[ PRICING ]

SmartTec pricing.

Simple, transparent pricing for battery-backed NVIDIA and Cerebras compute. Fixed-price proposals in 48 hours.

Get a quote in 48h

[ ON-DEMAND NVIDIA ]

GPUs by the hour.

Spin up, train, tear down. Spot / preemptible available.

Wondering whether these are good rates? We publish the whole market, not just ours — B200 capacity trades between $3.50 and $14.24/GPU-hr across 13 providers, at a $6.12 median. See the July 2026 market rates →

NVIDIA H100 / H200 8-GPU node
[ H100 / H200 ]
Hopper · SXM5
NVIDIA HGX B200 8-GPU baseboard
[ B200 ]
Blackwell · HGX
NVIDIA GB200 NVL72 rack
[ GB200 ]
Blackwell · NVL72
GPU
Price
Note
Reserve
H100 80GB
$2.40/GPU-hr
Spot: $1.20
H200 141GB
$3.20/GPU-hr
Spot: $1.60
B200
$4.80/GPU-hr
Q4 2026
GB200
$6.40/GPU-hr
Q4 2026
L40S
$1.60/GPU-hr
Best for batch

[ RESERVED CLUSTERS · H100 ]

Dedicated nodes, on independent power.

Yours alone, monthly or annual. Discounts on prepay.

Cluster
Monthly
Annual
Prepay discount
1 node (8×H100)
$12,800
$138,240
10% prepay
4 nodes (32×H100)
$46,080
$497,664
15% prepay
16 nodes (128×H100)
$174,080
$1,879,000
20% prepay
1× B200 (reserved)
$3,240
$34,992
$4.50/GPU-hr · Q4 2026
5× B200 (anchor block)
$16,200
$174,960
$4.50/GPU-hr · Q4 2026

[ CEREBRAS INFERENCE ]

The fastest tokens on earth.

Cerebras CS-3 system
[ CEREBRAS CS-3 ]
WSE-3 · 4 trillion tx
Cerebras CS-3 in production rack
[ IN PRODUCTION ]
2× CS-3 per rack · 25 kW each
Inference endpoint
$0.04/1K tokens
Bursty workload
Dedicated CS-3 system
$48,000/month
Committed inference
Wafer-scale cluster
Custom
Multi-system capacity

[ COLOCATION · ENTERPRISE ]

Bring your hardware.

Colocate onto our battery-backed power, or we'll build dedicated MW for you. Starting at 100kW. PPA, operating lease, or CAPEX structures available.


[ FAQ ]

Pricing questions.

01How is pricing structured?
On-demand by the GPU-hour, or reserved by node/cluster with prepay discounts. H100 on-demand starts at $2.40/GPU-hr (spot from $1.20). Reserve now for launch pricing.
02What GPUs are available on-demand?
H100, H200, and B200 on-demand at launch; GB200 as allocation opens. Spot / preemptible available at significant discount.
03What about Cerebras?
Dedicated wafer-scale systems for committed inference workloads. Inference endpoints billed per-token.
04Can I prepay for a discount?
Yes — multi-year and pre-paid contracts receive tiered discounts. 10% off annual prepay, 20% off 3-year prepay.
05Do you offer colocation pricing?
Yes — starting at 100kW. We'll size the right configuration during scoping.
06How do I cancel?
On-demand has no commitment. Reserved clusters are typically 12-month terms with 90-day cancellation notice. No early termination fees on monthly plans.

Reserve launch pricing.

Lock in early-partner pricing before Q4 2026 power-on. Three design-partner slots open.

Get a quote