The 10,500% GPU Price Gap: Same Card, 105× Price Difference (Sep 2026)

One NVIDIA V100 rents for $0.03/hr at one cloud and $3.18/hr at another — a 10,500% spread on the same GPU. The cheapest listing is rarely the whole story. This snapshot comes from the live GPU tracker at serverlist.xyz: 23 providers, 3,093 offers, prices updated hourly.

The spread table: where the arbitrage lives

Serverlist ranks GPU models by the gap between their cheapest and priciest live provider rate (per-GPU $/hr):

GPU model	Cheapest	Priciest	Spread
V100	HardwareHQ $0.03/hr	Naver Cloud GPU $3.18/hr	10 500%
H200	HardwareHQ $0.50/hr	YouGPU $13.87/hr	2 674%
A100	GPU Finder $0.27/hr	Naver Cloud GPU $4.64/hr	1 618%
H800	AutoDL $1.22/hr	JD Cloud GPU $12.78/hr	947% (only 2)
H100 80GB	Alibaba Cloud GPU $0.79/hr	CloudCompute $7.96/hr	907%
T4	Alibaba Cloud GPU $0.17/hr	Yandex Cloud $1.38/hr	712%
L40S	GPU Finder $0.38/hr	Naver Cloud GPU $3.05/hr	703%
A10	Alibaba Cloud GPU $0.19/hr	JD Cloud GPU $1.50/hr	690%
A100 80GB	GPU Finder $0.47/hr	CloudCompute $3.63/hr	672%

What the table does not say (the tracker flags it honestly): models marked "only 2" are priced by a single provider pair — the spread is a weak signal, not a market fact. V100's 10,500% gap is real, but most of it is one outlier quote at one end.

The heatmap: DEAL% = floor undercut vs provider median

The interesting signal isn't the raw cheapest price — it's how far a model's floor sits below its own provider median. Serverlist's DEAL% channel (raw 0–95%, n=43/117 models, hatch = single source) ranks exactly that:

• V100 $0.06/hr — DEAL +96%: floor undercuts its median by 96%
• H200 $0.50/hr — DEAL +86%
• A100 $0.27/hr — DEAL +82%
• MI300X $0.50/hr — DEAL +79%
• RTX 4090 $0.11/hr — DEAL +72%
• H100 80GB $0.80/hr — DEAL +68% (floor 68% under its median)

A high DEAL% means one provider aggressively undercuts the pack on that model — exactly where a spot-market buy lives. Hatch marks single-source models: no competition to arbitrage.

What's actually cheap right now (live offers)

Cheapest live configurations on page one of the catalog (62 pages, 3 093 offers):

• AutoDL RTX 3060 — $0.11/hr (CN, CHEAPEST badge) · RTX 2080 Ti $0.12 · RTX A4000/3080 Ti $0.13
• RunPod RTX 3070 — $0.13/hr · RTX A5000 $0.16 · RTX A4000 $0.17 · RTX 3080 $0.17
• Vast RTX 5070 — $0.14/hr (US) · RTX 5070 Ti $0.14 (VN) · RTX 5080 $0.16 (CA)
• Vast RTX 4090 — $0.38/hr (JP, 192 GB across 4×) · RTX 5090 $0.39/hr (CN)
• H100 80GB: from $0.79/hr (Alibaba) — Vast lists it at $2.80/hr (US)
• H200 from $0.50/hr (HardwareHQ) · B200 $6.63/hr (Vast, 4× 716 GB)

Filter by workload: LLM training (80GB+ VRAM, 4–8 GPUs — 503 configs), inference (single GPU under $1/hr — 1 709 configs), consumer RTX (770 configs), enterprise A100/H100/H200/MI300X (407 configs).

Why the gap exists — and what to do about it

• Regional arbitrage. The same card costs 10–20× more at Korean (Naver) or Chinese hyperscalers than on global spot markets. HardwareHQ and GPU Finder consistently sit at the floor.
• Vast/RunPod/AutoDL are the deep discounters on consumer and older datacenter cards — their prices track utilization, not cost-plus.
• New flagships have no floor yet. B200 at $6.63/hr vs H200 at $13.87 (YouGPU) — spreads collapse as supply lands.
• Check "only 2" before acting. A 947% H800 spread between two providers is two price lists, not a market.

Bottom line

• Cheapest serious inference box: RTX 3060/3070-class at $0.11–0.13/hr (AutoDL/RunPod).
• Cheapest H100 80GB: $0.79/hr at Alibaba vs $2.80 at Vast — shop the model floor, not the brand.
• A100-class: from $0.27/hr (GPU Finder).
• The 10,500% V100 spread is the poster child: never rent the first quote for a GPU — the same card is 105× cheaper one click away.

Live prices refresh hourly. Full interactive view — heatmap with DEAL/ΔPRICE/SUPPLY channels, spread table, workload presets, 3 093 searchable offers: serverlist.xyz/gpu.