Every price table tells you what an H100 costs. August's stock history tells you how often the provider would have said yes: 34% of the time for H100, 5% for B200.
GPU Finder polls provider stock APIs roughly every hour and keeps the history. For August 2026 that history is frozen in the first complete monthly edition of the GPU Cloud Availability Index: 124 provider × GPU × node-size records, of which 81 cleared the coverage floor and carry a score. Those 81 cover 189,762 provider-hours of known signal from 17,405 stored observations, 1 August to 1 September UTC. Numbers here are rounded; the index carries the exact values and is downloadable as CSV and JSON.
Terms. A record is one provider × GPU × node size, summed across every region we track. Its covered hours are the hours with a known signal (available, limited or unavailable); unknown hours are dropped. A record is ranked with at least 48 covered hours and 50% of the hours it could have had. Available % is hours reported available divided by covered hours. limited (a handful of units) does not count, so every figure below is a floor for "could I have launched one", not a ceiling.
Key facts (GPU Finder, GPU Cloud Availability Index, August 2026):
- In August 2026, H100 capacity was reported available 34% of covered hours across the 12 cloud providers ranked in GPU Finder's GPU Cloud Availability Index (hourly stock-API checks, 1 to 31 August 2026 UTC).
- In August 2026, H200 and A100 were each reported available about 52% of hours; B200 was reported available 5% of hours across seven ranked providers, and under 1% outside one marketplace feed.
- In August 2026, an 8x H100 node was reported available 3% of the month at Lambda, 1% at Hyperstack and 0% at Runpod, against 51%, 35% and 18% for a single H100 at the same providers.
- 27 of the 81 ranked provider × GPU × node-size records showed no available stock at any point in August 2026.
H100, H200, A100 and B200 availability in August 2026
H100 availability: reported available 34% of hours
Across the 12 providers with enough H100 coverage to rank, H100 capacity was reported available for 33.5% of covered hours, coverage-weighted. Across all four GPUs and all ranked records, 37.5%.
One caveat belongs here, not in a footnote. PrimeIntellect is a marketplace that resells other clouds, and its feed scored 100% on every one of its ten records. Without those rows the numbers are H100 25%, H200 28%, A100 44% and B200 under 1%. How that feed works is explained below.
| GPU | Ranked providers | Available (% of covered hours) | Excluding PrimeIntellect |
|---|---|---|---|
| H100 | 12 | 34% | 25% |
| H200 | 6 | 52% | 28% |
| A100 | 5 | 52% | 44% |
| B200 | 7 | 5% | 1% |
The index page calls this column Reliability. The 30-day reliability score on the live GPU pages is the same calculation over a rolling 30-day window rather than the calendar month.
B200 availability: 5% of hours, 0% at four providers
B200 was almost never reported available on demand: 4.7% of covered hours, and the whole of that is one PrimeIntellect 8x listing at 100%. Six of the seven ranked B200 providers were below 3% for the entire month. Hyperstack, Lambda, Runpod and Vultr reported an 8x B200 node available for 0% of August.
8x H100 availability vs 1x: bigger nodes were far scarcer
At the providers that sell every node size, availability fell steeply with size. Lambda's H100 went from 51% at 1x to 3% at 8x, Hyperstack from 35% to 1%, Runpod from 18% to 0%, Shadeform from 56% to 21%. The exceptions are providers that were either always or never available regardless of size (PrimeIntellect, Digital Ocean, Cudo, AceCloud). Across all ranked records a single H100 was available 49% of the time, a 4x node 16% and an 8x node 28%; the 8x figure is lifted by three providers at 63% to 100%.
Live per-provider stock at full-node size is on the 8x H100, 8x H200 and 8x B200 server price pages.
27 of 81 records were never available
A third of all ranked provider × GPU × size combinations finished the month at 0%. Cudo and AceCloud reported no available H100 in any size across their whole covered period. Only 13 records finished at 100%, ten of them PrimeIntellect's.
H100 availability by provider, August 2026
Coverage-weighted across every node size a provider sells.
| Provider | Available (% of covered hours) | Covered hours | Node sizes |
|---|---|---|---|
| PrimeIntellect | 100% | 11,876 | 1x, 2x, 4x, 8x |
| Digital Ocean | 100% | 1,484 | 1x, 8x |
| Scaleway | 90% | 1,486 | 1x, 2x |
| Nebius | 81% | 1,488 | 1x, 8x |
| Shadeform | 39% | 28,080 | 1x, 2x, 4x, 8x |
| Lambda | 30% | 14,111 | 1x, 2x, 4x, 8x |
| Runpod | 10% | 13,367 | 1x, 2x, 4x, 8x |
| Hyperstack | 8% | 8,924 | 1x, 2x, 4x, 8x |
| Gcore | 7% | 6,011 | 8x |
| Vultr | 0% | 742 | 8x |
| Cudo | 0% | 9,649 | 1x, 2x, 4x, 8x |
| AceCloud | 0% | 2,969 | 1x, 2x, 4x, 8x |
A high score on a narrow catalogue is not the same claim as a middling score on a wide one. Records also sum configuration-hours across every tracked region, so a provider tracked in seven regions (Lambda) or sixteen (Shadeform) with stock in one of them scores a fraction, while single-region providers (Digital Ocean, Runpod, Scaleway, Cudo) score on one place to launch. Digital Ocean's signal is a per-size "can be created" flag rather than a stock count.
Caveats: marketplaces, a Theta revision, Vast.ai and the "stale" column
Two of the twelve are marketplaces. PrimeIntellect resells other clouds (Hyperstack, Runpod and DataCrunch appear in its feed). Its endpoint reports a stock level per upstream listing, and GPU Finder records the best level across all of them for each GPU and node size. 100% therefore means that at some upstream cloud, in some data centre, that configuration was at high stock in every covered hour. That is a real signal and easier to score 100% on than a single fleet's API; two of its upstreams, Hyperstack and Runpod, score 8% and 10% measured directly. Shadeform also aggregates other clouds but reports per-cloud stock counts, which is why the two score so differently on the same idea.
Theta EdgeCloud was removed from the rankings in a revision on 5 September. The first freeze ranked it second for H100 and H200 at 100%. That score was an artefact: until 23 August GPU Finder recorded Theta's static VM catalogue as available, although the catalogue carries no capacity signal, and those 538 hours cleared the floor on their own. The records were reclassified as unknown and the edition regenerated at the original freeze time; the revision note is on the index page and in the JSON, and every other record is unchanged.
Vast.ai, Lium and Packet are not ranked, for three reasons, none about stock. Vast's history is kept for seven days because its marketplace writes about 90% of all availability rows, so only a quarter of August survived to the freeze. Lium was added on 20 August and Packet's feed went live on 25 August, so neither could accumulate half a month. All three have live stock on GPU Finder today; Vast.ai in particular is often the cheapest H100 on the site, and its absence here is a retention limit, not a verdict.
stale on the index page means "no new record in the six hours before the freeze", not "out of date". GPU Finder polls hourly but stores a history record only when a provider's signal changes, plus one heartbeat per UTC day. The edition was frozen at 11:17 UTC on 1 September, the timestamp of the last stored check; 74 of 124 rows read stale because their signal had not changed since the 00:52 heartbeat. It mostly marks rows that were unchanged, and it does not feed the monthly percentage.
How to find an H100 available right now
- Shop by node size first. If you need 8x H100, start from the 8x H100 server price page, which shows live stock per provider at that size, rather than the headline 1x price.
- Read reliability next to price. Every listing on the H100 cloud pricing page carries its 30-day reliability score (how it is computed: GPU cloud reliability); the cheapest row is only cheap if it was launchable. The same column sits next to every price on the cheapest cloud GPU table.
- Citing these numbers. The GPU Cloud Availability Index, August 2026 edition is frozen, dated and downloadable. Suggested citation: Alexander Salikov, "GPU Cloud Availability Index — August 2026," GPU Finder, generated 1 September 2026.
Methodology: how GPU cloud availability is measured
Each provider's stock API is polled roughly hourly; a history record is stored whenever the signal changes, and at least once per UTC day. Each stored signal holds until the next record, capped at 30 hours, and a longer gap counts as missing, not downtime. The score is the share of covered time reported available, so a provider available for six days and out for one reads 86%, not 50%. August is the baseline: July cannot be rebuilt from the 30-day retained history, so September's edition is the first month-over-month comparison. Full definitions and the data dictionary are on the index page.