AI Infrastructure & GPU
GPU rental medians, compute tokens, and the hyperscaler capex anchor from XBRL.
Open offerings
0
of 0 published in this category
Live series
6
latest print 2026-10-31
GPU Rental Index — H100 on-demand median
$4.09/hr
Daily · 2026-10-17
Opportunities · Pro
Open offerings and new filings
Flow · Pro
How much capital is forming, and how fast
Not collected yet. The source is identified above and this section fills in when the collector ships.
Price
What the category's market is doing
$2.75/hr
2026-10-17 · +0.0% vs prior print
Source: AltStreet · link · CSV
METHODOLOGY
Median hourly on-demand A100 80GB (SXM4 and PCIe) rate across live-verified sources, per GPU, same rules as the H100 series. Method v2 from 2026-10-03: the print is the median of each provider's own median, so every provider counts once and a marketplace's many listings do not outvote list-price clouds; the plain median over offers is kept in raw.offer_median. Before that date the series sat on one provider's list price because the Vast.ai marketplace scrape was returning 2 to 3 offers a day; prints before and after the change are not like for like. Vast.ai contributes up to its 64 cheapest listings per model and pricing type.
$4.09/hr
2026-10-17 · +0.0% vs prior print
Source: AltStreet · link · CSV
METHODOLOGY
Median hourly on-demand H100 (SXM5 and PCIe 80GB) rate across live-verified sources (source_kind api/scrape, verified within 48h) in the GPU project, per GPU. Computed by markets-collect from v_gpu_yield_rankings. Method v2 from 2026-10-03: the print is the median of each provider's own median, so every provider counts once and a marketplace's many listings do not outvote list-price clouds; the plain median over offers is kept in raw.offer_median. Before that date the series sat on one provider's list price because the Vast.ai marketplace scrape was returning 2 to 3 offers a day; prints before and after the change are not like for like. Vast.ai contributes up to its 64 cheapest listings per model and pricing type.
$0.34/hr
2026-10-17 · +0.0% vs prior print
Source: AltStreet · link · CSV
METHODOLOGY
Median hourly spot/interruptible RTX 4090 rate across live-verified sources, per GPU. Method v2 from 2026-10-03: the print is the median of each provider's own median, so every provider counts once and a marketplace's many listings do not outvote list-price clouds; the plain median over offers is kept in raw.offer_median. Before that date the series sat on one provider's list price because the Vast.ai marketplace scrape was returning 2 to 3 offers a day; prints before and after the change are not like for like. Vast.ai contributes up to its 64 cheapest listings per model and pricing type.
Outcome
What investors got versus what was promised
No primary data exists for this layer in this category. Nothing is charted rather than something misleading.