Accelerator · Tape rank 1 of 20 · NVIDIA
B200 SXM6 rental price
The NVIDIA B200 SXM6 rents for $6.69 per GPU-hour on-demand at Lambda as of 7 September 2026 (8× SXM6 instance.). Other sourced prints: $8.60 on-demand at CoreWeave, $4.26 spot at CoreWeave, $3.68 composite at SemiAnalysis. Booked / sales-gated.
Key numbers
| Buy-now price | $6.69 per GPU-hour | Lambda, on-demand, 8× SXM6 instance.. As of 7 Sep 2026. |
|---|---|---|
| 1-year term book | $9.86 | Lambda, 1CC 2w–1y. |
| Change, 1 quarter | 0.0% | Lambda $6.69 on 9 Jun 2026 to $6.69. |
| Change, 1 year | — | Lambda on-demand: no print a year ago. CoreWeave on-demand 1y: 0.0% ($8.6 → $8.6). |
| Since first print | +34% | $4.99 on 9 Jan 2026. |
| Availability | Booked / sales-gated | Crusoe sales-gated. SemiAnalysis (2 Apr 2026): rental capacity increasingly impossible; market-wide booked through August–September 2026. CoreWeave near-term capacity sold-out in August 2026 after a ~25% July raise. Lambda 1-Click Clusters print $9.86 / GPU-hr on the 16-GPU row (2 weeks–1 year); 1y+ is contact. |
| Memory | 180 GB HBM3e | |
| Tape rank | 1 of 20 | Score 3.00: liquidity 3, demand 3, frontier 3, each 0 to 3. Ordinal hygiene, not a valuation. |
Every sourced quote
Venue, term, price, date. The highlighted row is the display print: the most liquid public on-demand list, or a labeled spot print where on-demand is contact-sales.
| Venue | Term | $ per hour | As of | Note |
|---|---|---|---|---|
| on-demand | $6.69 | 7 Sep 2026 | 8× s×m · 8× SXM6 instance. | |
| on-demand | $8.60 | 7 Sep 2026 | ||
| spot | $4.26 | 7 Sep 2026 | ||
| composite | $3.68 | Apr 2026 | Public dashboard print. SA public prints stop at April 2026. | |
| Capacity Blocks | $12.36 | 7 Sep 2026 | cb · Not on-demand. | |
| 1y CUD | $11.12 | 31 Aug 2026 | On-demand N/A on the fetched list. | |
| listed, no number | — | 18 Aug 2026 | Sales-gated; no public dollar print in this snapshot. | |
| 1CC 2w–1y | $9.86 | 7 Sep 2026 | 1cc 16 · 1-Click Clusters, 16 GPU, reservation 2 weeks–1 year. Not instance OD. 1y+ is contact — no public card. | |
| 1CC 2w–1y | $9.36 | 7 Sep 2026 | 1cc 64 · 1-Click Clusters, 64 GPU, 2 weeks–1 year. Prefer the 16-GPU row on the tape. | |
| 1CC 2w–1y | $8.87 | 7 Sep 2026 | 1cc 256 · 1-Click Clusters, 256+ GPU, 2 weeks–1 year. Prefer the 16-GPU row on the tape. | |
| 1y+ | — | 18 Aug 2026 | 1cc 1y plus · Contact. No public 1y+ card. |
Price path
Steps, not candles. Each dot is a dated print; the line carries the last print forward until the next one. A dashed line is a series that stopped publishing.
Lambda on-demand printed $4.99 on 2026-01-09 and $6.69 on 2026-05-21. CoreWeave on-demand printed $8.60 on 2025-08-21 and $8.60 on 2025-08-21. Dated prints only; nothing between them is inferred.
What this page refuses to show
- Official NVIDIA MSRP
- 1h / 24h / 7d ticks (US lists do not tick daily)
- A blended OD / Capacity Block / SA composite average
- 1h / 24h / 7d ticks (US lists do not tick daily)
- An invented 30d / 90d / 1y without two dated prints
Questions people ask
How much does B200 SXM6 cost per hour?
$6.69 per GPU-hour on-demand at Lambda as of 7 Sep 2026 (8× SXM6 instance.). Other sourced prints: $8.60 on-demand at CoreWeave, $4.26 spot at CoreWeave, $3.68 composite at SemiAnalysis.
Why do B200 SXM6 prices differ so much between clouds?
Because the terms differ. On-demand is a list price with no commitment; spot is interruptible and priced to fill idle capacity; reserved and one-year contracts trade a commitment for a lower rate; AWS Capacity Blocks are scheduled reservations, not on-demand. The tape never averages these: each print carries its venue, term and date.
Has the B200 SXM6 price gone up or down?
Over one quarter, Lambda on-demand moved 0.0% ($6.69 on 9 Jun 2026 to $6.69).
Is the B200 SXM6 available to rent right now?
Crusoe sales-gated. SemiAnalysis (2 Apr 2026): rental capacity increasingly impossible; market-wide booked through August–September 2026. CoreWeave near-term capacity sold-out in August 2026 after a ~25% July raise. Lambda 1-Click Clusters print $9.86 / GPU-hr on the 16-GPU row (2 weeks–1 year); 1y+ is contact.
Other accelerators on the tape
Cite this page
CC BY 4.0 with attribution to compute.world. Vendor names and marks belong to their owners and identify the source of each quote.