Crusoe started by putting GPUs next to flared natural gas and has grown into one of the largest neoclouds by contracted power, with multi-gigawatt data centre projects. It sells H100, H200, A100 and AMD MI300X on demand, with Blackwell available through sales, and pitches itself on cost and sustainability at scale.
packet.ai does not build power plants. It squeezes more useful work out of each GPU through scheduling and sells the result hourly. This comparison looks at the on-demand rate cards, which overlap on the A100 80GB, and at the very different ways each company gets to a low number.
Platform overview: packet.ai vs Crusoe
What Crusoe offers
Crusoe's published on-demand per-GPU rates in August 2026: HGX H100 $3.90/hr, HGX H200 $4.29/hr, A100 80GB SXM $2.30/hr, AMD MI300X $3.45/hr, L40S about $1.50/hr. B200, GB200 NVL72 and MI355X are quoted by sales. Egress is free and committed-use discounts are sales-quoted.
- Energy-first infrastructure: data centres sited at stranded or renewable power
- Managed Kubernetes and Slurm, InfiniBand clusters, enterprise support
- Blackwell on-demand pricing not published; contact sales
- No consumer or workstation GPUs
What packet.ai delivers
Crusoe's cost story starts upstream, at the power source. packet.ai's starts at the GPU itself: the scheduler keeps a card doing useful work a higher share of the time, and that recovered utilisation is passed back as a lower published rate. On the one card both list, the A100 80GB, that shows up directly: $1.43/hr Dedicated on packet.ai against $2.30/hr on Crusoe, a 38% gap.
The bigger difference is what happens above Ampere. Crusoe's B200 is not on its rate card at all, it is quoted through sales. packet.ai publishes a B200 rate you can deploy against today.
- Dynamic: scheduler-enforced multi-tenant, published rate, no sales call. B200 starting at $3.75/hr, RTX 6000 Pro at $0.66/hr, both of which Crusoe does not list on-demand.
- 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 Crusoe has that packet.ai does not: published Hopper rates (H100 $3.90/hr, H200 $4.29/hr) that packet.ai's notify-list H100/H200 cannot yet match, AMD MI300X, and a genuine energy-sourcing story for buyers with sustainability requirements written into procurement. If you need Hopper today or a contracted power narrative, Crusoe has both. If you need Blackwell on demand without a sales conversation, packet.ai does.
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 Crusoe, packet.ai has a published B200 price where Crusoe has a sales form, and on the one card both publish, the A100 80GB, packet.ai is 38% cheaper at $1.43/hr versus $2.30/hr. Below the A100, Crusoe has only the L40S; packet.ai has L40S, RTX 6000 Pro and RTX 4090.
packet.ai vs Crusoe 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 | Crusoe |
|---|---|---|
| A100 80GB | $1.43/hr Dedicated−38% | $2.30/hr SXM |
| L40S (48GB) | $0.92/hr Dedicated | ~$1.50/hr |
| B200 (180–192GB) | $3.75/hr Dynamic · $6.99/hr Dedicated | Contact sales |
| RTX 6000 Pro (96GB) | $0.66/hr Dynamic | Not offered |
| RTX 4090 (24GB) | $0.39/hr Dedicated | Not offered |
| H100 / H200 | Launching soon (notify list) | H100 $3.90/hr · H200 $4.29/hr |
| AMD | Not offered | MI300X $3.45/hr; MI355X on request |
| Commitment model | On-demand; monthly up to 20% off; clusters ~30% below retail on term | On-demand for Hopper/Ampere; Blackwell and discounts sales-quoted |
| Orchestration | SSH instances, API, packet CLI | Managed Kubernetes, Slurm, bare metal |
| Data transfer | No ingress fee; egress $0.04/GB | Free egress |
| Energy story | Utilisation: more work per GPU-hour via scheduling | Siting: data centres at stranded and renewable power |
Detailed comparison
Two routes to a low price
Crusoe's cost advantage is upstream: cheaper power, often power that would otherwise be wasted, in data centres it builds itself. That shows up as competitive Hopper pricing and a strong story for buyers with sustainability targets.
packet.ai's cost advantage is downstream: its scheduler places multiple isolated workloads on a card so that the GPU is doing useful work a far higher share of the time, and the recovered utilisation is passed back as a lower hourly rate on Dynamic. Same peak performance, same VRAM, less waste. Different mechanism, same effect on your invoice, and arguably a more direct form of efficiency.
On-demand rate cards
The overlap is the A100 80GB: $1.43/hr on packet.ai Dedicated against $2.30/hr on Crusoe, a 38% gap. The L40S is $0.92/hr against about $1.50/hr. Crusoe has published Hopper rates (H100 $3.90, H200 $4.29) that packet.ai cannot match today because its H100 and H200 are still on the notify list.
On Blackwell the situation reverses: packet.ai publishes a B200 at $3.75/hr Dynamic and $6.99/hr Dedicated, and Crusoe asks you to contact sales. For a developer who wants one B200 this afternoon, that is the whole story.
Scale and platform
Crusoe is built for large, long commitments: managed Kubernetes and Slurm, InfiniBand clusters, enterprise support and multi-gigawatt build-outs. If you are placing a 1,000-GPU order with a sustainability clause in the contract, Crusoe is a natural shortlist entry.
packet.ai's clusters go to 1,024+ GPUs at wholesale rates roughly 30% below retail, quoted and provisioned over two to six weeks. For 8 to 64 GPUs on a 6–12 month term it competes well; for gigawatt-scale campus deals it is not the counterparty.
Who should choose which
Choose packet.ai if
- You need a B200 on-demand without an enterprise agreement
- You want the cheapest Ampere card on the market (A100 80GB at $1.43/hr, L40S at $0.92/hr)
- You want a single GPU, hourly, with no minimum node size
- A simpler GPU-first platform is more valuable than a full managed cloud stack
Choose Crusoe if
- You need an H100 or H200 at a published price today
- Sustainability provenance (stranded energy, carbon targets) is part of your procurement criteria
- Your workload needs a managed cloud layer: object storage, Kubernetes, and managed services
- You are running at enterprise or research scale and want a reserved agreement
Conclusion
Crusoe is a serious infrastructure company with a real energy advantage and a full Hopper lineup. packet.ai is cheaper on the one card they both publish, has a published Blackwell price where Crusoe has a form, and covers everything below the A100. For Hopper at scale, Crusoe; for on-demand Blackwell and Ampere, packet.ai.
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.