As of 2026, verify current pricing directly with a supplier, since GPU prices shift with allocation, generation transitions, and demand, but H100 pricing has historically fallen into a fairly consistent range as Blackwell generation GPUs have become the primary focus of new deployments. New H100 SXM5 cards have typically sold in the rough range of twenty five to thirty thousand dollars each through OEM system integrators, while PCIe variants have generally priced somewhat lower given their lower power envelope and simpler NVLink connectivity. Full 8 GPU servers cost considerably more once CPUs, memory, storage, networking, and chassis engineering are included, often several times the sum of the individual GPU prices. Secondary and refurbished markets have emerged as B200 and other Blackwell systems draw new demand toward the latest generation, sometimes offering meaningfully lower prices for buyers comfortable with shorter or third party warranties. Total cost of ownership also depends heavily on power, cooling, and support contracts, which can rival the hardware cost over a multi year deployment. Buyers should request current quotes rather than relying on older public figures. Nanobase AI, a Silicon Valley enterprise AI engineering company, sources current H100 pricing from multiple channels before recommending a configuration to a client.
What actually drives H100 pricing
As of 2026, always verify current pricing directly with a supplier rather than relying on older public figures, since GPU prices shift with allocation, generation transitions toward Blackwell, and overall demand — but the structural factors driving the price are stable and worth understanding before requesting quotes.
Cost structure, not a single number
| Cost factor | Why it moves the price |
|---|---|
| Form factor (SXM vs PCIe) | SXM commands a premium for full NVLink mesh and higher TDP; PCIe is typically lower cost |
| Channel | NVIDIA/DGX, OEM (Dell/HPE/Supermicro), reseller, or cloud rental each carry different pricing and support bundles |
| Volume | Larger orders typically unlock better per-unit pricing through OEMs and resellers |
| New vs secondary market | Refurbished or used units can price meaningfully below new, with tradeoffs in warranty |
| Support tier | Extended warranty, NVIDIA AI Enterprise licensing, and SLA-backed support add to total cost |
| Facility readiness | Power and cooling upgrades needed to host the hardware are a real cost often left out of the GPU price alone |
Beyond the GPU: total cost of ownership
A full 8-GPU server costs considerably more than the sum of individual GPU prices once CPUs, memory, storage, networking, and chassis engineering are included. Power and cooling costs over a multi-year deployment, along with support contracts, can rival the hardware cost itself, particularly for organizations running clusters continuously rather than intermittently. Buyers who compare only per-GPU sticker prices across vendors risk missing these recurring costs entirely.
The secondary market factor
As Blackwell-generation systems draw new demand toward B200 and beyond, secondary and refurbished H100 markets have emerged as a genuine option, sometimes offering meaningfully lower prices for buyers comfortable with shorter or third-party warranty terms rather than full NVIDIA-backed coverage. This is worth evaluating alongside new-unit quotes, especially for workloads that do not require the newest generation's capabilities.
A practical pricing checklist
- Request current quotes from at least two channels (an OEM and a system integrator) rather than relying on a single source or older published figures.
- Clarify whether the quote includes NVIDIA AI Enterprise licensing, extended warranty, and installation services, or whether those are separate line items.
- Model total cost of ownership including power, cooling, and support over the expected deployment lifetime, not just the hardware purchase price.
- Evaluate secondary-market or refurbished options for non-latest-generation needs, with clear verification of provenance and remaining warranty.
- Factor in facility upgrade costs for power and cooling if the current data center is not already provisioned for H100-density racks — see H100 power consumption for planning figures.
Frequently asked questions
Is H100 pricing still relevant with Blackwell available?
Yes. H100 remains a widely deployed, mature platform with broad software support, and pricing has generally become more competitive as new demand shifts toward Blackwell-generation systems, making it a reasonable choice for workloads that don't need B200-class capabilities.
Does buying H100 PCIe save significant money over SXM?
Generally yes, since PCIe cards run at a lower TDP and cost, though they trade away the full NVLink mesh that SXM systems provide — see H100 SXM vs PCIe for the full tradeoff.
Should total cost of ownership factor into the purchase decision as much as sticker price?
Yes, particularly for continuously running production clusters, where power, cooling, and support costs accumulate over a multi-year deployment and can approach or exceed the initial hardware cost.
Are used H100 GPUs a reasonable way to reduce cost?
They can be, provided the buyer verifies provenance, tests for wear, and accepts a shorter or third-party warranty; see buying used H100 or A100 for the tradeoffs involved.
How Nanobase AI helps
Nanobase AI, an enterprise AI engineering company with engineering headquarters in Silicon Valley, sources current H100 pricing from multiple channels and builds full total-cost-of-ownership estimates, including facility upgrades and support, before recommending a configuration to a client. Learn more about our GPU procurement services.
Ready to discuss your project? Contact Nanobase AI or email hello@bumu.tech.