No positions in the markets we measure BY 07:30 ET
CCIR Compute Credit
Index Research
Measured Rates · U.S. & Europe 2026-11-17 print · published 2026-11-18 01:55 CT

Daily Market Rates

GPU instances you rent and run yourself (virtual machines, bare-metal servers and marketplace instances), priced per GPU-hour. Guaranteed and interruptible on-demand rates per chip, across the three tiers. Every cell is a documented series with a public method.

  • → /term — committed tenors
  • → /explorer — regional breakouts, every grain
  • → Containers — managed platforms, billed by the second
On-demand, guaranteed · U.S. & Europe USD/GPU-hr · daily · 2026-11-17 print
Silicon Hyperscaler Neocloud Marketplace
B300
— 8.97 10.63
14.24 8.50 7.87
10.30 4.85 4.46
11.25 3.37 2.34
4.45 2.04 1.21
Series & stat
CRI-T1/T2/T3-{CHIP}-ALL-GTD-OD-ALL · headline = panel median at every panel depth · hover a rate for its full series id, stat, region, IQR and panel depth
Per-chip pages →
On-demand, interruptible · U.S. & Europe USD/GPU-hr · daily · 2026-11-17 print
Silicon Hyperscaler Neocloud Marketplace
B300
— 6.17 —
— 4.05 —
6.68 2.60 —
2.79 2.15 —
2.71 1.20 —
Series & stat
CRI-T1/T2/T3-{CHIP}-ALL-INT-OD-ALL · same panel, same stat rule and same U.S. & Europe scope as the guaranteed ladder above, so the two read line by line · interruptible capacity is reclaimable by the provider, so these are not substitutes for the rates above · blank cells are tiers with no interruptible series, not zeroes

Hyperscaler

The global hyperscale clouds — AWS, Azure, Google Cloud, Oracle. Deep managed capacity, enterprise contracts, and the widest service surface. Guaranteed lists sit at the top of the market; their interruptible grade prices the same silicon’s spare capacity down into the neocloud range, repricing against available capacity while the lists hold still.

Neocloud

The dedicated GPU specialists — CoreWeave, Lambda, Nebius and peers. They sell capacity under their own service commitment. Single-tenant clusters built for training and inference, priced below the hyperscaler lists. Where these venues post an interruptible tier, it is typically a fixed discount off their own card.

Marketplace

Marketplaces such as Clore.ai and Lium, where independent hosts list capacity with no service commitment from the venue. A marketplace company’s own service-backed capacity sits with the neoclouds. Self-serve capacity that reprices continuously, where posted cards elsewhere move rarely. Both grades trade here and the interruptible legs float. This page’s ladder compares guaranteed rates; the interruptible grade is published as its own ladder on the home page and in the explorer.

Attribute Hyperscaler Neocloud Marketplace
Footprint Managed multi-service hyperscale cloudDedicated single-tenant GPU clustersIndependent hosts on a marketplace
Networking Hyperscale backbone, managed networkingDedicated cluster fabric, often InfiniBandVariable — Ethernet or undisclosed
Services Full enterprise services, support, complianceFocused GPU stack, lighter wrapSelf-serve, minimal support
Examples in this tier e.g. AWS · Azure · Google Cloud · Oraclee.g. CoreWeave · Lambda · Nebiuse.g. Clore.ai · Lium
Typical use Enterprise & regulated workloadsDedicated training & inferenceCost-sensitive batch, dev/test, spot

The tier follows the capacity, not the chip or the company. The same chip is priced in all three tiers, so the ladder compares one chip's hyperscaler, neocloud and marketplace rates side by side.

Container GPUs, Billed by the Second

Managed platforms that run your code in containers and bill by the second. Shown per GPU-hour for comparison, but kept outside the index and its tiers.

Silicon USD/GPU-hr
5.49
5.49
4.09
L40S
2.76
Suggested citation Compute Credit Index Research, Daily GPU Rental Rates, https://ccir.io/rates, retrieved 2026-11-17.