNeocloud | Training and inference with a full managed platform | H100 $4.50 from October 1, 2026, up from $3.85 | Two price increases in about three months |
| CrusoeNeocloud | Energy-first data centers, training and inference | H100 about $3.90 on demand; B200 and GB200 by quote | Newest GPUs only through sales |
| Together AINeocloud and model platform | Teams that want model APIs and GPU clusters from one provider | Clusters by quote | Best value when you use the platform, not just the GPUs |
| FluidstackNeocloud | Dedicated clusters for AI labs and large enterprises | Not published | No longer rents individual GPU instances |
| Voltage ParkNeocloud | H100 capacity at a published low price | H100 from $1.99 | Narrower GPU range |
| Latitude.shDedicated GPU servers (Megaport) | Inference fleets across many metros, bare metal | Eight-GPU B300 server from $64 an hour, about $8 per GPU-hour | Cluster networking over Ethernet rather than InfiniBand |
| RunpodMarketplace and cloud | Small teams, experiments, per-second billing | H100 from about $1.99 on the community tier | Community capacity varies by host |
| Vast.aiMarketplace | The lowest prices for interruptible work | H100 often under $1.50 | Interruptible capacity and varying hosts |
| AWS, Azure, Google CloudHyperscalers | Teams already built on one cloud | H100 about $6.88 on AWS and up to $6.98 on Azure, on demand before discounts | Highest list prices; AWS raised H200 prices 15% in January 2026 |
Sources, September 2026 unless noted: CoreWeave node rate via Spheron, July 2026, and price increase from its earnings call; Lambda via UsagePricing; Nebius via its customer notice, as reported; Crusoe via Spheron, July 2026; Fluidstack via Cloud GPUs; Runpod via GPU Cloud Cost; Voltage Park and Latitude.sh from their published price lists; marketplace and hyperscaler rates via CloudZero, August 2026; AWS's H200 increase via Cast AI.
Send the GPU, how many, for how long and where. You'll hear back within 24 hours from the person who will run your search. No signup, no call needed.
Why Compare GPU Clouds Through Us Instead of Going Direct
Going direct gets you a rate card, a sales call and terms you negotiate alone, one provider at a time. Here is the difference.
| Going direct | Through us | |
|---|---|---|
| Price | The provider's rate card | Negotiated against what the market actually pays |
| Cost to you | Nothing extra | Nothing extra; the provider pays us |
| Options compared | Only the ones you find and call | The options that fit your requirement, compared for you |
| Contract terms | Negotiated alone, against a deal team | Negotiated by someone who knows what providers accept |
| Time to a reply | Weeks of calls | A reply within 24 hours from the person doing the work |
| After you sign | An account rep who works for the provider | Someone on your side through delivery and renewal |
1. Rising prices make the negotiation worth more
When list prices are climbing, the rate card is the worst price you can pay, not a fixed one. We know what comparable deployments actually sign for, and we negotiate against the market rather than accepting the first number. Where more than one provider can serve a requirement, they compete for it.
2. You do not pay more for it
The provider you choose pays us through its channel program, from the budget it would otherwise spend on its own sales team. Going direct does not get you a lower price; it only means nobody is comparing for you.
3. The terms matter as much as the rate
Delivery dates with remedies if they slip, the right to move to newer GPUs during the term, network and storage performance in writing, bandwidth charges and exit rights. With providers citing sold-out capacity, a contract without a committed delivery date is a price for capacity you may not get. Our GPU rental guide covers each term in detail.
4. We know where capacity is, not just what is listed
Published availability and delivered availability are different things. We take live AI requirements to providers and operators across North America, and Data Center Knowledge led its August 2026 reporting on usable AI capacity with our findings.
Why buyers are looking for CoreWeave alternatives in 2026
CoreWeave is the largest of the specialist GPU clouds and, for the biggest training clusters, one of the strongest. The reasons buyers look elsewhere are mostly about access and price, not quality. Its July increase of about 25 percent across its lineup lands mainly in negotiated contracts and renewals, and management has described near-term capacity as effectively sold out, with demand from several customers for each GPU it brings online. Its best pricing also goes to large, long commitments.
For a team that needs a few dozen GPUs next month, or a steady inference fleet rather than a training supercluster, that combination points to CoreWeave competitors instead. The practical question is not which provider is best in general, but which one can deliver your requirement, on your timeline, at a price that holds up.
The Main CoreWeave Competitors, One by One
Lambda
Lambda sells self-serve GPU instances from one to eight GPUs and 1-Click Clusters of 16 to 2,000+ interconnected H100 or B200 GPUs, with no egress fees, according to Contrary Research and UsagePricing. It is the closest like-for-like CoreWeave alternative for teams that want clusters without an enterprise-scale commitment. Its on-demand prices rose in 2026, and on-demand stock is first come, first served, so reservations matter for anything steady.
Nebius
Nebius offers GPUs from the L40S to the GB300, a managed platform around them, and published hourly rates. It is a strong fit for teams that want the platform as well as the GPUs. From October 1, 2026, its H100 rate is $4.50 a GPU-hour, up from $3.85, and its B300 rate is $9.50, up from $7.85, according to its customer notice, as reported, so committed pricing is worth negotiating before renewals.
Crusoe
Crusoe builds and runs its own energy-first AI data centers and sells GPU capacity and managed inference on them. It publishes on-demand rates for the H100, H200 and A100, about $3.90 for an H100, while B200 and GB200 capacity is by quote.
Together AI
Together AI combines a large model API platform with GPU clusters for training and inference. It suits teams that want to run open models through an API and rent their own clusters from the same provider. For pure GPU rental, compare its cluster pricing against the neoclouds above.
Fluidstack
Fluidstack has shifted to building large dedicated clusters for a small number of major customers, including Anthropic's multi-site buildout in Texas and New York, and no longer rents individual GPU instances. It is a CoreWeave competitor at the very top of the market, not an option for a few servers. An investor memo reported by Forbes projects it will manage up to 1.3 gigawatts of power this year.
Voltage Park
Voltage Park runs H100s across six US data centers and publishes H100 pricing from $1.99 an hour. It suits teams that want H100 capacity at a published low price and do not need the newest generation.
Latitude.sh
Latitude.sh, owned by the network provider Megaport, rents dedicated GPU and CPU servers on demand in 26 locations worldwide, with published pricing and discounts for monthly and yearly terms. Megaport has raised capital to build an on-demand GPU pool for AI inference, including 2,048 NVIDIA B300 GPUs coming online this fall. It fits inference fleets that need servers in several metros, with private connections to the major clouds.
Runpod and Vast.ai
Runpod and Vast.ai serve the lower-cost end of the market. Runpod bills by the second and offers both community and secure capacity, with H100s from about $1.99 an hour on its community tier. Vast.ai is a marketplace that matches buyers with independent hosts, which produces the lowest prices, often under $1.50 for an H100, in exchange for interruptible capacity and more variation between hosts. Both suit experiments and short jobs better than production inference.
The hyperscalers
AWS, Microsoft Azure and Google Cloud publish the highest GPU list prices, about $6.88 an H100 GPU-hour on AWS and up to $6.98 on Azure on demand, according to CloudZero, and AWS raised H200 prices by 15 percent in January 2026. What they offer is integration with everything else you run there. Many AI companies keep the application on a hyperscaler and rent GPUs from a neocloud or dedicated provider, joined by a private connection, which the data center connectivity guide explains.
Alternatives to Nebius, Lambda, Runpod and Modal
The same comparison works from any starting point. Among Nebius competitors, the closest Nebius alternatives are Lambda, Crusoe and CoreWeave, and Latitude.sh for inference fleets across several regions. Among Lambda competitors, the strongest Lambda alternatives are CoreWeave and Nebius at cluster scale, and Voltage Park for H100 capacity at a published low price. Among Runpod competitors, the best Runpod alternatives are Vast.ai for the lowest prices, and Lambda or a dedicated server provider once a workload needs to run steadily in production. For Together AI alternatives, the neoclouds for clusters, paired with a separate model API if you need one. For Modal alternatives, Runpod for per-second serverless capacity, or a dedicated GPU server provider once a workload runs steadily enough that paying by the second costs more than a monthly server.
CoreWeave vs Lambda
Both run large clusters on NVIDIA’s current generation. CoreWeave is built around the largest enterprise and AI lab commitments; Lambda makes clusters easier to reach at smaller scale, with published cluster pricing and shorter reservations. For a few hundred GPUs or fewer, Lambda is usually easier to get; for the largest training runs, CoreWeave is the more likely home.
Runpod vs Vast.ai
Runpod is a managed cloud with a secure tier and per-second billing; Vast.ai is an open marketplace with the lowest prices and more variation between hosts. Runpod is the safer choice for anything customer-facing; Vast.ai for batch work that can stop and restart.
How to choose among GPU clouds
Size and networking. A single server or a small inference fleet can run almost anywhere. Training across dozens of servers needs a cluster with high-speed interconnect, which narrows the field quickly.
Term. On-demand suits experiments and bursts. Anything that runs steadily for months costs much less reserved, and in a rising market a reservation also locks in today’s price.
GPU generation. The newest GPUs are the hardest to get and the most expensive. A model that fits comfortably on H100s or H200s may cost less to serve there, with more providers able to deliver it.
Location. Inference close to users cuts latency, and some teams need capacity in a specific country or region for data residency.
The contract. Delivery dates, upgrade rights, bandwidth and exit terms decide whether a low rate stays low. This is the part most buyers negotiate least.
Not sure which of these fits? Describe the workload, and we will tell you which providers can serve it and what it should cost.
Renting versus owning
At steady scale over several years, owning GPU servers and placing them in colocation can cost less than renting, with the hardware still yours at the end. The AI and GPU colocation guide covers that path, the data center cost guide covers what it takes, and a fixed-fee rent-versus-own analysis is available through FinOps consulting.
What this page cannot tell you
It cannot tell you which provider has capacity for your requirement next month, because that changes week to week. It cannot tell you what a provider will sign for, because negotiated pricing is not published. And it cannot tell you which contract terms a provider will concede, because that depends on how full it is and how much it wants the business. Those are what we find out for you.
How the practice works
Metro Colo Advisory is an independent infrastructure advisory practice working across North America. We do not own GPUs, run a cloud or operate data centers.
Placement. You give us the requirement. We find who can deliver it, benchmark the price against the market, negotiate the terms and stay on the account. The provider you choose pays us; you pay nothing. If nothing we can place fits, we tell you.
Consulting, when the question comes first. Whether to rent or own, how much capacity you need, or what a fleet costs over three years. That is FinOps consulting, at a fixed fee. For investors evaluating a GPU cloud or AI infrastructure business, see data center due diligence.
Frequently Asked Questions
Who are CoreWeave's main competitors?
The other specialist GPU clouds, Lambda, Nebius, Crusoe, Together AI, Fluidstack and Voltage Park; the hyperscalers, AWS, Microsoft Azure and Google Cloud; dedicated GPU server providers such as Latitude.sh; and marketplaces such as Runpod and Vast.ai. Fluidstack competes at the very top of the market, the marketplaces at the lower end.
What is the best CoreWeave alternative?
It depends on the requirement. For clusters of up to a few hundred GPUs, Lambda and Nebius are the closest matches. For steady inference across several locations, a dedicated GPU server provider. For the lowest price on interruptible work, a marketplace. The best alternative is the one that can deliver your GPU, count and term on your timeline, which is what we find out for you.
Did CoreWeave raise its prices?
Yes. On its second-quarter 2026 earnings call, CoreWeave said it raised prices by about 25 percent across its lineup in July, and described near-term capacity as effectively sold out. The increase shows up mainly in negotiated contracts and renewals rather than the public rate card.
Are GPU cloud prices going up in 2026?
At the major neoclouds, yes. CoreWeave raised prices about 25 percent in July, Lambda raised its on-demand H100 rate from $2.99 to $3.99, and Nebius’s H100 rate rose to $4.50 on October 1, from $3.85. Marketplace prices for older GPUs have stayed low. Rising list prices make reserved terms and negotiated pricing worth more.
Who are Nebius's competitors?
CoreWeave, Lambda and Crusoe are the closest, with Together AI for teams that also want model APIs, and dedicated GPU server providers such as Latitude.sh for inference fleets across several regions.
Runpod vs Vast.ai: which is better?
Runpod for anything customer-facing, because it offers a secure tier and a managed platform. Vast.ai for the lowest price on work that can be interrupted and restarted. Neither is designed for large production clusters.
What is a neocloud?
A cloud built mainly to rent GPUs for AI, rather than a general-purpose cloud with GPUs added. CoreWeave, Lambda, Nebius and Crusoe are the best-known neocloud providers.
Which GPU providers do you work with?
We work through channel partnerships with providers of dedicated GPU servers, fully managed GPU hosting, and GPU capacity inside managed private clouds, including options audited for regulated data. Which provider fits depends on your GPU, size, region and how much you want managed for you. If a requirement needs something our partners cannot supply, we will tell you rather than stretch the fit.
Do I pay anything to use Metro Colo Advisory?
No. The provider you choose pays us through its channel program, and you are under no obligation to use us. A formal rent-versus-own analysis for a large fleet is a separate fixed-fee engagement.
Related Reading
Renting GPUs: the GPU rental guide, bare metal, and data center connectivity for private connections between your GPUs and clouds.
Owning instead: AI and GPU colocation, direct to chip cooling, wholesale colocation and the data center cost guide.
Costs and decisions: FinOps consulting, the cloud repatriation guide, and data center due diligence for investors.
Tell Us What You Need to Run
The GPU, how many, for how long, where, and what the workload is. That's enough to start, and you'll hear back within 24 hours from the person who will run your search.
Price My GPU RequirementNorth American coverage. Independent of any single GPU provider. No cost to buyers.
