Skip to content
HN On Hacker News ↗

Compute Cheap — The world’s cheapest GPU compute

▲ 60 points 20 comments by EmiCorleone 3d ago HN discussion ↗

Pangram verdict · v3.3

We believe that this text is a mix of AI and human-written content.

42 %

AI likelihood · overall

Mixed
61% human-written 39% AI-generated
SEGMENTS · HUMAN 1 of 2
SEGMENTS · AI 1 of 2
WORD COUNT 209
PEAK AI % 71% · §2
Analyzed
Sep 3
backend: pangram/v3.3
Segments scanned
2 windows
avg 105 words each
Distribution
61 / 39%
human / AI fraction
Verdict
Mixed
Pangram v3.3

Article text · 209 words · 2 segments analyzed

Human AI-generated
§1 Human · 24%

The world's cheapestGPU compute.H200 and H100 capacity for training, inference, and everything in between. Historically delivering some of the lowest prices for frontier AI.Explore GPUsRequestH100 on-demand pricing$2.038/ GPU-hourUpdated Sep 2, 2026H100 Compute Cheap on-demand price $2.038 per GPU-hour on Sep 2, 2026MarketCompute Cheap priceSelect Spot or On-demand and H100 or H200, then hover or focus the chart to compare the daily market benchmark with the Compute Cheap price. When the chart is focused, use the left and right arrow keys to inspect dates.Daily H100 on-demand market and Compute Cheap GPU-hour prices across the latest 31-day window.Lowest-price GPU capacityCheapest on spot. Cheapest on-demand.NVIDIA H200141 GBHBM3eLaunch sizes1×2×4×8×Spot$1.773/hrOn-demand$2.995/hrNVIDIA H10080 GBHBMSXM*Launch sizes1×2×4×8×Spot$1.187/hrOn-demand$2.038/hr*H100 memory specification shown for the SXM configuration.Compute Cheap H100 and H200 Spot and on-demand GPU-hour prices.

§2 AI · 71%

The market comparisons are refreshed daily above.The economicsWhy is it so cheap?01Competition sets the floor.The marketplace model lets capacity owners compete for workloads instead of hiding prices behind fixed tiers.02Idle capacity gets a job.Cheaper GPU clouds make use of available hardware, turning underused infrastructure into useful compute.03Reserved, not idle.Prepaid blocks are quoted at the on-demand rate after availability is confirmed. No multi-year contracts; capacity is reserved after payment.04A smaller stack.We are built around the GPU—not a maze of managed services and layers of enterprise overhead.