GPU EconomyCloud GPU prices, read every hour
Latest read 97 providers tracked13,250 prices read in 24 h
H100 index$3.5334 providersWidestV100 32GB20.3×Mover, 7dRTX A4000+8.5%

Analysis · · figures frozen on that day

Phala H200 went $4.80 to $5.50, B300 $6.50 to $8.50, October 19, 2026

Every figure below is a reading this site published - a rate read on a provider's own page, or a value of the index computed from such rates - listed with its source under the text and frozen as of that day: prices move, and a dated piece says what was true on its date. The live tables are linked beside each figure.

Phala Cloud’s GPU TEE page raised two of its three GPU rates between our reads of October 19, 2026 at 18:40 and 19:39 UTC. The H200 141GB went from $4.80 to $5.50 a GPU-hour on-demand (+14.6%), and the B300 288GB on its 1-month term from $6.50 to $8.50 a GPU-hour (+30.8%). The H100 80GB kept its $3.08 in the same reads.

What moved, and by how much

Both old figures had been the same in every hourly read since we first read the page on October 11. The H200 rose by $5.50 - $4.80 = $0.70 a GPU-hour and the B300 by $8.50 - $6.50 = $2.00. The page prints a minimum beside each price: “24h minimum” for the on-demand H100 and H200, “30d minimum” for the 1-month B300. At that minimum, one H200 went from $115.20 ($4.80 × 24) to $132.00 ($5.50 × 24), and one B300 for 30 days from $4,680 ($6.50 × 720) to $6,120 ($8.50 × 720).

Phala sells confidential-computing GPUs: the page says its prices include Intel TDX and NVIDIA confidential computing readiness, and we record the figure as printed, one GPU per row. Each card on the page also shows a “Slot” price labelled reserved with no term; we do not read it, so no figure here comes from it.

Where the new rates sit

Our card pages compare each provider’s cheapest on-demand rate, its rate card. In the build of 05:39 UTC on October 20, the median of 50 H200 rate cards on /gpu/h200 was $4.65, with the middle half between $3.68 and $5.97. Phala’s $5.50 ranked 35th of 50 from the cheapest. Phala’s unchanged H100 sits on the card page for an H100 whose form factor the seller does not state: there the median was $3.80 over 25 rate cards, and $3.08 ranked 9th of 25.

The B300 rate is a commitment, so it stands beside the other committed B300 rates on /reserved rather than in the on-demand median. In the same build those were Paperspace at $7.94 a GPU-hour for 12 months, and Civo at $6.49 for one year and $5.49 for three. Phala’s $8.50 is the only 1-month term among the four, and the highest of them. Nothing here is forecast: this is what the reads returned.

Who moved the figure

The Phala parser module was not changed in the days of the move, and the H100 row it reads from the same page kept its figure, so the H200 and B300 changes came from Phala’s page. Both new figures were read again in all 12 hourly reads from 19:39 UTC on October 19 to 06:39 UTC on October 20.

How to check these figures yourself

Every figure above is listed under this text with the page it links and the time we read it. The live rates are on Phala’s provider page and its H200 and B300 pages; the medians are on the H200 and H100 card pages, and the committed rates on /reserved. Those pages update with every hourly read; this article keeps the figures of October 19 and 20.

Every figure in this piece, and where it was read

As published on October 20, 2026. Each price links to the page it was read from; the last column links the page that keeps it current.
FigureAs publishedRead from, whenLive page
Phala Cloud, NVIDIA H200 141GB, on-demand with a 24h minimum, the last read before the move$4.80 per GPU-hour, listed since our first read of 2026-10-11source
/providers/phala/h200
Phala Cloud, NVIDIA H200 141GB, on-demand with a 24h minimum$5.50 per GPU-hour, up from $4.80, +14.6%, listed since 2026-10-19 19:39 UTCsource
/providers/phala/h200
Phala Cloud, NVIDIA B300 288GB, 1-month term with a 30d minimum, the last read before the move$6.50 per GPU-hour, listed since our first read of 2026-10-11source
/providers/phala/b300
Phala Cloud, NVIDIA B300 288GB, 1-month term with a 30d minimum$8.50 per GPU-hour, up from $6.50, +30.8%, listed since 2026-10-19 19:39 UTCsource
/providers/phala/b300
Phala Cloud, NVIDIA H100 80GB, on-demand with a 24h minimum, unchanged in the same reads$3.08 per GPU-hour, listed since our first read of 2026-10-11source
/providers/phala/h100
Median rate card, H200, on-demand, as /gpu/h200 prints it$4.65 per GPU-hour, 50 rate cards, middle half $3.68 to $5.97; Phala rate-card rank 35 of 50source
/gpu/h200
Median rate card, H100 (form factor not stated), on-demand, as /gpu/h100 prints it$3.80 per GPU-hour, 25 rate cards; Phala rate-card rank 9 of 25source
/gpu/h100
Other committed B300 rates live in the same buildPaperspace 12-month $7.94, Civo 1-year $6.49 and 3-year $5.49 per GPU-hoursource
/reserved

Answers the reader question: check my provider's rates before I renew. Written because Phala Cloud's GPU TEE page returned new rates for the H200 and the 1-month B300 in our read of 2026-10-19 at 19:39:52 UTC, the first change to either since our first read of 2026-10-11; the Phala parser module was not changed in the days of the move. All of it is open data under CC BY 4.0; the numbers here are a snapshot, the index and the card pages are current.

Earlier pieces: — RunPod vs Vast.ai, October 19, 2026: Vast.ai lower on 15 of 23 GPUs · — Lium first read October 17, 2026: 8×A100 80GB node $13.44 an hour · — Lyceum H100 went $3.49 to $4.69 an hour, B300 $11.09, October 16, 2026. All of them: Analysis.

Data updated: