Vultr is an independent general-purpose cloud with a large global footprint and a straightforward developer console, and it has built out a GPU catalogue on top: NVIDIA A100, L40S, A40, GH200, H100, B200 and AMD MI300X, sold as Cloud GPU instances or bare metal. For a team already on Vultr for compute and storage, adding GPUs in the same account is the easy path.
packet.ai is only GPUs. It does not have 32 regions or a general VM catalogue. On the cards the two share, it is markedly cheaper. This comparison is about whether GPU-in-a-general-cloud convenience is worth the difference.
Platform overview: packet.ai vs Vultr
What Vultr offers
Vultr's published GPU rates verified in late August 2026: A100 PCIe 80GB $2.397/hr, L40S $1.671/hr, A40 $1.712/hr, GH200 $1.99/hr per GPU, and B200 around $8.50/hr per GPU normalised from its 8× bare-metal node. H100 and AMD MI300X are available in larger shapes; many high-end configurations are reserved-contract rather than genuine on-demand.
- 32 regions across six continents; GPUs available in a subset of them
- General cloud services (VMs, Kubernetes, object and block storage, load balancers) alongside GPUs
- Hourly billing; bandwidth allowances with overage at $0.01/GB
- High-end GPUs often sold as reserved bare metal rather than on-demand singles
What packet.ai delivers
packet.ai does one thing: GPU compute. No VMs, no Kubernetes, no load balancers. That single focus is what makes it 40–56% cheaper than Vultr on every card they share. Instead of building a general cloud that also rents GPUs, packet.ai uses hosted·ai's intelligent scheduler to keep GPUs busy across tenants and passes that utilisation back as lower prices, without compromising performance.
On the cards both platforms list, the rate difference is consistent: A100 80GB at $1.43/hr versus Vultr's $2.397/hr, L40S at $0.92/hr versus $1.671/hr, B200 from $3.75/hr Dynamic versus roughly $8.50/hr. packet.ai also carries RTX 6000 Pro and RTX 4090 that Vultr does not list.
Two ways to buy on packet.ai, mix both on one account:
- Dynamic: scheduler-enforced multi-tenant GPU with the same peak performance and VRAM as Dedicated. RTX 6000 Pro at $0.66/hr, B200 at $3.75/hr. Right for dev, bursty work and batch inference.
- Dedicated: the whole card, exclusively yours, with a 99% SLA and zero scheduler interference. RTX 4090 at $0.39/hr, L40S at $0.92/hr, A100 80GB at $1.43/hr, B200 at $6.99/hr. For production serving where p99 latency matters.
What packet.ai gives up is Vultr's 32 regions and its surrounding services: VMs, Kubernetes, block storage and DNS. Nine sites across the US and EU, object storage, and GPUs. If that geography covers you and your team manages its own infra layer, you pay for GPU hours you actually need, not the surrounding general cloud.
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 Vultr, packet.ai is 40% cheaper on the A100 80GB ($1.43 versus $2.397), 45% cheaper on the L40S ($0.92 versus $1.671) and 56% cheaper on the B200 ($3.75 Dynamic versus about $8.50), and it adds RTX 6000 Pro and RTX 4090 that Vultr does not list. Vultr keeps the regions and the general-cloud services.
packet.ai vs Vultr 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 | Vultr |
|---|---|---|
| A100 80GB | $1.43/hr Dedicated−40% | $2.397/hr PCIe |
| L40S (48GB) | $0.92/hr Dedicated−45% | $1.671/hr |
| B200 (180–192GB) | $3.75/hr Dynamic · $6.99/hr Dedicated | ~$8.50/hr per GPU (8× bare metal) |
| RTX 6000 Pro (96GB) | $0.66/hr Dynamic | Not offered |
| RTX 4090 (24GB) | $0.39/hr Dedicated | Not offered (A40 $1.712/hr, A16 in some regions) |
| H100 (80GB) | Launching soon (notify list) | Available, largely in reserved bare-metal shapes |
| GH200 / AMD | Not offered | GH200 $1.99/hr · MI300X available |
| Regions | US (4 sites) + EU (5 sites) | 32 regions; GPUs in a subset |
| Billing | Hourly; monthly commits up to 20% off | Hourly; reserved contracts for large shapes |
| Data transfer | No ingress fee; egress $0.04/GB | Bandwidth allowance then ~$0.01/GB |
| General cloud services | Object storage only | VMs, Kubernetes, block/object storage, LBs, DNS |
Detailed comparison
Rate card overlap
Three cards appear on both rate cards and packet.ai is cheaper on all of them. The A100 80GB is $1.43/hr against $2.397/hr, the L40S is $0.92/hr against $1.671/hr, and the B200 is $3.75/hr Dynamic against roughly $8.50/hr per GPU on Vultr's 8× bare-metal node. Even packet.ai's Dedicated B200 at $6.99/hr is 18% under Vultr.
Vultr has GH200 at a competitive $1.99/hr and AMD MI300X for teams that want either; packet.ai is NVIDIA-only and does not carry Grace Hopper. Vultr also has H100 in larger shapes, mostly reserved.
Regions and the general-cloud advantage
Vultr's 32 regions span the Americas, Europe, Asia, Australia and Africa, and GPU capacity exists in a meaningful subset. A team serving users in Singapore or São Paulo can put inference near them. packet.ai's nine sites are all US and EU today, with APAC rolling out.
The other Vultr advantage is everything that is not a GPU: cheap VMs, managed Kubernetes, block and object storage, load balancers, all in one account with one bill. packet.ai has object storage and GPUs, and expects you to bring the rest.
Bandwidth
Vultr's bandwidth is cheaper: generous allowances and $0.01/GB overage against packet.ai's $0.04/GB egress. For a GPU workload that streams large outputs to users this matters; for most training and inference it is a small fraction of the compute bill, where packet.ai's 40–56% advantage dominates.
Who should choose which
Choose packet.ai if
- GPU-hours are your primary cost and the 46–55% lower rate on A100 and L40S matters at your scale
- You want an RTX 6000 Pro or B200 that is not in Vultr's catalogue
- Your users and workloads are in North America and Europe, where packet.ai has capacity
- You want a Dynamic tier that undercuts Vultr's L40S price without any reservation
Choose Vultr if
- You need GPU capacity in APAC, Latin America, or other regions where packet.ai has no presence
- Your team already runs on Vultr and adding a GPU keeps everything on one account
- You need GPU alongside Vultr Kubernetes, managed databases, or Object Storage on one bill
- H100 SXM in multiple global regions is the card you need today
Conclusion
Vultr is a good general cloud that also rents GPUs, and its global footprint is real. packet.ai is a GPU cloud that is 40–56% cheaper on every shared card. If you need the regions or the surrounding services, pay Vultr's GPU premium. If you need the GPU, packet.ai is the cheaper hour.
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.