Google Cloud accelerator-optimized VM pricing

Accelerator-optimized VM table. Default region on load is us-central1 (Iowa) - the region selector ships with aria-selected=true on Iowa, so the top table is us-central1 prices; request ?region=us-east4 returned byte-identical figures, i.e. the page did not honour the query param server-side. a3-highgpu-8g (8x H100) = $88.490000119/h on-demand = $11.06/H100; a3-megagpu-8g (8x H100) = $93.400712807 = $11.68/H100; a3-ultragpu-8g (8x H200) = $84.806908493 = $10.60/H200; a2-ultragpu-8g (8x A100 80GB) = $40.550383123 = $5.07/A100; a2-highgpu-8g = $29.38708 = $3.67/A100; g2-standard-8 (1x L4) = $0.853624312. Also carries spot and 1y/3y CUD columns for some SKUs and states accelerator-optimized machines are NOT eligible for SUDs or flexible CUDs. Odd number to note: a3-highgpu-8g shows a 1-year CUD ($61.38) HIGHER than its on-demand rate ($88.49 is OD, but 3-year CUD is $38.86) - the column ordering in the rendered table is ambiguous, so treat the CUD figures as suspect and only the on-demand column as solid.