Vultr Cloud GPU Pricing 2026: H100, A100, GH200 Cost vs Spheron

Vultr cloud GPU pricing is refreshingly public. GH200 at $1.99/hr, A100 PCIe around $2.40/hr, and an 8x H100 bare metal server at roughly $23.92/hr, all bookable without a "request a quote" form standing between you and a number. That's rare enough in the neocloud market that it changes what kind of post we can write: instead of the usual caveat about sales-gated pricing, we can run an honest, apples-to-apples comparison against Spheron's live on-demand rates.

We pulled Vultr's numbers from pricing trackers DeployBase and ComputePrices, cross-checked where they diverge, and matched them against Spheron's live rates from the Spheron GPU pricing API, split strictly by instanceType so on-demand is compared to on-demand and spot to spot, never mixed. If you're comparing more than one partially-transparent neocloud, our Denvr Dataworks pricing breakdown covers a provider with a nearly identical publish-some-gate-some pattern, just on different GPUs.

TL;DR: Vultr Cloud GPU Pricing 2026 vs Spheron

At 05 Aug 2026 rates, Vultr was cheaper for GH200 and H100 bare metal, while Spheron was cheaper on every A100 tier and is the only one of the two selling spot.

Spheron rates are live as of 17 Sep 2026. Spheron A100 rental is where the 05 Aug 2026 gap was widest.

Vultr Cloud GPU Pricing 2026: The Published Rate Card

Here's the short version. These are the lowest published or tracked on-demand rates for Vultr's main AI accelerators, per GPU-hour:

Sources: DeployBase's Vultr GPU cloud pricing guide and ComputePrices' Vultr tracker. The two trackers disagree on H100 and A100 SXM framing, which we'll unpack below, because it actually matters for what you end up paying.

Pricing fluctuates based on GPU availability. The prices above are based on 05 Aug 2026 and may have changed. Check current GPU pricing → for live rates.

Vultr Cloud GPU Instance Pricing by Model

Vultr's lineup spans NVIDIA GH200, HGX H100, A100 (PCIe and HGX), L40S, A40, and A16, with early-access HGX B200 and, as of December 2025, preemptible AMD MI325X/MI355X plans layered on top (Vultr GH200 launch announcement). The three models below are the ones with real, comparison-shoppable self-serve numbers.

Vultr H100 8x Bare Metal Pricing

Vultr's primary self-serve H100 product is an 8x HGX H100 80GB bare metal server, not a single-GPU cloud instance. DeployBase tracks the node at roughly $23.92 per hour total, which works out to about $2.99 per GPU-hour (DeployBase pricing guide). That's the number to budget against if you're renting the way Vultr appears to sell it: a full 8-GPU chassis with bare-metal access, not a slice of a shared instance.

A separate tracker tells a different story. ComputePrices lists a "Vultr H100 SXM" line item at $2.30 per GPU-hour on what it describes as a per-GPU cloud VM basis (ComputePrices Vultr tracker), a lower number on a different packaging than DeployBase's flat bare-metal rate. We're flagging the discrepancy instead of picking a favorite, because it changes your math: if Vultr genuinely sells single or paired H100s at $2.30/hr, that's your number. If the real self-serve product is the 8-GPU node, $2.99/hr per GPU, with $23.92/hr as a hard floor whether you use all 8 GPUs or not, is what you're planning around.

Vultr A100 PCIe Cloud GPU Pricing

Vultr's A100 80GB PCIe on-demand rate runs about $2.397 per GPU-hour, per DeployBase's tracked rate card, on an instance configuration with 24 vCPUs and 80GB of system RAM alongside the 80GB of GPU memory (source). A separate ComputePrices listing puts A100 SXM pricing higher, at $2.60 per GPU-hour (source). SXM's NVLink interconnect and higher inter-GPU bandwidth typically carry a premium over PCIe, and Vultr's own numbers follow that pattern.

Vultr GH200 Grace Hopper Pricing

GH200 is Vultr's cheapest listed GPU and the one it's marketed hardest. DeployBase tracks Vultr's GH200 (96GB) on-demand rate at $1.99 per GPU-hour (source), the lowest self-serve number anywhere in this comparison. Vultr rolled GH200 out across its full data center footprint rather than a handful of flagship regions: at launch, Vultr's rollout spanned its entire global cloud footprint, 32 data center locations at the time, across the US, Latin America, EMEA, and APAC (DataCenterNews Asia coverage of the launch).

J.J. Kardwell, CEO of Vultr's parent company Constant, framed the launch as a coverage play as much as a hardware one: "With our global rollout of the NVIDIA GH200 Grace Hopper Superchip, AI innovators now have access to the most powerful GPU for AI inference and the ability to access it worldwide to support their local market latency, data sovereignty, compliance, and privacy goals."

For the architecture behind why that unified-memory design matters for inference workloads specifically, see our GH200 Grace Hopper architecture and performance guide.

Pricing fluctuates based on GPU availability. The prices above are based on 05 Aug 2026 and may have changed. Check current GPU pricing → for live rates.

Reserved and Committed-Use Discounts Explained

Vultr advertises annual prepayment discounts on top of its hourly on-demand rates, but it doesn't publish the exact percentage on its own rate card. Third-party benchmarking guide VendorBenchmark estimates GPU committed-use discounts in the 20-35% range depending on GPU family and commitment depth, breaking it down further: 1-year terms typically land around 15-22% (up to 20-28% with negotiating leverage), and 3-year terms run 22-30% (up to 28-38%) (VendorBenchmark's Vultr pricing analysis).

DeployBase's guide gives a concrete example: an A100 PCIe instance at $2.397/hr on-demand drops to roughly $1.80/hr on a 1-year prepaid term, close to a 25% discount, working out to an estimated $5,230 in annual savings at full-time utilization (source).

The number that actually matters here isn't the discount percentage. It's whether you'll run the reserved capacity enough to clear it. A 25% cut on hardware you use 40% of the time is a worse deal than paying on-demand only for the hours you need. Our serverless vs on-demand vs reserved GPU guide walks through that break-even math in more detail, and the logic holds regardless of whose reserved tier you're evaluating.

Vultr vs Spheron: On-Demand H100 Cost Compared

This is where published numbers meet published numbers. Spheron's rates below are pulled live from the Spheron GPU pricing API each time the page renders, filtered strictly by instanceType so dedicated and spot offers stay separate.

At the node level, compare Vultr's $2.99/hr per GPU bare-metal rate against Spheron's live on-demand column. ComputePrices tracks a lower $2.30/hr figure for Vultr, though we'd weight DeployBase's bare-metal number more heavily since it matches the packaging Vultr actually appears to sell as its flagship H100 product.

Where Spheron differs is flexibility Vultr's rate card doesn't offer at all. Vultr has no published spot or preemptible H100 tier, while H100 GPU rental on Spheron spot runs $2.79/hr. Spot capacity can be reclaimed without notice, so it fits workloads that tolerate interruption. If yours can't, and you need the full 8-GPU chassis with NVLink, Vultr's bare-metal rate is genuinely competitive.

Pricing fluctuates based on GPU availability. Spheron rates above are live as of 17 Sep 2026; other providers reflect their most recent published rates and may have changed. Check current GPU pricing → for live rates.

Vultr vs Spheron: A100 and GH200 Cost Compared

A100 is where Spheron's marketplace has historically been strongest against Vultr's published $2.397/hr (PCIe) and $2.60/hr (SXM) rates, and Spheron also sells A100 on spot, which Vultr does not. Compare the live columns above before you commit. Our A100 deployment guide covers the SXM vs PCIe and MIG-partitioning tradeoffs that determine which A100 tier actually fits a given workload.

GH200 is where Vultr's rate card is strongest. Its $1.99/hr rate is the cheapest published Vultr number in this comparison; check it against the live on-demand rate for Spheron GH200 instances. Worth noting: Spheron's marketplace aggregates live offers across 5+ providers, and GH200 listings come and go with partner supply. If GH200's unified memory is specifically what your workload needs, Vultr's number is the one to compare against.

Pricing fluctuates based on GPU availability. Spheron rates above are live as of 17 Sep 2026; other providers reflect their most recent published rates and may have changed. Check current GPU pricing → for live rates.

What's Included (and What's Not): Regions, Storage, Interconnect

Vultr bundles compute, listed storage, and interconnect into the hourly rate across its A100, H100, and GH200 tiers, no separate networking line item on the published cards (DeployBase's guide). What that rate does buy you is genuinely wide geographic reach: Vultr's GH200 footprint alone spans 32 global data center locations across the US, Latin America, EMEA, and APAC (DataCenterNews Asia), which matters if data sovereignty or local-market latency is part of your deployment decision, not just raw GPU-hour cost.

What isn't itemized on the public trackers is support tier, SLA terms, or uptime guarantees, standard omissions for a self-serve rate card, but not something you can comparison-shop without asking Vultr directly. The same goes for the exact terms of that annual prepayment discount: real percentage, minimum commitment, and overage handling all live behind a sales conversation once you're past the on-demand number. Our AI buyer's guide covers the fuller checklist of hidden costs and contract terms worth asking any GPU provider before you sign, not just Vultr.

Spheron's model is different in kind: it aggregates live offers across 5+ providers rather than running its own fixed regional footprint, with per-minute billing and instances deploying in under two minutes once you've picked a rate. That trades Vultr's single-vendor consistency for the ability to compare regions and hardware partners side by side inside one marketplace.

Real-World Monthly Cost at 200 Hours/Month

200 hours a month is a reasonable proxy for a team running GPUs on weekdays during business hours, common for iterative fine-tuning or inference load testing rather than 24/7 production serving. This is a worked example frozen at the rates recorded on 05 Aug 2026 (Spheron's figures then: GH200 $3.02/hr, A100 PCIe $1.48/hr on-demand and $1.20/hr spot, A100 SXM $1.82/hr, H100 SXM $3.38/hr on-demand and $2.91/hr spot, H100 PCIe spot $2.21/hr). Spheron's live rates are in the tables above:

• GH200, Vultr: $1.99 × 200 = $398/month
• GH200, Spheron on-demand (05 Aug 2026): $3.02 × 200 = $604/month
• A100 PCIe, Vultr: $2.397 × 200 ≈ $479/month
• A100 PCIe, Spheron on-demand (05 Aug 2026): $1.48 × 200 = $296/month
• A100 PCIe, Spheron spot (05 Aug 2026): $1.20 × 200 = $240/month
• A100 SXM, Vultr: $2.60 × 200 = $520/month
• A100 SXM, Spheron on-demand (05 Aug 2026): $1.82 × 200 = $364/month
• H100, Vultr (8x bare metal, per GPU): $2.99 × 200 = $598/month
• H100 SXM, Spheron on-demand (05 Aug 2026): $3.38 × 200 = $676/month
• H100 SXM, Spheron spot (05 Aug 2026): $2.91 × 200 = $582/month
• H100 PCIe, Spheron spot (05 Aug 2026): $2.21 × 200 = $442/month

On 05 Aug 2026, Vultr won on GH200 and on H100's flagship bare-metal packaging, and Spheron won on every A100 tier and on H100 spot. Neither number stays fixed for long. Spheron's rates move with real-time availability, and Vultr's published card can change too. Run this with your own utilization before you commit a budget to it.

The monthly figures above are a worked example at rates recorded on 05 Aug 2026. Spheron's live rates are in the tables earlier in this post, as of 17 Sep 2026. Check current GPU pricing → for live rates.

When Vultr's Model Makes Sense vs a GPU Marketplace

The tier-by-tier math doesn't produce one universal winner, and pretending it does would be dishonest:

• You need GH200 and want the lowest published rate anywhere. Vultr's $1.99/hr beats every other number in this comparison, including its own A100 and H100 rates.
• You need a full 8-GPU H100 bare-metal chassis and want a fixed, self-serve node rate. Vultr's $23.92/hr ($2.99/GPU-hr) is a fixed node rate; compare it with Spheron's 8x H100 on-demand node at $21.20/hr.
• You need A100, in any form factor. Spheron's A100 runs $1.43/hr on-demand and $1.19/hr on spot, against Vultr's published $2.397/hr (PCIe) and $2.60/hr (SXM).
• Your H100 workload can tolerate interruption. Spheron sells H100 spot at $2.79/hr; Vultr publishes no spot H100 tier at all.
• You want to compare multiple providers' hardware and regions in one place. Spheron aggregates offers across 5+ providers, so you're comparing vendors side by side rather than shopping one company's own data centers. Vultr's tradeoff is the opposite: a single vendor's infrastructure across a genuinely wide 32-region footprint, with one rate card to reason about instead of several.

For the wider market beyond just Vultr and Spheron, our GPU cloud pricing comparison across 15+ providers and top 10 cloud GPU providers roundup put more rate cards side by side. Deployment docs for spinning up an instance on Spheron are at docs.spheron.ai.

Vultr's rate card is one of the few in this market you can actually read without a sales call, and GH200 in particular is hard to beat on price. For A100 at rates Vultr's own card can't match, or H100 spot pricing with no minimum commitment, check Spheron's live numbers today.

Spheron H100 instances →