Menu

Hyperstack GPU prices: A100, H100, H200, B200 per hour

We may earn commissions when you shop through the links below. Learn more

NexGen Cloud's GPU cloud: per-minute NVIDIA GPU VMs in Canada, the US and Norway.

GPU cloudInference hostingPrivate AI hosting

GPU-focused cloud · Headquarters: London, United Kingdom

From published sources, not hands-on tested

From
$0.50≈ €0.447 (estimated conversion from $0.50)/GPU-hrRTX A6000 48GB · On-demandChecked hyperstack.cloud
GPU families
6
B200 · H200 · A100 · +3 more · 8 models
Regions
3
None in the EU
Billing
Per minute
Hourly, spot, reserved, serverless, usage-based

Prices checked 42 sourcesSources (42)How we check

In short

Hyperstack is the GPU-as-a-Service platform of London-based NexGen Cloud, launched in 2023. It rents NVIDIA GPU virtual machines by the minute from prepaid credit, priced per GPU-hour: for example H100 PCIe $2.50, H100 SXM $3.20, H200 SXM $3.99 and B200 $6.00 on-demand. The larger GPUs are sold only as 8-GPU VMs.

Read full summary

Spot VMs cost 20% less for five GPU types, and 'starting from' reservation rates are published. Self-serve regions are CANADA-1 (Quebec), US-1 (Texas) and NORWAY-1 (Vestland). Most GPU types run only in CANADA-1, and the Norwegian region lists only RTX A4000 GPUs. Data transfer is free and block volumes cost about $0.07 per GB-month. On-demand Kubernetes, S3-compatible object storage (CANADA-1 only) and a vLLM-based Dedicated Inference wizard are also offered.

From

$0.50≈ €0.447 (estimated conversion from $0.50)/GPU-hrRTX A6000 48GB · On-demandChecked hyperstack.cloud
Visit Hyperstack Opens hyperstack.cloud in a new tab · Check the final plan, price and terms there before you order.Compare with alternatives to Hyperstack
Setup guides for Hyperstack
Billing increment
Per minute
Egress
No egress feesSource
Persistent storage
Yes
Managed service
Partial

Pricing

Prices checked 30 pricesHow we check

Billing: per minute, hourly, spot, reserved, serverless, usage-based. Lowest pay-as-you-go GPU price: $0.50≈ €0.447 (estimated conversion from $0.50) per GPU-hour (RTX A6000 48GB, on-demand). Other price types start lower: $0.35≈ €0.313 (estimated conversion from $0.35) per GPU-hour (RTX A6000 48GB, reserved).

Prices in USD, as listed, excluding VAT.≈ Estimated conversion from the listed USD price at the ECB reference rate of 23 Oct 2026 (1 EUR = 1.1193 USD). Billing currency not verified: providers may bill in USD, and your bank's rate and fees differ. Prices exclude VAT.

GPU offers25

Hyperstack GPU offers: 25 offers grouped by GPU; prices per GPU-hour, excl. VAT.

NVIDIA B200 SXM 180GB180 GB VRAM · 2 offers · from $5.10≈ €4.56 (estimated conversion from $5.10) per GPU-hour

Checked hyperstack.cloud

  • Reserved8-GPU node$5.10≈ €4.56 (estimated conversion from $5.10)/GPU-hrVaries (CANADA-1) · monthly · Not verified
  • On-demand8-GPU node · 252 vCPU · 2048 GB RAM · 41 TB disk$6.00≈ €5.36 (estimated conversion from $6.00)/GPU-hrVaries (CANADA-1) · per minute · Not verified

NVIDIA H200 SXM 141GB141 GB VRAM · 2 offers · from $2.79≈ €2.49 (estimated conversion from $2.79) per GPU-hour

Checked hyperstack.cloud

  • Reserved8-GPU node$2.79≈ €2.49 (estimated conversion from $2.79)/GPU-hrVaries (CANADA-1) · monthly · Not verified
  • On-demand8-GPU node · 176 vCPU · 1800 GB RAM · 48100 GB disk$3.99≈ €3.56 (estimated conversion from $3.99)/GPU-hrVaries (CANADA-1) · per minute · Limited
Show all 25 offers (from $0.35≈ €0.313 (estimated conversion from $0.35) per GPU-hour)

NVIDIA A100 PCIe 80GB80 GB VRAM · 5 offers · from $0.95≈ €0.849 (estimated conversion from $0.95) per GPU-hour

Checked hyperstack.cloud

  • Reserved$0.95≈ €0.849 (estimated conversion from $0.95)/GPU-hrVaries (CANADA-1) · monthly · Not verified
  • Reserved8-GPU node$0.98≈ €0.876 (estimated conversion from $0.98)/GPU-hrVaries (CANADA-1) · monthly · Not verified
  • Spot28 vCPU · 120 GB RAM · 850 GB disk$1.08≈ €0.965 (estimated conversion from $1.08)/GPU-hrVaries (CANADA-1) · per minute · Available
  • On-demand28 vCPU · 120 GB RAM · 850 GB disk$1.35≈ €1.21 (estimated conversion from $1.35)/GPU-hrVaries (CANADA-1) · per minute · Available
  • On-demand8-GPU node · 252 vCPU · 1920 GB RAM · 6600 GB disk$1.40≈ €1.25 (estimated conversion from $1.40)/GPU-hrVaries (CANADA-1) · per minute · Limited

NVIDIA A100 SXM4 80GB80 GB VRAM · 2 offers · from $1.36≈ €1.22 (estimated conversion from $1.36) per GPU-hour

Checked hyperstack.cloud

  • Reserved8-GPU node$1.36≈ €1.22 (estimated conversion from $1.36)/GPU-hrUS (US-1) · monthly · Not verified
  • On-demand8-GPU node · 192 vCPU · 960 GB RAM · 16100 GB disk$1.60≈ €1.43 (estimated conversion from $1.60)/GPU-hrUS (US-1) · per minute · Limited

NVIDIA H100 PCIe 80GB80 GB VRAM · 5 offers · from $1.75≈ €1.56 (estimated conversion from $1.75) per GPU-hour

Checked hyperstack.cloud

  • Reserved$1.75≈ €1.56 (estimated conversion from $1.75)/GPU-hrVaries (CANADA-1) · monthly · Not verified
  • Reserved8-GPU node$1.82≈ €1.63 (estimated conversion from $1.82)/GPU-hrVaries (CANADA-1) · monthly · Not verified
  • Spot28 vCPU · 180 GB RAM · 850 GB disk$2.00≈ €1.79 (estimated conversion from $2.00)/GPU-hrVaries (CANADA-1) · per minute · Available
  • On-demand28 vCPU · 180 GB RAM · 850 GB disk$2.50≈ €2.23 (estimated conversion from $2.50)/GPU-hrVaries (CANADA-1) · per minute · Limited
  • On-demand8-GPU node · 252 vCPU · 1440 GB RAM · 6600 GB disk$2.60≈ €2.32 (estimated conversion from $2.60)/GPU-hrVaries (CANADA-1) · per minute · Limited

NVIDIA H100 SXM 80GB80 GB VRAM · 3 offers · from $2.72≈ €2.43 (estimated conversion from $2.72) per GPU-hour

Checked hyperstack.cloud

  • Reserved8-GPU node$2.72≈ €2.43 (estimated conversion from $2.72)/GPU-hrVaries (CANADA-1, US-1) · monthly · Not verified
  • On-demand8-GPU node · 192 vCPU · 1800 GB RAM · 32100 GB disk$3.20≈ €2.86 (estimated conversion from $3.20)/GPU-hrVaries (CANADA-1) · per minute · Limited
  • On-demand8-GPU node · 192 vCPU · 960 GB RAM · 20100 GB disk$3.20≈ €2.86 (estimated conversion from $3.20)/GPU-hrUS (US-1) · per minute · Not verified

NVIDIA L40 48GB48 GB VRAM · 3 offers · from $0.70≈ €0.625 (estimated conversion from $0.70) per GPU-hour

Checked hyperstack.cloud

  • Reserved$0.70≈ €0.625 (estimated conversion from $0.70)/GPU-hrVaries (CANADA-1) · monthly · Not verified
  • Spot28 vCPU · 58 GB RAM · 825 GB disk$0.80≈ €0.715 (estimated conversion from $0.80)/GPU-hrVaries (CANADA-1) · per minute · Available
  • On-demand28 vCPU · 58 GB RAM · 825 GB disk$1.00≈ €0.893 (estimated conversion from $1.00)/GPU-hrVaries (CANADA-1) · per minute · Available

NVIDIA RTX A6000 48GB48 GB VRAM · 3 offers · from $0.35≈ €0.313 (estimated conversion from $0.35) per GPU-hour

Checked hyperstack.cloud

  • Reserved$0.35≈ €0.313 (estimated conversion from $0.35)/GPU-hrVaries (CANADA-1) · monthly · Not verified
  • Spot28 vCPU · 58 GB RAM · 100 GB disk$0.40≈ €0.357 (estimated conversion from $0.40)/GPU-hrVaries (CANADA-1) · per minute · Limited
  • On-demand28 vCPU · 58 GB RAM · 100 GB disk$0.50≈ €0.447 (estimated conversion from $0.50)/GPU-hrVaries (CANADA-1) · per minute · Limited

Prices are per single GPU-hour, excluding VAT. Multi-GPU nodes are divided by their GPU count; the node size is shown with the price type. Each row shows when the price was observed and links to the page it was read from.

Research notes for these offers25
NVIDIA B200 SXM 180GB · Reserved · 8-GPU node
'NVIDIA B200' reserved from $5.10 per GPU-hour (15% below the $6.00 on-demand rate). 8 x $5.10 = $40.8/hr for an 8-GPU VM. Pricing-page 'Reservation Pricing - Starting from' rate per GPU-hour. Reservations are contracts for a number of GPUs over a defined duration at an agreed discount, arranged via the reservation form or support/sales and invoiced monthly; the contract length, minimum GPU count and region that unlock this 'starting from' rate are not published. Availability is not verified. VRAM listed by Hyperstack as 192 GB; listed here as the B200 SXM 180GB (same HGX B200 GPU).
NVIDIA B200 SXM 180GB · On-demand · 8-GPU node
Pricing-page row 'NVIDIA B200' ($6.00 per GPU-hour, VRAM listed as 192 GB); the HGX B200 product page repeats '$6.00/hr on-demand' and 'up to 192 GB HBM3e'. Listed as the B200 SXM 180GB, which is the same HGX B200 GPU (NVIDIA's current HGX B200 datasheet rates it at 180 GB; 192 GB was the originally announced figure). 8-GPU flavor n3-B200-SXM6x8 only: 8 x $6.00 = $48.0/hr per VM; 252 CPUs, 2,048 GB RAM, 1,000 GB root + 40,000 GB ephemeral; flavor card shows no hibernation, snapshots or SR-IOV support. The B200 flavor is also absent from the high-speed networking (SR-IOV) GPU list. B200 did not appear in the live stock feed at 18:05:01, 18:47:08 or 18:56:29 UTC (26 Sep 2026), and the product page's calls to action lead to a reservation form, so on-demand availability is not verified. CPU cores, RAM and local storage are included in the per-GPU rate on GPU flavors (Resource Pricing docs). Included storage is the root disk plus the ephemeral disk. Billed per minute from prepaid credit (pricing-page FAQ, billing-policies docs); taxes added by jurisdiction. Hyperstack lists 'CPUs' per flavor (the pricing page calls them 'pCPUs'); whether these are physical cores or vCPU threads is not stated consistently.
NVIDIA H200 SXM 141GB · Reserved · 8-GPU node
'NVIDIA H200 SXM' reserved from $2.79 per GPU-hour (30% below the $3.99 on-demand rate). 8 x $2.79 = $22.32/hr for an 8-GPU VM. Pricing-page 'Reservation Pricing - Starting from' rate per GPU-hour. Reservations are contracts for a number of GPUs over a defined duration at an agreed discount, arranged via the reservation form or support/sales and invoiced monthly; the contract length, minimum GPU count and region that unlock this 'starting from' rate are not published. Availability is not verified.
NVIDIA H200 SXM 141GB · On-demand · 8-GPU node
NVIDIA H200 SXM 141 GB at $3.99 per GPU-hour. Sold only as the 8-GPU flavor n3-H200-SXM5x8 (CANADA-1): 8 x $3.99 = $31.92/hr per VM. Flavor: 176 CPUs, 1,800 GB RAM, 100 GB root + 48,000 GB ephemeral NVMe; supports SR-IOV network optimization, hibernation and snapshots. Live stock: out of stock at 18:05:01, 18:47:08 and 18:56:29 UTC (26 Sep 2026). CPU cores, RAM and local storage are included in the per-GPU rate on GPU flavors (Resource Pricing docs). Included storage is the root disk plus the ephemeral disk. Billed per minute from prepaid credit (pricing-page FAQ, billing-policies docs); taxes added by jurisdiction. Hyperstack lists 'CPUs' per flavor (the pricing page calls them 'pCPUs'); whether these are physical cores or vCPU threads is not stated consistently.
NVIDIA A100 PCIe 80GB · Reserved
'NVIDIA A100' reserved from $0.95 per GPU-hour (30% below the $1.35 on-demand rate). Pricing-page 'Reservation Pricing - Starting from' rate per GPU-hour. Reservations are contracts for a number of GPUs over a defined duration at an agreed discount, arranged via the reservation form or support/sales and invoiced monthly; the contract length, minimum GPU count and region that unlock this 'starting from' rate are not published. Availability is not verified.
NVIDIA A100 PCIe 80GB · Reserved · 8-GPU node
'NVIDIA A100 NVLink' reserved from $0.98 per GPU-hour (30% below the $1.40 on-demand rate). 8 x $0.98 = $7.84/hr for an 8-GPU VM. Pricing-page 'Reservation Pricing - Starting from' rate per GPU-hour. Reservations are contracts for a number of GPUs over a defined duration at an agreed discount, arranged via the reservation form or support/sales and invoiced monthly; the contract length, minimum GPU count and region that unlock this 'starting from' rate are not published. Availability is not verified.
NVIDIA A100 PCIe 80GB · Spot
Spot VM table 'NVIDIA A100' $1.08 per GPU-hour (20% below the $1.35 on-demand rate). Same hardware as the A100-80G-PCIe flavors (1x shown). No termination notice; no hibernation, snapshots or boot-from-volume. Live stock on 26 Sep 2026: 4 x 1-GPU, 2 x 2-GPU and 1 x 4-GPU spot configurations at 18:05:01, 18:47:08 and 18:56:29 UTC. The docs also list an A100-80G-SXM4 spot flavor, but no rate for it is published. Billed per minute from prepaid credit (pricing-page FAQ, billing-policies docs); taxes added by jurisdiction.
NVIDIA A100 PCIe 80GB · On-demand
Pricing-page row 'NVIDIA A100' ($1.35 per GPU-hour) = flavor family A100-80G-PCIe in CANADA-1. Specs shown for the 1-GPU flavor n3-A100x1 (28 CPUs, 120 GB RAM, 100 GB root + 750 GB ephemeral); 2x, 4x and 8x flavors exist. Live stock on 26 Sep 2026: 6 x 1-GPU and 2 x 2-GPU configurations at 18:05:01 UTC; 6 x 1-GPU, 3 x 2-GPU and 1 x 4-GPU at 18:47:08 and 18:56:29 UTC. CPU cores, RAM and local storage are included in the per-GPU rate on GPU flavors (Resource Pricing docs). Included storage is the root disk plus the ephemeral disk. Billed per minute from prepaid credit (pricing-page FAQ, billing-policies docs); taxes added by jurisdiction. Hyperstack lists 'CPUs' per flavor (the pricing page calls them 'pCPUs'); whether these are physical cores or vCPU threads is not stated consistently.
NVIDIA A100 PCIe 80GB · On-demand · 8-GPU node
Pricing-page row 'NVIDIA A100 NVLink' ($1.40 per GPU-hour) = flavor A100-80G-PCIe-NVLink (n3-A100x8-NVLink): A100 PCIe 80 GB cards with NVLink bridges, listed as the A100 PCIe 80GB. 8-GPU flavor only: 8 x $1.40 = $11.2/hr per VM; 252 CPUs, 1,920 GB RAM, 100 GB root + 6,500 GB ephemeral; SR-IOV network optimization supported; NVMe passthrough available only to contracted customers. Live stock: out of stock at 18:05:01, 18:47:08 and 18:56:29 UTC (26 Sep 2026). CPU cores, RAM and local storage are included in the per-GPU rate on GPU flavors (Resource Pricing docs). Included storage is the root disk plus the ephemeral disk. Billed per minute from prepaid credit (pricing-page FAQ, billing-policies docs); taxes added by jurisdiction. Hyperstack lists 'CPUs' per flavor (the pricing page calls them 'pCPUs'); whether these are physical cores or vCPU threads is not stated consistently.
NVIDIA A100 SXM4 80GB · Reserved · 8-GPU node
'NVIDIA A100 SXM' reserved from $1.36 per GPU-hour (15% below the $1.60 on-demand rate). 8 x $1.36 = $10.88/hr for an 8-GPU VM. Pricing-page 'Reservation Pricing - Starting from' rate per GPU-hour. Reservations are contracts for a number of GPUs over a defined duration at an agreed discount, arranged via the reservation form or support/sales and invoiced monthly; the contract length, minimum GPU count and region that unlock this 'starting from' rate are not published. Availability is not verified.
NVIDIA A100 SXM4 80GB · On-demand · 8-GPU node
NVIDIA A100 SXM 80 GB at $1.60 per GPU-hour. Sold only as the 8-GPU flavor n3-A100-SXM4x8 in US-1 (Texas): 8 x $1.60 = $12.8/hr per VM; 192 CPUs, 960 GB RAM, 100 GB root + 16,000 GB ephemeral; no SR-IOV network optimization on this flavor (not in the high-speed networking GPU list). Live stock: out of stock at 18:05:01, 18:47:08 and 18:56:29 UTC (26 Sep 2026). CPU cores, RAM and local storage are included in the per-GPU rate on GPU flavors (Resource Pricing docs). Included storage is the root disk plus the ephemeral disk. Billed per minute from prepaid credit (pricing-page FAQ, billing-policies docs); taxes added by jurisdiction. Hyperstack lists 'CPUs' per flavor (the pricing page calls them 'pCPUs'); whether these are physical cores or vCPU threads is not stated consistently.
NVIDIA H100 PCIe 80GB · Reserved
'NVIDIA H100' reserved from $1.75 per GPU-hour (30% below the $2.50 on-demand rate). Pricing-page 'Reservation Pricing - Starting from' rate per GPU-hour. Reservations are contracts for a number of GPUs over a defined duration at an agreed discount, arranged via the reservation form or support/sales and invoiced monthly; the contract length, minimum GPU count and region that unlock this 'starting from' rate are not published. Availability is not verified. The H100 PCIe product page repeats 'Starts from $1.75/hour'.
NVIDIA H100 PCIe 80GB · Reserved · 8-GPU node
'NVIDIA H100 NVLink' reserved from $1.82 per GPU-hour (30% below the $2.60 on-demand rate). 8 x $1.82 = $14.56/hr for an 8-GPU VM. Pricing-page 'Reservation Pricing - Starting from' rate per GPU-hour. Reservations are contracts for a number of GPUs over a defined duration at an agreed discount, arranged via the reservation form or support/sales and invoiced monthly; the contract length, minimum GPU count and region that unlock this 'starting from' rate are not published. Availability is not verified. The pricing page shows no reservation link for this row.
NVIDIA H100 PCIe 80GB · Spot
Spot VM table 'NVIDIA H100 PCIe' $2.00 per GPU-hour (20% below the $2.50 on-demand rate; Hyperstack describes spot as a fixed-percentage discount). Same hardware as the on-demand H100-80G-PCIe flavors (1x shown). Spot VMs can be deleted at any time without notice and support no hibernation, snapshots or boot-from-volume; public IPs bill at the standard rate. Live stock on 26 Sep 2026: 5 x 1-GPU and 2 x 2-GPU spot configurations at 18:05:01 UTC; 4 x 1-GPU and 1 x 2-GPU at 18:47:08 and 18:56:29 UTC. A Hyperstack tutorial quotes the same $2.00 for the 8x H100-80G-PCIe-NVLink spot flavor; that rate is not on the pricing page and is not recorded separately. Billed per minute from prepaid credit (pricing-page FAQ, billing-policies docs); taxes added by jurisdiction.
NVIDIA H100 PCIe 80GB · On-demand
Pricing-page row 'NVIDIA H100' ($2.50 per GPU-hour; the H100 PCIe product page states the same rate) = flavor family H100-80G-PCIe in CANADA-1. Specs shown for the 1-GPU flavor n3-H100x1 (28 CPUs, 180 GB RAM, 100 GB root + 750 GB ephemeral); 2x, 4x and 8x flavors exist (8x: 252 CPUs, 1,440 GB RAM, 6,500 GB ephemeral); the high-speed networking docs list H100 PCIe as SR-IOV capable. Live stock on 26 Sep 2026: 3 x 1-GPU configurations at 18:05:01 UTC, out of stock at 18:47:08 UTC, 2 x 1-GPU at 18:56:29 UTC; 0 at 2x, 4x and 8x each time. Because it went in and out of stock between 18:05 and 18:56 UTC, availability is shown as Limited rather than Available. CPU cores, RAM and local storage are included in the per-GPU rate on GPU flavors (Resource Pricing docs). Included storage is the root disk plus the ephemeral disk. Billed per minute from prepaid credit (pricing-page FAQ, billing-policies docs); taxes added by jurisdiction. Hyperstack lists 'CPUs' per flavor (the pricing page calls them 'pCPUs'); whether these are physical cores or vCPU threads is not stated consistently.
NVIDIA H100 PCIe 80GB · On-demand · 8-GPU node
Pricing-page row 'NVIDIA H100 NVLink' ($2.60 per GPU-hour) = flavor H100-80G-PCIe-NVLink (n3-H100x8-NVLink): H100 PCIe 80 GB cards joined by NVLink bridges, so listed as the H100 PCIe 80GB (form factor per flavor docs: 'PCIe + NVLink'). 8-GPU flavor only: 8 x $2.60 = $20.8/hr per VM; 252 CPUs, 1,440 GB RAM, 100 GB root + 6,500 GB ephemeral; SR-IOV network optimization supported. Live stock: out of stock at 18:05:01, 18:47:08 and 18:56:29 UTC (26 Sep 2026); the spot variant was also out of stock. CPU cores, RAM and local storage are included in the per-GPU rate on GPU flavors (Resource Pricing docs). Included storage is the root disk plus the ephemeral disk. Billed per minute from prepaid credit (pricing-page FAQ, billing-policies docs); taxes added by jurisdiction. Hyperstack lists 'CPUs' per flavor (the pricing page calls them 'pCPUs'); whether these are physical cores or vCPU threads is not stated consistently.
NVIDIA H100 SXM 80GB · Reserved · 8-GPU node
'NVIDIA H100 SXM' reserved from $2.72 per GPU-hour (15% below the $3.20 on-demand rate). 8 x $2.72 = $21.76/hr for an 8-GPU VM. Pricing-page 'Reservation Pricing - Starting from' rate per GPU-hour. Reservations are contracts for a number of GPUs over a defined duration at an agreed discount, arranged via the reservation form or support/sales and invoiced monthly; the contract length, minimum GPU count and region that unlock this 'starting from' rate are not published. Availability is not verified. The H100 SXM product page repeats 'starting from $2.72/hour'.
NVIDIA H100 SXM 80GB · On-demand · 8-GPU node
NVIDIA H100 SXM 80 GB at $3.20 per GPU-hour (pricing page and H100 SXM product page). Sold only as the 8-GPU flavor n3-H100-SXM5x8; CANADA-1 variant: 192 CPUs, 1,800 GB RAM, 100 GB root + 32,000 GB ephemeral: 8 x $3.20 = $25.6/hr per VM. Live stock: H100-80G-SXM5 out of stock in CANADA-1 at 18:05:01, 18:47:08 and 18:56:29 UTC (26 Sep 2026). The product page also says clusters of up to 16,384 H100 SXM are 'only available through Hyperstack's Supercloud reservation'. CPU cores, RAM and local storage are included in the per-GPU rate on GPU flavors (Resource Pricing docs). Included storage is the root disk plus the ephemeral disk. Billed per minute from prepaid credit (pricing-page FAQ, billing-policies docs); taxes added by jurisdiction. Hyperstack lists 'CPUs' per flavor (the pricing page calls them 'pCPUs'); whether these are physical cores or vCPU threads is not stated consistently.
NVIDIA H100 SXM 80GB · On-demand · 8-GPU node
NVIDIA H100 SXM 80 GB at $3.20 per GPU-hour; the pricing page publishes one rate with no regional variation. US-1 (Texas) variant of flavor n3-H100-SXM5x8: 192 CPUs, 960 GB RAM, 100 GB root + 20,000 GB ephemeral: $25.6/hr per VM. At 18:05:01, 18:47:08 and 18:56:29 UTC (26 Sep 2026) the live stock feed for US-1 listed only 'H100-80G-SXM5 Spot' (one 8x configuration available) and no on-demand H100 SXM entry, so on-demand availability is not verified. The flavor catalog lists no H100 SXM5 spot flavor; the spot entry appears only in the stock feed. A US-1 spot rate for H100 SXM is not published and is not recorded. US-1 does not support Kubernetes clusters (FAQ). CPU cores, RAM and local storage are included in the per-GPU rate on GPU flavors (Resource Pricing docs). Included storage is the root disk plus the ephemeral disk. Billed per minute from prepaid credit (pricing-page FAQ, billing-policies docs); taxes added by jurisdiction. Hyperstack lists 'CPUs' per flavor (the pricing page calls them 'pCPUs'); whether these are physical cores or vCPU threads is not stated consistently.
NVIDIA L40 48GB · Reserved
'NVIDIA L40' reserved from $0.70 per GPU-hour (30% below the $1.00 on-demand rate). Pricing-page 'Reservation Pricing - Starting from' rate per GPU-hour. Reservations are contracts for a number of GPUs over a defined duration at an agreed discount, arranged via the reservation form or support/sales and invoiced monthly; the contract length, minimum GPU count and region that unlock this 'starting from' rate are not published. Availability is not verified.
NVIDIA L40 48GB · Spot
Spot VM table 'NVIDIA L40' $0.80 per GPU-hour (20% below the $1.00 on-demand rate). Same hardware as the L40 flavors (1x shown). No termination notice; no hibernation, snapshots or boot-from-volume. Live stock on 26 Sep 2026: 4 x 1-GPU, 2 x 2-GPU and 1 x 4-GPU spot configurations at 18:05:01 UTC; 5 / 2 / 1 at 18:47:08 and 18:56:29 UTC. Billed per minute from prepaid credit (pricing-page FAQ, billing-policies docs); taxes added by jurisdiction.
NVIDIA L40 48GB · On-demand
NVIDIA L40 48 GB at $1.00 per GPU-hour, flavor family L40 in CANADA-1. Specs shown for the 1-GPU flavor n3-L40x1 (28 CPUs, 58 GB RAM, 100 GB root + 725 GB ephemeral); the pricing page's 'Max RAM (GB) per GPU' column says 120 GB, which no listed L40 flavor reaches (8x = 464 GB). 2x, 4x and 8x flavors exist. Live stock on 26 Sep 2026: available at 18:05:01 UTC (28 x 1-GPU, 13 x 2-GPU, 5 x 4-GPU, 1 x 8-GPU) and at 18:47:08 and 18:56:29 UTC (26 / 11 / 5 / 1). CPU cores, RAM and local storage are included in the per-GPU rate on GPU flavors (Resource Pricing docs). Included storage is the root disk plus the ephemeral disk. Billed per minute from prepaid credit (pricing-page FAQ, billing-policies docs); taxes added by jurisdiction. Hyperstack lists 'CPUs' per flavor (the pricing page calls them 'pCPUs'); whether these are physical cores or vCPU threads is not stated consistently.
NVIDIA RTX A6000 48GB · Reserved
'NVIDIA A6000' reserved from $0.35 per GPU-hour (30% below the $0.50 on-demand rate). Pricing-page 'Reservation Pricing - Starting from' rate per GPU-hour. Reservations are contracts for a number of GPUs over a defined duration at an agreed discount, arranged via the reservation form or support/sales and invoiced monthly; the contract length, minimum GPU count and region that unlock this 'starting from' rate are not published. Availability is not verified.
NVIDIA RTX A6000 48GB · Spot
Spot VM table 'NVIDIA A6000' $0.40 per GPU-hour (20% below the $0.50 on-demand rate). Same hardware as the RTX-A6000 flavors (1x shown). No termination notice; no hibernation, snapshots or boot-from-volume. Live stock: out of stock at 18:05:01, 18:47:08 and 18:56:29 UTC (26 Sep 2026). Billed per minute from prepaid credit (pricing-page FAQ, billing-policies docs); taxes added by jurisdiction.
NVIDIA RTX A6000 48GB · On-demand
NVIDIA RTX A6000 48 GB at $0.50 per GPU-hour, flavor family RTX-A6000 in CANADA-1. Specs shown for n3-RTX-A6000x1 (28 CPUs, 58 GB RAM, 100 GB root; no ephemeral disk listed for 1x); 2x-8x flavors add 200-1,400 GB ephemeral. Live stock on 26 Sep 2026: out of stock at every GPU count at 18:05:01 and 18:56:29 UTC; one 1-GPU configuration at 18:47:08 UTC. CPU cores, RAM and local storage are included in the per-GPU rate on GPU flavors (Resource Pricing docs). Included storage is the root disk plus the ephemeral disk. Billed per minute from prepaid credit (pricing-page FAQ, billing-policies docs); taxes added by jurisdiction. Hyperstack lists 'CPUs' per flavor (the pricing page calls them 'pCPUs'); whether these are physical cores or vCPU threads is not stated consistently.

Compare these prices with other providers’: H100 prices comparedH200 prices compared.

Plans5

CPU instance5 plans · from $0.35≈ €0.313 (estimated conversion from $0.35) per hour

  • n1-cpu-small (CPU only)4 vCPU · 4 GB RAM · 100 GB storage · VariesChecked hyperstack.cloud$0.35≈ €0.313 (estimated conversion from $0.35)/hr
  • n1-cpu-medium (CPU only)8 vCPU · 8 GB RAM · 100 GB storage · VariesChecked hyperstack.cloud$0.65≈ €0.581 (estimated conversion from $0.65)/hr
  • n1-cpu-large (CPU only)16 vCPU · 16 GB RAM · 200 GB storage · VariesChecked hyperstack.cloud$1.30≈ €1.16 (estimated conversion from $1.30)/hr
  • n1-cpu-xlarge (CPU only)16 vCPU · 32 GB RAM · 100 GB storage · VariesChecked hyperstack.cloud$1.90≈ €1.70 (estimated conversion from $1.90)/hr
  • n1-cpu-xxlarge (CPU only)32 vCPU · 64 GB RAM · 100 GB storage · VariesChecked hyperstack.cloud$3.74≈ €3.34 (estimated conversion from $3.74)/hr
Research notes for these plans5
n1-cpu-small (CPU only)
On-demand CPU-only flavor at $0.35/hr (4 CPUs, 4 GB RAM, 100 GB root disk as listed on the pricing page). No monthly price is published; a 730-hour month at this rate would be $255.50. Flavor docs list it in NORWAY-1 and CANADA-1. Data transfer is free; a public IP adds $0.00672043/hr. Billed per minute from prepaid credit. The pricing page's column header reads 'GPU Cores' for these CPU-only flavors (apparent typo).
n1-cpu-medium (CPU only)
On-demand CPU-only flavor at $0.65/hr (8 CPUs, 8 GB RAM, 100 GB root disk as listed on the pricing page). No monthly price is published; a 730-hour month at this rate would be $474.50. Flavor docs list it in NORWAY-1 and CANADA-1. Data transfer is free; a public IP adds $0.00672043/hr. Billed per minute from prepaid credit. The pricing page's column header reads 'GPU Cores' for these CPU-only flavors (apparent typo).
n1-cpu-large (CPU only)
On-demand CPU-only flavor at $1.30/hr (16 CPUs, 16 GB RAM, 200 GB root disk as listed on the pricing page). No monthly price is published; a 730-hour month at this rate would be $949.00. Flavor docs list it in CANADA-1 with a 200 GB root disk and in NORWAY-1 with a 100 GB root disk; CPU-only flavors are billed per resource, so the NORWAY-1 hourly cost may differ (not published). Data transfer is free; a public IP adds $0.00672043/hr. Billed per minute from prepaid credit. The pricing page's column header reads 'GPU Cores' for these CPU-only flavors (apparent typo).
n1-cpu-xlarge (CPU only)
On-demand CPU-only flavor at $1.90/hr (16 CPUs, 32 GB RAM, 100 GB root disk as listed on the pricing page). No monthly price is published; a 730-hour month at this rate would be $1,387.00. Flavor docs list it in CANADA-1 only, with an additional 512 GB ephemeral disk. Data transfer is free; a public IP adds $0.00672043/hr. Billed per minute from prepaid credit. The pricing page's column header reads 'GPU Cores' for these CPU-only flavors (apparent typo).
n1-cpu-xxlarge (CPU only)
On-demand CPU-only flavor at $3.74/hr (32 CPUs, 64 GB RAM, 100 GB root disk as listed on the pricing page). No monthly price is published; a 730-hour month at this rate would be $2,730.20. Flavor docs list it in CANADA-1 only, with an additional 1,024 GB ephemeral disk and SR-IOV network optimization support. Data transfer is free; a public IP adds $0.00672043/hr. Billed per minute from prepaid credit. The pricing page's column header reads 'GPU Cores' for these CPU-only flavors (apparent typo).

Storage and data transfer

Shared Storage Volume (network SSD block storage, 'Cloud-SSD')
$0.071≈ €0.063 (estimated conversion from $0.071) per GB-month
Block storage · PersistentPublished as $0.000096774 per GB per hour; x 730 h = $0.07065 per GB-month (docs: 'approximately $0.07 per GB/month'). Docs: Shared Storage Volumes - Hyperstack Docs (opens in a new tab)
MoreReplicated across servers; up to 10 volumes per VM; one VM per volume at a time; billed while AVAILABLE, IN-USE or RESERVED. Available in all regions subject to each region's 'volumes' feature flag. Snapshots and the root disks of hibernated VMs are also billed at this Cloud-SSD rate (Resource Pricing docs). Kubernetes CSI driver available.
Object Storage (S3-compatible)
$0.014≈ €0.013 (estimated conversion from $0.014) per GB-month
Object storage · PersistentPublished as $0.000019397 per GB per hour; x 730 h = $0.01416 per GB-month (docs: 'approximately $14.29 per TB per month'). Docs: Object Storage - Hyperstack Docs (opens in a new tab)
MoreCANADA-1 only (endpoint ca1.obj.nexgencloud.io), no geo-replication or regional redundancy, 4+2 erasure coding, no server-side encryption (client-side encryption recommended), ~40 Gb/s aggregate public uplink. No request or transfer fees are listed. Access keys are suspended while a prepaid balance is negative; data is deleted after 30 days of negative balance.
Root disk (local NVMe, included in GPU flavors)
$0.00≈ €0.00 (estimated conversion from $0.00) per GB-month
Local NVMe · PersistentSize fixed by flavor (100 GB on most GPU flavors, 1,000 GB on B200/B300) and cannot be changed after creation. Docs: Resource Pricing - Hyperstack Docs (opens in a new tab)
MoreIncluded at no extra cost on GPU flavors: the Resource Pricing docs say local storage (root and ephemeral disk) is included 'however large the flavor's allocation is' and only GPUs are billed. The Root Disk docs page's own pricing line says customers 'are billed based on the size of the root disk for the virtual machine flavor used', which is literally true only for CPU-only flavors: they bill local storage per GB, folded into their published hourly prices. Persists through stop and hibernation (moved to Cloud-SSD and billed at that rate while hibernated) but is deleted with the VM and lost if the VM is rescheduled after a hardware failure. Large-root-disk ('-bigroot') flavors exist via API.
Ephemeral disk (local NVMe, included in most GPU flavors)
$0.00≈ €0.00 (estimated conversion from $0.00) per GB-month
Local NVMe · Not persistentMounted at /ephemeral; size scales with GPU count (e.g. 750 GB on 1x H100 PCIe, 48,000 GB on 8x H200). Included in the GPU price (Resource Pricing docs). Wiped on hibernation or deletion. Docs: Ephemeral Storage - Hyperstack Docs (opens in a new tab)
Egress (data out)
$0.00≈ €0.00 (estimated conversion from $0.00) per GB
Pricing page lists 'Egress/Ingress traffic: Free' and its FAQ says there are no charges for bandwidth or data transfer; the About page also states 'no egress or ingress fees'. Pricing page: Hyperstack AI Cloud Pricing (On-Demand, Reservation, Spot VM, CPU, Storage, Other Services) (opens in a new tab)
MorePublic IP: $0.00672043 per hour while attached (about $4.91 per 730-hour month), also billed during hibernation if retained. Object storage public transfers share a ~40 Gb/s aggregate uplink in CANADA-1. No fair-use cap is documented.

Before you compare

From Hyperstack’s record: its first stated uses and limitation, how its prices are set, and the providers compared below.

Suited to
  • Single-GPU to 8-GPU VMs on A100 80GB PCIe, H100 PCIe and L40 for fine-tuning and inference, billed per minute with no data-transfer fees
  • 8-GPU H100 SXM (CANADA-1 and US-1) and H200 SXM (CANADA-1) VMs with SR-IOV networking of up to 350 Gbps between VMs for multi-node jobs (8x B200 is also listed in CANADA-1, but its flavor does not support SR-IOV network optimization)
All 4 uses it suits
Main limitation
The only European region is NORWAY-1 (Norway, EEA but not EU), and its only GPU is the RTX A4000; no EU-member-state region is documented, although Hyperstack's security page mentions EU regions
All 10 limitations
Price basis
GPU prices per GPU-hour, billed per minute; egress included; 4 storage options listed
All prices, storage and egress
Alternatives
Nebius provider pageDigitalOcean provider pageOVHcloud provider pageLambda provider page
Compare side by side with HyperstackHow they differ

Who it fits

Based on Hyperstack’s public documentation, checked . We don’t score providers. How we build records

Suited to

  • Single-GPU to 8-GPU VMs on A100 80GB PCIe, H100 PCIe and L40 for fine-tuning and inference, billed per minute with no data-transfer fees
  • 8-GPU H100 SXM (CANADA-1 and US-1) and H200 SXM (CANADA-1) VMs with SR-IOV networking of up to 350 Gbps between VMs for multi-node jobs (8x B200 is also listed in CANADA-1, but its flavor does not support SR-IOV network optimization)
  • Interruptible batch jobs on spot VMs at 20% below on-demand (H100 PCIe, A100 PCIe, L40, RTX A6000)
  • Teams that want a private OpenAI-compatible vLLM endpoint set up by a wizard and billed at the VM's hourly rate

Good fit for

  • Fine-tuningStrong fitSingle-GPU A100 80GB PCIe ($1.35/hr) and L40 ($1.00/hr) VMs were in stock on 26 Sep 2026, and single-GPU H100 PCIe ($2.50/hr) went in and out of stock. Billing is per minute, hibernation pauses GPU charges, and an official Unsloth QLoRA tutorial is published.
  • Batch inferenceStrong fitSpot VMs at 20% below on-demand for H100 PCIe, A100 PCIe, L40 and RTX A6000, free ingress and egress, and vLLM/SGLang tutorials. Spot VMs can be reclaimed without notice, so jobs need checkpointing.
  • TrainingSuitable8-GPU VMs with H100 SXM, H200 SXM, B200 and H100/A100 PCIe-NVLink. SR-IOV networking of up to 350 Gbps between VMs is available on the H100 SXM, H100 PCIe, H100 PCIe-NVLink, H200 and A100 PCIe-NVLink flavors in CANADA-1 and US-1, but not on B200 or A100 SXM4. A Hyperstack tutorial runs a 32-GPU SGLang job across four H100 PCIe-NVLink VMs over Ethernet and states that on-demand nodes are not joined by InfiniBand. All of these 8-GPU on-demand flavors were out of stock or absent from the live stock feed on 26 Sep 2026, and large clusters are sold through reservations.
  • Real-time inferenceSuitableDedicated Inference creates a private, API-key-protected vLLM endpoint on a VM billed at the flavor's hourly rate, and AI Studio offers per-token serverless inference on third-party hosted models. Each endpoint is a single VM in one region; autoscaling is not documented.
Show 3 more
  • RAG applicationsSuitableGPU VMs for self-hosted models, persistent block volumes, S3-compatible object storage (CANADA-1 only) and AI Studio knowledge bases with cited answers; no managed vector database is documented.
  • RenderingSuitableL40 and RTX A6000 GPUs are available by the minute and Hyperstack markets rendering workloads, but only Linux images are supported (no Windows).
  • Private LLM hostingSuitableSelf-hosted models on dedicated GPUs in VMs, private Dedicated Inference endpoints, a published GDPR DPA and a single-tenant Secure Private Cloud option (consultation only). European hosting is limited to Norway (EEA, not EU), whose only GPU is the RTX A4000.

Look elsewhere for

  • AI agentsPartial fitHyperstack publishes OpenClaw and n8n tutorials and offers CPU-only VMs in CANADA-1 and NORWAY-1, but the smallest CPU flavor costs $0.35/hr (4 CPUs, 4 GB RAM) and there is no managed agent-hosting product.

Not sure Hyperstack fits what you want to build?Help me choose

Strengths and limitations

Summarised from the research record for Hyperstack. We don’t score providers.

Strengths

  • Per-GPU on-demand rates are published for 13 GPU types, and CPU, RAM and local NVMe on GPU flavors are included in the GPU rate
  • Per-minute billing, with hibernation that stops GPU charges and bills only the saved root disk, any retained public IP and any attached volumes (restoring needs the flavor to be in stock again)
  • No charge for ingress or egress traffic; a public IPv4 address costs $0.00672043 per hour
  • Live GPU stock per region and GPU count is shown on a public docs dashboard and returned by the /core/stocks API (API key required)
Show 3 more
  • REST API with Python, Go and JavaScript SDKs, a Terraform provider and a hosted MCP server; on-demand Kubernetes with unbilled master nodes
  • Official tutorials cover Ollama, vLLM (single, multi-GPU and Kubernetes), SGLang, n8n, OpenClaw and Unsloth fine-tuning
  • SOC 2 Type 2 attestation (NexGen Cloud) and a published DPA covering EU and UK GDPR with EU Standard Contractual Clauses

Limitations

  • The only European region is NORWAY-1 (Norway, EEA but not EU), and its only GPU is the RTX A4000; no EU-member-state region is documented, although Hyperstack's security page mentions EU regions
  • Most GPU types are only in CANADA-1; US-1 has only 8x H100 SXM and 8x A100 SXM flavors, and object storage exists only in CANADA-1
  • H200, H100 SXM, B200, A100 SXM and the NVLink PCIe variants come only as 8-GPU VMs (for example 8x H200 = $31.92/hr)
  • GPU stock is thin and volatile: in the live stock feed at 18:05, 18:47 and 18:56 UTC on 26 Sep 2026, H200, H100 SXM, the H100 and A100 NVLink flavors, A100 SXM and B300 were out of stock each time, on-demand H100 PCIe (single-GPU only) and RTX A6000 went in and out of stock, and B200 and RTX PRO 6000 never appeared in the feed
Show 6 more
  • Prepaid credit is the default: at zero balance on-demand VMs are hibernated if their flavor supports it, while VMs whose flavor lacks hibernation (spot, large-root-disk and, per its flavor card, B200) and Kubernetes clusters are deleted; reaching $0 does not stop billing, and postpaid accounts are arranged through sales
  • Spot VMs can be terminated without notice and support no hibernation, snapshots or boot-from-volume
  • Snapshots are capped at 100 GB and object storage has no server-side encryption
  • The published SLA is 99.5% monthly uptime, with account-credit rebates that must be claimed within 5 calendar days (Service Level Addendum v1, 15 Aug 2023)
  • Reservation prices are 'starting from' rates with no published term, minimum size or region
  • CPU-only VMs start at $0.35/hr for 4 CPUs and 4 GB RAM, and Windows images are not supported

Infrastructure

Regions

3 regions across 3 countries; none is recorded as an EU region.

Europe (1)
NORWAY-1 (Vestland, Norway)
North America (2)
CANADA-1 (Quebec, Canada); US-1 (Texas, United States)

Source: Docs: Regions & Data Centers - Hyperstack Docs (opens in a new tab)

Region notes3
CANADA-1 (Quebec, Canada)
Hyperstack's region documentation describes this region as sustainably powered (green); AIHostCompare has not verified that claim. SR-IOV high-speed networking in network-optimized environments. Hosts most GPU flavors: B300, B200, H200 SXM, H100 PCIe / PCIe-NVLink / SXM5, A100 PCIe / PCIe-NVLink, RTX PRO 6000 SE, L40 and RTX A6000, plus CPU flavors, Kubernetes and the only object storage. Live stock on 26 Sep 2026: 6 of 14 GPU models in stock at 18:05, 18:47 and 18:56 UTC, although the mix changed between those times (on-demand H100 PCIe and RTX A6000 flipped). Status page 'Canada 1 Datacentre': Operational, 100% uptime over 90 days. The status page also lists a 'Canada 2 Datacentre' (100%) that is not documented as a deployable region.
US-1 (Texas, United States)
Standard-energy region with SR-IOV networking. GPU flavors: 8x H100 SXM5 and 8x A100 SXM4 only; no CPU-only flavors; Kubernetes not supported (FAQ); volumes supported since 2025. Live stock on 26 Sep 2026 at 18:05, 18:47 and 18:56 UTC: A100 SXM4 out of stock, one 8x H100 SXM5 spot configuration available. Status page 'US 1 DataCenter': Operational, 99.92% uptime over 90 days.
NORWAY-1 (Vestland, Norway)
Europe, but outside the EU (Norway is an EEA member, so GDPR applies). Hyperstack's region documentation describes it as sustainably powered (not verified by AIHostCompare); no high-speed networking. GPU flavors: RTX A4000 only (1x-10x; not in the AIHostCompare catalog, so no offers are listed); CPU flavors small/medium/large; Kubernetes supported. Live stock on 26 Sep 2026 at 18:05, 18:47 and 18:56 UTC: RTX A4000 out of stock. Status page 'Norway 1 Datacentre': Operational, 100% uptime over 90 days.

Frameworks

How Hyperstack works with common AI serving, agent and training frameworks. Support levels, strongest first: an official template, documentation, a community guide, or manual setup on a general-purpose machine.

Hosting

Support for running the framework itself on Hyperstack.

vLLM hosting providers
Official template
The Dedicated Inference 'Deploy AI' wizard (released 24 Sep 2026) provisions a GPU VM running a pinned vLLM version with an API-key-protected OpenAI-compatible endpoint. Docs: Dedicated Inference - Hyperstack Docs (opens in a new tab)
MoreOfficial docs tutorials also cover single-GPU, multi-GPU and Kubernetes vLLM.
n8n
Documented setup
Hyperstack tutorial (updated 17 Feb 2026) runs the n8n container alongside vLLM on a 2x A100 PCIe VM; AI Studio docs also describe n8n as an integration of the serverless API. Docs: How to Self-Host n8n with Llama 3 on a Cloud GPU Server - Hyperstack (opens in a new tab)
Ollama hosting providers
Documented setup
Official docs tutorial runs the ollama/ollama container on a GPU VM with a CUDA + Docker image; companion Open WebUI tutorial. Docs: Running Ollama - Hyperstack Docs (opens in a new tab)
OpenClaw hosting providers
Documented setup
Hyperstack tutorial (updated 26 Aug 2026) installs OpenClaw with its Docker setup script on a VM and adds Tailscale, UFW and Docker sandboxing; no template. Docs: Securing OpenClaw on Hyperstack: Safe AI Agent Deployment (opens in a new tab)
PyTorch
Documented setup
No preinstalled PyTorch image; CUDA images are provided and the official Unsloth fine-tuning tutorial installs PyTorch via pip. Docs: Fine-Tuning LLMs with Unsloth - Hyperstack Docs (opens in a new tab)
SGLang
Documented setup
Hyperstack tutorials (e.g. Kimi K3 across four 8x H100 PCIe-NVLink VMs, updated 26 Aug 2026) run SGLang containers (lmsysorg/sglang:kimi-k3 image, over Ethernet with NCCL_IB_DISABLE=1); no one-click template. Docs: Deploy Kimi K3 on GPU Cloud for Multi-Node 2.8T Inference - Hyperstack (opens in a new tab)
Dify
Manual setup
No Hyperstack guide found (as of 26 Sep 2026). Docker-enabled Ubuntu images allow a self-managed Docker Compose install (not tested). Docs: Operating System Images - Hyperstack Docs (opens in a new tab)
Flowise
Manual setup
No Hyperstack guide found (as of 26 Sep 2026). Could run as a container on a Docker-enabled VM (not tested). Docs: Operating System Images - Hyperstack Docs (opens in a new tab)
Hermes Agent
Manual setup
No Hyperstack guide found (as of 26 Sep 2026). Could be self-installed on a Linux VM with root and Docker (not tested). Docs: Operating System Images - Hyperstack Docs (opens in a new tab)
LangGraph
Manual setup
No Hyperstack LangGraph deployment tutorial found as of 26 Sep 2026; a general Hyperstack blog explainer on agentic AI frameworks mentions LangGraph.
MoreAs a Python library it can run on any VM or against the OpenAI-compatible endpoints (not tested).
Text Generation Inference
Manual setup
No Hyperstack TGI deployment guide found (as of 26 Sep 2026; Hyperstack articles mention TGI only as a compatible engine); TGI's container could run on the CUDA + Docker images (not tested). Docs: Operating System Images - Hyperstack Docs (opens in a new tab)

Hyperstack is not listed on the Hermes Agent hosting page. That page lists providers that host the framework with an official template, documentation or a community guide.

Operations

Managed service: partial. Each feature below states what the provider documents, with the source it was checked against. “Not verified” means no reliable source was found.

Features

Compute and access

Custom containers (Docker)
Yes
Ubuntu 22.04/24.04 CUDA images are offered 'with Docker'; official tutorials state these include the NVIDIA Container Toolkit. Docs: Operating System Images - Hyperstack Docs (opens in a new tab)
Root access
Yes
Full Linux VMs; tutorials log in as 'ubuntu' with sudo, and the FAQ documents root access via recovery mode. Windows is not supported. Docs: FAQ - Hyperstack Docs (opens in a new tab)
SSH access
Yes
Docs: Virtual Machine Overview - Hyperstack Docs (opens in a new tab)
API access
Yes
REST API (Infrahub) with an equivalent for every console action; the docs home lists Python, Go and JavaScript SDKs and a Terraform provider, and the release notes add a hosted MCP server (26 Aug 2026). Docs: Virtual Machine Overview - Hyperstack Docs (opens in a new tab)
Jupyter notebooks
Yes
Optional Jupyter Notebook auto-install at deployment (port 8888, needs a public IP). Docs: Virtual Machine Overview - Hyperstack Docs (opens in a new tab)
Templates
Partial
Pre-built OS images (NVIDIA driver/CUDA, optional Docker), saved provisioning profiles and the Dedicated Inference wizard; no application marketplace (the legacy Marketplace template and deployment APIs were removed on 17 Sep 2026). Docs: Release Notes - Hyperstack Docs (opens in a new tab)
Multi-GPU instances
Yes
Flavors up to 8 GPUs per VM (10 for RTX A4000); A100/H100/H200/B200/B300 SXM and the PCIe-NVLink flavors come only as 8-GPU VMs. Docs: Flavors - Hyperstack Docs (opens in a new tab)
MoreSR-IOV networking of up to 350 Gbps between VMs applies only to network-optimized flavors (A100 PCIe-NVLink, H100 PCIe, H100 PCIe-NVLink, H100 SXM5, H200 and B300 per the high-speed networking docs) in network-optimized environments in CANADA-1 and US-1; B200 and A100 SXM4 flavors are not network-optimized. The same docs say SR-IOV works only with the Ubuntu 22.04 R535 CUDA 12.2 image, which conflicts with B300 accepting only the R595 image.
Bare metal
No
Self-serve compute is virtual machines only. Secure Private Cloud (single-tenant dedicated hardware, consultation only) does not state that it is bare metal. Docs: Virtual Machine Overview - Hyperstack Docs (opens in a new tab)
Kubernetes
Yes
On-demand clusters in CANADA-1 and NORWAY-1 only (FAQ); US-1 does not support clusters. Docs: Kubernetes - Hyperstack Docs (opens in a new tab)
MoreWorker nodes are billed at flavor rates; master, load balancer and bastion nodes are not billed (Resource Pricing docs). CSI driver for persistent volumes. Clusters cannot be hibernated (delete to stop billing), and public IPs on bastion and load-balancer nodes are billed separately (FAQ).
Show all 21 features

Management

Managed service
Partial
Core product is self-managed VMs. Docs: Dedicated Inference - Hyperstack Docs (opens in a new tab)
MoreManaged elements: on-demand Kubernetes (master nodes not billed), the Dedicated Inference wizard (vLLM endpoint on a VM in your account) and AI Studio serverless inference.
Serverless
Partial
AI Studio offers serverless, OpenAI-compatible inference billed per 1M tokens, now only on third-party hosted models (Hyperstack-hosted models and fine-tuning were removed on 25 Aug 2026). Docs: AI Studio Billing - Hyperstack Docs (opens in a new tab)
MoreThere is no serverless GPU container or function product; GPU compute is VM-based.

Storage and data

Persistent storage
Yes
Shared Storage Volumes (block, persist independently of VMs) in all regions; S3-compatible object storage in CANADA-1. Docs: Shared Storage Volumes - Hyperstack Docs (opens in a new tab)
Snapshots
Yes
VM snapshots (incremental after the first), max 100 GB, billed at the Cloud-SSD rate on consumed storage; not available on spot, large-root-disk, Kubernetes-node, boot-from-volume, Dedicated Inference or B200 VMs. Docs: Snapshots - Hyperstack Docs (opens in a new tab)
Managed backups
Partial
No scheduled or managed backup service is documented; manual snapshots (100 GB cap) and copying data to volumes or object storage are the documented options. Docs: Snapshots - Hyperstack Docs (opens in a new tab)

Billing

Per-second billing
No
Billing is per minute; rates are quoted per hour. Docs: Accounts, Billing Policies, and Usage Monitoring - Hyperstack Docs (opens in a new tab)
Spot pricing
Yes
Spot VMs for H100 PCIe, RTX Pro 6000 SE, A100 PCIe, L40 and RTX A6000 at 20% below on-demand; reclaimable without notice. Pricing page: Hyperstack AI Cloud Pricing (On-Demand, Reservation, Spot VM, CPU, Storage, Other Services) (opens in a new tab)
Reserved pricing
Yes
'Starting from' reservation rates for 12 GPU models; contract terms not published. Pricing page: Hyperstack AI Cloud Pricing (On-Demand, Reservation, Spot VM, CPU, Storage, Other Services) (opens in a new tab)

Regions and compliance

EU region
No
No EU region is documented: the only European region is NORWAY-1 (Vestland, Norway), an EEA member where GDPR applies, but not an EU member state. Docs: Regions & Data Centers - Hyperstack Docs (opens in a new tab)
MoreIts only GPU flavor is the RTX A4000 (not in the AIHostCompare catalog, out of stock at 18:05 UTC on 26 Sep 2026), plus CPU flavors. Hyperstack's security page mentions 'data centre regions in the EU', but no EU-member-state region appears in the docs, flavor catalog, stock feed or status page.
SOC 2 report
Yes
SOC 2 Type 2 attestation held by parent company NexGen Cloud; ISO 27001 and HIPAA listed as upcoming (not achieved). Docs: Secure By Design - Hyperstack (opens in a new tab)
GDPR data processing agreement
Yes
Published DPA template covering EU GDPR and UK GDPR, with EU SCCs (module 3) plus the UK Addendum for restricted transfers; the sub-processor schedule is blank in the published template. Docs: NexGen Cloud Limited Data Processing Agreement (opens in a new tab)
Customer-managed keys (BYOK)
No
No customer-managed or customer-supplied key option is documented. Docs: Object Storage - Hyperstack Docs (opens in a new tab)
MoreObject storage explicitly lacks server-side encryption (client-side encryption recommended). The security page claims encryption at rest without describing key management.

SLA and support

Uptime SLA
99.5% uptime
NexGen Cloud Limited Service Level Addendum, Version 1 - 15 August 2023: minimum 99.5% monthly uptime for GPU Processing, CPU Processing, Memory and Storage Capacity, excluding scheduled maintenance, force majeure and customer-caused issues; rebates are pro-rata account credits capped at the fees for the affected service, claimed by email within 5 calendar days of the billing cycle; the customer may terminate after three consecutive months below target. Docs: NexGen Cloud Limited Service Level Addendum (opens in a new tab)
MoreThe SLA page also carries a hidden copy of the addendum that states 100.0%; only the visible 99.5% is used. Weekly platform maintenance windows are posted on the status page. The docs' '99.982% annual uptime' refers to Tier 3 data-center design, not a contractual SLA.
Support channels
4 listed
  • Support Center (Zendesk): tickets and live chat
  • Email: support@hyperstack.cloud
  • Sales: sales@hyperstack.cloud, +44 (0) 203 475 3402
  • Status page with email, SMS, Slack, Teams, webhook and RSS subscriptions
Premium postpaid accounts include 'prioritized support' (billing-policies docs). Docs: Support - Hyperstack Docs (opens in a new tab)
MoreThe H100 SXM page advertises 24/7 assistance for Supercloud (reservation) customers. Response-time targets and support tiers are not published.

Sources and verification

42 sources, accessed between 26 Sep 2026 and 20 Oct 2026. Official provider pages are listed first.

How we verify

Prices, regions, storage, features, framework support, SLA and support channels link to the page they were read from, and every price shows when it was checked. Fit labels, strengths and limitations summarise the same research record. The Hyperstack record was last checked on 26 Sep 2026. A check date older than 7 days is marked “Stale” with an amber dot, and one older than 30 days “Needs refresh”.

Last verified
26 Sep 2026, 19:00 UTC
Checked · Stale
Method
Source-by-source review by AI agents, independently re-checked
Confidence
High
Checked by
AI research agent; every value re-checked against its source by a separate AI agent

Verification notes

On 26 September 2026 we checked every price, VRAM figure and flavor spec in this listing against Hyperstack's live pricing pages, documentation and stock dashboard. Its resource-pricing docs still say resources are billed for every hour they exist; the listing follows the pricing FAQ and billing docs, which state per-minute billing. Availability comes from three readings of the stock feed that day, and stock changes often. Later corrections are listed on the methodology page.

Commercial relationship

AIHostCompare has no active affiliate link with Hyperstack; the Visit provider link goes to its official website. Commissions never change ordering, alternatives, fit labels or conclusions. Read the affiliate disclosure.

Pricing

  1. 1.Hyperstack AI Cloud Pricing (On-Demand, Reservation, Spot VM, CPU, Storage, Other Services) (opens in a new tab) · Official pricing page · accessed
  2. 2.NVIDIA RTX Pro 6000 SE - Hyperstack (opens in a new tab) · Official pricing page · accessed
  3. 3.Rent NVIDIA H100 PCIe GPU - Hyperstack (opens in a new tab) · Official pricing page · accessed
  4. 4.Rent NVIDIA H100 SXM - Hyperstack (opens in a new tab) · Official pricing page · accessed
  5. 5.Rent NVIDIA HGX B200 GPU - Hyperstack (opens in a new tab) · Official pricing page · accessed

Documentation

  1. 6.2025 Release Notes - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  2. 7.About Hyperstack (opens in a new tab) · Official docs · accessed
  3. 8.AI Studio - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  4. 9.AI Studio Billing - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  5. 10.Contracts - Hyperstack Docs (opens in a new tab) · Official docs · accessed
Show all 42 sources

Documentation

  1. 11.Dedicated Inference - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  2. 12.Deploy Kimi K3 on GPU Cloud for Multi-Node 2.8T Inference - Hyperstack (opens in a new tab) · Official docs · accessed
  3. 13.Ephemeral Storage - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  4. 14.FAQ - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  5. 15.Fine-Tuning LLMs with Unsloth - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  6. 16.Flavors - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  7. 17.GPU Stock Information (live GPU stock dashboard) - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  8. 18.High-Speed Networking w/ SR-IOV - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  9. 19.How to Self-Host n8n with Llama 3 on a Cloud GPU Server - Hyperstack (opens in a new tab) · Official docs · accessed
  10. 20.Hyperstack Documentation (home) (opens in a new tab) · Official docs · accessed
  11. 21.Kubernetes - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  12. 22.Object Storage - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  13. 23.Operating System Images - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  14. 24.Public IPs and Internet Accessibility - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  15. 25.Regions & Data Centers - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  16. 26.Release Notes - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  17. 27.Resource Pricing - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  18. 28.Root Disk - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  19. 29.Running Ollama - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  20. 30.Running vLLM - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  21. 31.Secure By Design - Hyperstack (opens in a new tab) · Official docs · accessed
  22. 32.Secure Private Cloud - Hyperstack (opens in a new tab) · Official docs · accessed
  23. 33.Securing OpenClaw on Hyperstack: Safe AI Agent Deployment (opens in a new tab) · Official docs · accessed
  24. 34.Shared Storage Volumes - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  25. 35.Snapshots - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  26. 36.Spot VMs - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  27. 37.Support - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  28. 38.Virtual Machine Overview - Hyperstack Docs (opens in a new tab) · Official docs · accessed

Terms & policies

  1. 39.Accounts, Billing Policies, and Usage Monitoring - Hyperstack Docs (opens in a new tab) · Official docs · accessed
  2. 40.NexGen Cloud Limited Data Processing Agreement (opens in a new tab) · Official docs · accessed
  3. 41.NexGen Cloud Limited Service Level Addendum (opens in a new tab) · Official docs · accessed

Status

  1. 42.Hyperstack Service Status (opens in a new tab) · Official status page · accessed

Alternatives to Hyperstack

Providers that share a hosting category with Hyperstack, the most relevant first, spread so that every provider is listed on several provider pages. Not a ranking.

Also sharing categories: Lambda profile

Compare all 4 side by side

Related comparisons and guides

Categories that list Hyperstack