NVIDIA H200 SXM Rental Price
All H200 SXM rental rates, live: $/hr, ₹/hr, $/day, $/month across 5+ providers — on-demand and spot. Not the hardware price. The actual cost to run your workload on the 141GB Hopper GPU.
Updated from live feed
H200 SXM Rental — live providers
| Provider | On-demand $/hr | Spot $/hr | ₹/hr | $/day | $/month | Spot saving |
|---|---|---|---|---|---|---|
| Loading live rental prices… |
*Live gpuprice.in feed (pricing-data.json, source Parallel.ai), updated daily. Spot saving = % cheaper than that provider's on-demand. $/day = 24h, $/month = 730h, ₹/hr ≈ ₹83/USD.
H200 specs that matter for rental
The H200 is the big-VRAM Hopper card: 141GB of HBM3e — nearly 2× the H100's 80GB — fits models an 80GB card physically cannot run in a single GPU (100-200B-parameter class at lower precision). Crucially, the H200 keeps the full H100 compute (989 FP16 TFLOPS), so you gain VRAM headroom without sacrificing throughput.
The payback is scarcity: only ~5 providers serve H200 on the live feed, and spot is extremely thin (Nebius at $2.00/hr is currently the only spot listing). With less market competition, prices run a little above H100 per hour (~$2.80 vs $2.50) — you pay a modest premium to fit a bigger model on one card instead of sharding across two H100s.
How much will your H200 workload cost?
Enter hours — get cost for on-demand or spot, and what you'd pay on a comparable setup.
Need $/1M-token cost? Use the Workload Cost Calculator → to price a real inference workload on the H200 vs any GPU that fits your model.
On-demand H200
Pick a provider + enter hours
Spot H200 (interruptible)
Pick a provider + enter hours
Example: 1,000 hours of 120B-model inference on the H200 = ~$2,800 on-demand at the cheapest rate, vs ~$2,000 on spot (spot is scarce, though — check availability first). If your model fits an 80GB card, the H100 is cheaper; the H200 pays off only when you genuinely need the 141GB.
H200 vs alternatives: rent by what fits
Usually the choice is VRAM, not brand. Cheapest live on-demand rate per GPU.
| GPU | Arch | VRAM | Cheapest $/hr | Cheapest ₹/hr |
|---|
H200 vs H100
Both are Hopper cards with identical 989 FP16 TFLOPS compute. The H200's edge is 141GB vs 80GB — it runs a 100-200B model on one GPU where an H100 needs two (and sharded serving/overhead). Roughly $0.30/hr more (~$2.80 vs $2.50). Choose H200 when the model doesn't fit 80GB; choose H100 when it does.
H200 vs B200
The B200 (192GB, Blackwell) fits even bigger 400B-class MoE models and is ~2× the FP8 throughput, but it is a premium/$4.50+/hr flagship with only 2 providers. The H200 is the pragmatic middle ground: 141GB at near-H100 pricing beats paying flagship prices unless you truly need 192GB or Blackwell speed.
H200 vs A100
The A100 (80GB, $1.67/hr) is far cheaper but can't fit anything over 80GB and is ~3× slower. If VRAM is your limiter and your model is 80-140GB, the H200 is the single-GPU answer; if it fits 80GB, the A100 saves money on batch work.
H200 rental vs buying
An H200 SXM is roughly $30,000-40,000 to buy (≈₹25 lakh+ in India), plus hosting, power and depreciation. Renting at the cheapest $2.80/hr is about $2,044/month. Rent unless you run it 24/7 at high utilisation with a multi-year runway.
The strategic angle: the H200 lets you defer buying by renting the specific VRAM tier you need — 80GB today, 141GB tomorrow — without owning expensive hardware that gets obsolete. See full ₹-based TCO in GPU rental vs buy in India.
Cheapest H200 rental (right now, live): $2.80/hr on-demand at Spheron; $3.10/hr at Vast.ai. Note the thin market — spot is currently only Nebius at $2.00/hr, and E2E/AWS/Azure list no H200. Rates move weekly — this page is the live source.
H200 Rental — FAQ
Honest answers for teams renting NVIDIA H200 SXM compute.
How much does it cost to rent an NVIDIA H200 SXM?▾
Live on this page: H200 on-demand from $2.80/hr (≈₹232/hr) at Spheron, $3.10/hr at Vast.ai. Spot from $2.00/hr (Nebius — currently the only spot listing). A day is ~$67; a 730-hour month ~$2,044.
What makes the H200 different from renting an H100?▾
Identical compute (989 FP16 TFLOPS) but 141GB vs 80GB VRAM — it runs 100-200B-parameter models on a single GPU where the H100 needs two. Costs ~$0.30/hr more. Choose H200 when the model exceeds 80GB; H100 when it fits.
Why is H200 spot so hard to find / rental supply thin?▾
The H200 has fewer providers (only ~5 on the feed) and almost no spot market because demand for big-VRAM capacity outstrips supply. Unlike the H100 or A100, you often can't fall back to cheap spot — reserve on-demand capacity instead.
What can I run on a rented H200 (141GB)?▾
Larger open LLMs (100-200B-parameter class) at lower precision in a single GPU, high-throughput inference, fine-tuning, and big-batch training. For 400B-class MoE that exceeds 141GB, step up to the B200 (192GB).
Which provider rents the H200 cheapest?▾
Spheron leads on-demand at $2.80/hr; Vast.ai at $3.10/hr; Nebius $3.50/hr. Spot is Nebius only at $2.00/hr. Availability is thinner than H100/A100, so check live rates before planning capacity.