Paperspace Pricing 2026: A4000 to H100 GPU Cost vs Spheron

Paperspace's rate card looks reasonable at a glance: $0.45/hr for an entry GPU, $5.95/hr for an H100. What the headline numbers don't show is that five of Paperspace's GPU tiers, including A100 and A100-80G, sit behind a paid Gradient subscription plan. This post is the exact rate card, what the subscription actually costs once you do the math correctly, and how the real numbers stack up against Spheron.

TL;DR: Paperspace Pricing vs Spheron (2026)

Paperspace's on-demand rates span $0.45/hr (M4000) to $5.95/hr (H100). A100, A100-80G, A6000, A5000, and V100 require a Growth-plan subscription, metered at $0.058/hr up to a $39/month cap, before you can launch them at all. H100 is the one high-end tier that doesn't carry that subscription-plan requirement in Paperspace's own pricing table.

Spheron's A100 80GB is roughly 53% cheaper than Paperspace's A100-80G rate, and Spheron's H100 is roughly 50% cheaper than Paperspace's H100 rate, both before Paperspace's Growth surcharge is added on top of the A100 tiers. Spheron requires no subscription plan for any GPU. If you've already decided Paperspace isn't the right fit on price, our Paperspace alternatives roundup compares ten providers side by side.

Pricing fluctuates based on GPU availability. The prices above are based on 06 Aug 2026 and may have changed. Check current GPU pricing → for live rates.

Paperspace GPU Pricing by Tier: A4000 to H100

Paperspace's parent company, DigitalOcean, acquired Paperspace for $111 million in July 2023 to expand its AI compute offering. Paperspace's brand and its Machines/Gradient product line still operate as a distinct product from DigitalOcean's own GPU Droplets, which is a separate SKU under the same parent company with its own pricing and billing rules. Don't confuse the two when comparing rates.

Full On-Demand Rate Card (M4000 through H100)

This is the complete GPU Machines rate card from DigitalOcean's official Paperspace pricing documentation, current as of the page's own last-verified date of 13 Jul 2026:

Read that table by tier, not just by price. The M4000 through A4000 range is available on the free or Pro plan, so you can test those without paying for Growth. A5000 and up, including both A100 variants, require Growth specifically. H100 stands apart: it's the most expensive card on the list, but it's the one high-end GPU that doesn't appear tied to a subscription plan in this table. Multi-GPU nodes scale linearly off the same per-GPU rate: Paperspace's own example puts an 8x H100 machine at 8 x $5.95 = $47.60/hr, and the same math holds for A100 (8x = $24.72/hr) and A100-80G (8x = $25.44/hr).

None of Spheron's catalog overlaps with Paperspace's workstation-tier cards (M4000 through A6000, V100). Spheron's GPU lineup starts at A100 80GB and runs through H100, H200, and B200, so the comparison below is apples-to-apples only from A100 up.

Multi-GPU Pricing and What Paperspace Doesn't Publish

Multi-GPU nodes are straightforward multiplication of the single-GPU rate, no volume discount applied. What isn't in the current documentation is any committed-use or reserved-pricing tier: no 6-month, 12-month, or multi-year discount schedule for any GPU, and no flat monthly rate for a dedicated GPU as an alternative to hourly billing. Every GPU on Paperspace's rate card today is on-demand hourly, full stop. That's worth knowing if you're coming from a provider that publishes reserved pricing and expect to find an equivalent lever here; as of this post's publish date, there isn't one.

How Paperspace Bills You: Hourly, Tied to Power State

Paperspace's documentation is direct about this: "All Paperspace resources are billed on a per-hour basis," and "Machine compute charges apply only while a machine is powered on." Shutting a machine down stops compute billing, even though the machine still exists in your account. Attached storage, public IPs, and other add-ons keep billing hourly until you destroy them, not just power them off.

This is a meaningful contrast with DigitalOcean's own GPU Droplets, a separate product line, which bill per second for as long as the Droplet exists, powered on or off. Paperspace's docs call this out explicitly as the difference between the two products. If you're evaluating both under the DigitalOcean umbrella, don't assume Paperspace's billing behavior carries over to GPU Droplets, or the reverse; see our DigitalOcean GPU Droplet pricing breakdown for that product's numbers. Paperspace also doesn't charge for ingress or egress bandwidth on Machines, which some competitors do meter separately.

The Growth Subscription: What $0.058/hr and a $39 Cap Actually Buys You

Here's where the math needs to be done carefully, because the common framing of "Paperspace charges $39/month to unlock A100" isn't quite what the pricing table says. Growth is metered, not a flat subscription fee:

Growth accrues at $0.058 per hour of use, capping at $39 for the month. Divide $39 by $0.058 and the cap only kicks in around 672 hours of usage, close to running a machine nonstop for 28 days straight. Below that, the plan behaves like a small, consistent surcharge rather than a flat tax you pay regardless of how much you use the GPU:

The practical read: if you're using an A100 or A100-80G lightly, you're not paying anywhere near $39 a month for the privilege, you're paying an extra $0.058 on every hour you actually run the machine. If you're running it close to continuously, the cap means the surcharge per hour actually shrinks. Either way, it's real money layered on top of the listed rate for exactly the GPUs (A5000, A6000, V100, A100, A100-80G) where Paperspace's own table shows Growth is required, and it stacks with the compute rate rather than replacing it.

Paperspace vs Spheron: Cost Per GPU-Hour Compared

A100 and H100 Side-by-Side, On-Demand and Effective Rate

Spheron's live GPU pricing as of 06 Aug 2026, pulled from Spheron's pricing page:

Paperspace's listed A100-80G at $3.18/hr is more than double Spheron's $1.48/hr on-demand rate, before Growth is even added. Paperspace's H100 at $5.95/hr, also PCIe form factor, is essentially double Spheron's $2.98/hr H100 PCIe rate. Neither comparison requires any commitment on Spheron's side.

Effective Hourly Cost Once You Add the Growth Subscription

Adding the $0.058/hr Growth surcharge to the A100 tiers (the typical case for anyone using less than roughly 672 hours a month):

Spheron aggregates supply from 5+ providers rather than a single data center footprint, which is part of why on-demand rates land where they do without needing a subscription tier to unlock the higher-end cards. For teams running fault-tolerant training jobs, Spheron's spot rates push the gap even wider: $1.19/hr for A100 80GB and $2.07/hr for H100, both without a Growth plan or any minimum commitment.

For a wider read on what these two GPUs are actually good for at these price points, see our A100 vs H100 specs and pricing guide. If your workload needs multi-GPU NVLink bandwidth rather than the PCIe form factor both Paperspace and Spheron list here, the tradeoffs are covered in our H100 NVL vs SXM5 vs PCIe guide.

Hidden Costs Beyond the Listed Hourly Rate

The per-hour GPU rate isn't the full bill. Watch for these on Paperspace:

• The Growth subscription meter. Not optional if you want A100, A100-80G, A6000, A5000, or V100. Budget $0.058/hr on top of the compute rate for those tiers specifically.
• Storage billed hourly, separate from compute. Growth includes 50 GB; beyond that, extra storage runs $0.29/GB/month, charged whether or not the attached machine is running.
• Storage and IP charges continue after shutdown. Powering off a machine stops compute billing, but attached storage and public IPs keep accruing hourly charges until you destroy them, not just stop them.
• No published committed-use discount. If your cost model assumes a reserved-pricing tier exists to bring the hourly rate down at scale, it doesn't currently, so budget at the on-demand rate.

Spheron's model is simpler on this front: no subscription tier gates any GPU, compute billing is per minute with no charges after you stop the instance, and there's no separate storage meter layered under a GPU-only workload. Full billing details are documented at docs.spheron.ai.

Who Paperspace's Pricing Still Makes Sense For

Paperspace's lower tiers, M4000 through A4000, are genuinely inexpensive and available without Growth, which makes them reasonable for light prototyping, small Jupyter notebook sessions, or testing a workload before committing real budget. If you're already inside the Gradient ecosystem for Notebooks, Workflows, or Deployments and your usage is modest, the Growth surcharge is a small, predictable add-on rather than a dealbreaker.

Where the math stops working is any A100 or H100 workload run with any regularity. Once you're past a handful of test runs, at $3.09-$5.95/hr on-demand plus the subscription meter on the A100 tiers, the cost gap against a platform with no subscription requirement and roughly half the per-GPU rate compounds fast, especially for training runs or production inference that run for hours or days at a time. For H100 GPU rental on Spheron or A100 GPU rental on Spheron, current on-demand and spot rates are visible with no subscription and no sign-up gate to check them. For a market-wide view beyond just these two providers, our GPU cloud pricing comparison 2026 covers 15+ providers side by side.

Paperspace's A100 and H100 rates run roughly double Spheron's on-demand pricing before any subscription surcharge is added. Spheron's A100 80GB and H100 are available with no Growth plan, no commitment, and per-minute billing.

Rent A100 or H100 on Spheron →