On October 7, 2026, two GPU clouds we read every hour printed higher H100 rates. In our read at 19:40:43 UTC, Lyceum’s price page listed an H100 80 GB on demand at $3.49 a GPU-hour, up from $2.79 an hour earlier, a rise of 25.1%. In our read at 08:40:55 UTC, GMI Cloud’s page listed its H100 “from” $2.60 a GPU-hour, up from $2.00, a rise of 30.0%. Neither page names the form factor, so both rates sit on our H100 80GB card, where the median rate card at 06:41 UTC on October 8 was $3.51 across 20 rate cards.
Lyceum changed eight rates in one read
Lyceum publishes one table, captioned “USD per GPU hour, billed per second with no subscription”, with an on-demand and a spot column for each card. Four on-demand rates moved in the 19:40:43 UTC read: the H100 from $2.79 to $3.49, the L40S from $1.19 to $1.45 (up 21.8%), the A100 80 GB from $1.59 to $1.89 (up 18.9%) and the H200 from $4.29 to $4.49 (up 4.7%). The B200 at $6.49 and the B300 at $7.99 did not move.
All five spot rates we had been reading moved in the same read, by more: H100 from $1.10 to $1.79, up 62.7%; H200 from $1.60 to $2.29, up 43.1%; B200 from $2.40 to $3.29, up 37.1%; B300 from $2.99 to $3.99, up 33.4%; A100 from $0.80 to $0.99, up 23.7%.
Lyceum had redesigned that page earlier: from 06:40 UTC on October 5 our parser could not find its tables, and on October 7 at 09:03 UTC we changed the module to read the new one. The read at 09:41:09 UTC printed the old figures to the cent, and so did every hourly read until 18:40:53 UTC. The module was not touched again before 19:40:43 UTC, so this move is on Lyceum’s page, not in our reading of it.
At the new rate, a month of one H100 running all the time is $3.49 × 730 h, about $2,548, against $2.79 × 730 h, about $2,037, before the move. Against the card medians printed on October 8, Lyceum’s $3.49 H100 sits just below the $3.51 median, its $4.49 H200 below the H200 median of $4.54 (43 rate cards), and its L40S at $1.45 and A100 at $1.89 above their medians of $1.29 and $1.80.
GMI Cloud moved three “from” rates
GMI Cloud’s pricing page prints one figure per card as “from $X /GPU-hour”, an advertised starting rate, and our tables tag it that way. In the 08:40:55 UTC read on October 7, three of its four cards moved: H100 from $2.00 to $2.60, H200 from $2.60 to $3.20 (up 23.1%) and B200 from $4.00 to $5.00 (up 25.0%). GB200 stayed at $8.00. The read at 07:40:56 UTC still printed $2.00 for the H100, and the GMI module was not changed in that window. After the move, three of the 20 H100 rate cards on October 8 listed lower than GMI’s $2.60, and eight of the 43 H200 rate cards listed lower than its $3.20.
The Price Index is computed from the H100 SXM card only. Both rates here sit on the H100 80GB card, whose pages do not name a form factor, so neither move enters the index.
Checking these figures yourself
Every rate above was read from lyceum.technology/resources/pricing or gmicloud.ai/en/pricing at the minute named; both pages are public and print the dollar figure per GPU-hour as quoted here. /providers/lyceum and /providers/gmi-cloud list each current rate with the date it began standing, /changelog lists these moves, and /prices.json carries each row with its read time. Nothing here is forecast and nothing is a recommendation: these are the figures two price pages printed on the dates named.