CoreWeave went public, signed multi-billion-dollar capacity deals and builds Kubernetes-native GPU infrastructure for the largest AI labs and enterprises. Its on-demand catalogue is priced per 8-GPU node, its committed-use discounts run to 60%, and its published H100 and B200 rates sit closer to AWS than to RunPod.
packet.ai is a single-card-up cloud. You can rent one RTX 4090 for $0.39/hr or one B200 for $3.75/hr, no contract. This comparison is for teams who have outgrown a laptop GPU, are getting CoreWeave's name in vendor shortlists, and want to know whether that scale is something they actually need yet.
Platform overview: packet.ai vs CoreWeave
What CoreWeave offers
CoreWeave's published on-demand rates in August 2026, normalised per GPU from 8-GPU instances: HGX H100 $6.16/hr (spot $2.46), HGX H200 $6.31/hr (spot $2.62), HGX B200 $8.60/hr ($68.80 for the node, spot $4.26), A100 80GB SXM $2.70/hr, GB200 NVL72 $10.50/hr. RTX Pro 6000 is listed at $2.50/hr on reserved terms. Egress is free.
- Kubernetes-native (CKS), managed Slurm on Kubernetes (SUNK), Tensorizer, high-performance object and file storage
- Committed-use discounts up to 60%; most revenue is multi-year contracts
- Spot capacity at roughly 40% of on-demand for interruptible work
- 8-GPU nodes are the default unit; single-GPU on-demand is limited
What packet.ai delivers
CoreWeave is built for the multi-year, 8-GPU-node customer: reserved GB200 NVL72 racks, CKS Kubernetes, SUNK-managed Slurm, and committed-use discounts up to 60% for teams that sign long contracts. That is a real, well-built platform for a specific buyer. It is not built for a team that wants one GPU today without a sales call.
packet.ai is exactly that: single-card on-demand, published rates, no contract. On the card both platforms sell at comparable scale, packet.ai's A100 80GB Dedicated is $1.43/hr against CoreWeave's $2.70/hr on-demand SXM rate, a 47% gap. The B200 comparison is starker: packet.ai Dynamic at $3.75/hr against CoreWeave's $8.60/hr per-GPU on-demand rate, a 56% gap, even before CoreWeave's spot pricing which still requires accepting interruption.
- Dynamic: scheduler-enforced multi-tenant, single card, hourly. RTX 6000 Pro at $0.66/hr, B200 at $3.75/hr. No node minimum, no multi-year commitment.
- Dedicated: a whole card, single-tenant, 99% SLA. RTX 4090 at $0.39/hr, L40S at $0.92/hr, A100 80GB at $1.43/hr, B200 at $6.99/hr.
What CoreWeave has that packet.ai does not: GB200 NVL72 racks, Kubernetes-native infrastructure built for the platform, and committed-use pricing that beats packet.ai's rate card at true multi-year, multi-thousand-GPU scale. If you are standing up infrastructure for a frontier training run with a contract to match, CoreWeave is the more complete platform. If you need one GPU running in minutes at a published rate, packet.ai is built for exactly that.
Billing is hourly with monthly commits up to 20% off. There are no platform fees and no ingress charges; egress is $0.04/GB. Up to 2 TB of local NVMe per node is included. Dynamic instances are SSH-ready in under five minutes, Dedicated in five to ten. Capacity is live in the US (California, Virginia, Texas, Oregon) and Europe (Frankfurt, Amsterdam, Paris, London, Dublin). H100 SXM, H200 and RTX 5090 are on the notify list.
Against CoreWeave, packet.ai is simply a different-sized product: one GPU instead of eight, an hour instead of a year, a rate card instead of a quote. On the two SKUs both publish, the A100 80GB and the B200, packet.ai's on-demand hour is roughly half of CoreWeave's, and Dynamic B200 is below CoreWeave's spot price.
packet.ai vs CoreWeave at a glance
Published starting rates in USD per GPU-hour, on-demand unless noted. Percentages compare the first price in each cell.
| Category | packet.aiYOU | CoreWeave |
|---|---|---|
| A100 80GB | $1.43/hr Dedicated−47% | $2.70/hr (SXM, in node) |
| B200 (180–192GB) | $3.75/hr Dynamic · $6.99/hr Dedicated−56% | $8.60/hr on-demand per GPU (8× node) · $4.26/hr spot |
| RTX 6000 Pro (96GB) | $0.66/hr Dynamic−74% | $2.50/hr reserved only |
| L40S (48GB) | $0.92/hr Dedicated | L40 from ~$1.25/hr per GPU |
| RTX 4090 (24GB) | $0.39/hr Dedicated | Not offered |
| H100 / H200 | Launching soon (notify list) | H100 $6.16/hr · H200 $6.31/hr per GPU (8× nodes) |
| Smallest unit | 1 GPU, hourly | Typically 8-GPU instance |
| Commitment model | On-demand; monthly commits up to 20% off; clusters ~30% below retail on term | Committed-use discounts up to 60%, multi-year contracts common |
| Orchestration | SSH instances, API, packet CLI; bring your own K8s or Slurm | Managed Kubernetes (CKS), SUNK Slurm, Tensorizer |
| Data transfer | No ingress fee; egress $0.04/GB | Free egress |
| Time to capacity | Under 5 min (Dynamic); clusters 2–6 weeks | Minutes for on-demand nodes when available; contracts for scale |
Detailed comparison
Pricing: two different units
CoreWeave publishes prices for 8-GPU HGX instances. Divide by eight and the B200 is $8.60/hr per GPU; packet.ai sells the same card at $3.75/hr Dynamic and $6.99/hr Dedicated. Even CoreWeave's spot B200 at $4.26/hr is above packet.ai's non-interruptible Dynamic rate. The A100 80GB is $2.70/hr on CoreWeave against $1.43/hr on packet.ai.
Those gaps narrow sharply under a CoreWeave contract. A 60% committed-use discount takes the B200 to roughly $3.44/hr, in the same range as packet.ai's own 12-month cluster pricing of $2.80–3.20/GPU-hr. In other words: at 64+ nodes on multi-year terms, the two converge. Below that, packet.ai is a fraction of the on-demand cost.
Platform: managed Kubernetes versus a GPU with SSH
CoreWeave's real product is the platform around the GPUs: Kubernetes with GPU-aware scheduling, Slurm on Kubernetes for HPC-style batch, Tensorizer for fast model loading, and storage tuned for training I/O. If you have a platform team that thinks in Helm charts, that is a serious advantage.
packet.ai hands you a GPU instance with NVMe and a CUDA stack and gets out of the way. You can run k3s, Ray or Slurm on it; nothing is managed for you. For a two-person team shipping an inference endpoint, that is a feature. For a fifty-person ML platform group it may be a gap.
Who each one is built for
CoreWeave's customers are labs and enterprises buying thousands of GPUs on multi-year commitments, with the compliance, networking and support expectations that come with that. Its on-demand catalogue exists, but it is not the centre of the business.
packet.ai's customers are developers and teams who want one to a few dozen GPUs now, hourly, at a rate that leaves budget for the actual work, with a path to bare metal and clusters as they grow. Different buyer, different product, and the pricing reflects it.
Who should choose which
Choose packet.ai if
- You need 1 to 64 GPUs on-demand, hourly, today - no contract, no sales call
- You want the lowest published B200 or A100 rate without a multi-year commit
- Your workload fits RTX 6000 Pro, L40S or RTX 4090 - cards CoreWeave does not serve on-demand
- You want SSH access to a GPU, not a Kubernetes cluster to operate and staff
Choose CoreWeave if
- You are buying hundreds or thousands of GPUs on multi-year reserved terms
- You want managed Kubernetes, Slurm and enterprise support built in
- You need H100 or H200 at scale with InfiniBand fabric available today
- Your procurement process expects an enterprise vendor with public-company financials
Conclusion
CoreWeave is where you go when you already know you need a thousand GPUs and a platform team to run them. packet.ai is where you go when you need one, or eight, or sixty-four, and want to pay for the hour rather than the year. On the same B200 the on-demand price difference is more than 2×, and it only closes once CoreWeave's multi-year discounts kick in.
Same silicon. Smarter economics.
Every rate on this page is a published starting price you can deploy against today. Dynamic launches in under five minutes; no credit card to start.
Additional resources
- packet.ai GPU pricing: the full Dynamic, Dedicated and Clusters rate card.
- packet.ai documentation: API, CLI and platform guides.
- How intelligent scheduling works: why Dynamic keeps peak performance at a lower price.
- Clusters and wholesale quotes: multi-node InfiniBand capacity at ~30% below retail.
- Rent GPUs on packet.ai: specs, pricing and workload fit for every card.