Discover how much you could save running AI workloads locally on AMD Client devices

Start with a quick estimate using the preset or custom values. Then drill into advanced options to customize even more to fit your situation. In either case, share your results and pull in your AMD rep to discuss best options.

AMD AI Token
How to Use To use this calculator, enter your team size, analysis period, API workload, and choose the best Cloud API model. The calculator will then show you your estimated costs and recommend best fit for your infrastructure.
Provide basic information to determine potential AI savings.

1 Your AI usage

25 seats

Choose the period of time for this analysis.

Allocate the percent of users to different workloads. Total must equal 100%.

Usage listed is tokens per user per day.

AI Office Worker Coding & Advanced Reasoning
Light574K input + 57K output
%
Medium5.74M input + 574K output
%
Heavy16.7M input + 1.67M output
%
CustomSet your own
%
Input (MTok/day) Output (MTok/day)

Enter the % of users per model (must total 100%). Pricing is input/output per MTok per month.

Note that models listed are for pricing assessment only and not used for AI performance comparison in the configuration recommendations.

Gemini Pro$2 / $12 MTok
%
Claude Sonnet$3 / $15 MTok
%
Claude Opus$5 / $25 MTok
%
GPT 5.5$5 / $30 MTok
%
CustomSet your own
%

Cloud cost is calculated from API token consumption only. No seat, subscription, or flat monthly cloud price is used.

Pricing based on publicly available data as of July 2026. Claude Sonnet includes September 2026 pricing change. Prices subject to change.

How much AI work runs locally on AMD? (0% = all cloud, 100% = all local)

0% 100%

Move right to reduce cloud spend. Move left to preserve access to cloud model features for complex tasks.

🖥️ AMD hardware configuration

Each device uses its own specs for throughput, power, and runtime. Enter units to deploy — a flag appears if the combined fleet falls short of your token demand.

Changing values updates Your Results.

Ryzen™ AI 9 HX 470Laptop · iGPU NPU
⚠️ Combined fleet capacity falls short of total token demand.
Ryzen AI Max+ 395Workstation laptop · large iGPU
⚠️ Combined fleet capacity falls short of total token demand.
Radeon AI PRO R9700Desktop / workstation · dGPU
⚠️ Combined fleet capacity falls short of total token demand.

⚡ Energy & operations

Set how long devices stay powered on. This drives your electricity cost estimate, which may differ from active AI usage hours.

☁️ Cloud API inputs

Each column below corresponds to a Cloud AI model you assigned users to above. All values are prefilled from those quantities — edit any field to override. The hybrid mix % sets how many of that model's users are served locally instead.

🧮 Cloud pricing modifiers

Prompt caching, batch, regional and long-context logic from the math concept.

Every default is illustrative for demonstration and must be validated against the latest AMD product data and current cloud pricing.

2 Your results

Total cost includes Cloud API costs (for Cloud and Hybrid), hardware and power costs (for Local and Hybrid) where appropriate.

Projected savings at 50% Hybrid
$—
Cloud only vs Hybrid AI.
Max potential savings (100% Hybrid)
$—
Cloud only vs Local AI.
Local only
3-yr total cost
$—
Monthly cost (included in total cost)
$—
Hybrid at 50% mix
3-yr total cost
$—
Monthly cost (included in total cost)
$—
Cloud only
3-yr total cost
$—
Input $— + output $—
Monthly cost (included in total cost)
$—
Break-even — Local vs Cloud
—
to recover hardware
Break-even — Hybrid vs Cloud
—
hybrid payback
Cumulative cash spend over 36 months Cloud Local Hybrid
CapEx vs. OpEx — recommended path (3 years)
CapEx $—OpEx $—
3-yr savings
$—
Local advantage vs. cloud

Beyond cost — why teams move local

✓ Data stays in-house ✓ Predictable latency ✓ Offline / air-gapped ✓ No per-token metering ✓ No vendor lock-in

Beyond cost — why teams move local

✓ Data stays in-house✓ Predictable latency✓ Offline / air-gapped✓ No per-token metering✓ No vendor lock-in
Your AMD configuration
1 ×AMD Ryzen™ AI Max+ 395

Purchase 1 unit now — one device for the local user.

Per-user capacity is calculated from separate input processing and output generation time, with a hard ceiling of one device per user.

Note: AI model performance will vary by platform so you will need to ensure that the models you want to run locally will meet your performance needs. This calculator does not consider your performance needs.

What to do next

—

—

Disclaimers and notes
  • AI Workload is entered as input and output tokens per person per day across four tiers: Light (574K input / 57K output), Baseline (5.74M input / 574K output), Heavy (16.7M input / 1.67M output), and Custom (user-defined in MTok/day). Monthly cloud API volume = users per tier × daily tokens × active days per month.
  • Cloud AI model pricing is based on publicly available data as of July 2026. Prices subject to change. The calculator supports multiple models simultaneously — each with its own user count and price. A weighted average is applied across models for blended cost calculations. Supported models: OpenAI GPT 5.5 ($5/$30 per MTok), Claude Sonnet ($3/$15), Claude Opus ($5/$25), Gemini Pro ($2/$12), and Custom (user-defined). Pricing is input/output per MTok per month.
  • Local (AMD) capacity is evaluated per user per workload tier. Each tier's token demand must fit within one device's daily inference window (hours/day × throughput). The calculator assigns the lowest-cost AMD device that fits each tier. Available devices: AMD Ryzen™ AI 9 HX 470, AMD Ryzen™ AI Max+ 395, AMD Radeon™ AI PRO R9700. Custom workload users whose demand exceeds all local device capacities are routed to cloud API regardless of Hybrid Mix setting.
  • Hardware costs include upfront purchase price and ongoing electricity costs (watts × hrs/day × active days × $/kWh). Hardware is sized for current demand. Hardware life is set to the selected analysis period — full cost is reflected within the chosen window.
  • Hybrid deploys AMD hardware for all users, then splits monthly token demand by the Hybrid Mix %: that percentage of tokens is processed locally on AMD devices, and the remainder is sent to cloud API. Example: at 50% hybrid mix, all 25 users have an AMD device, but half their monthly token volume runs locally and half goes to cloud. Custom-workload users whose per-person demand exceeds local device capacity are always routed to cloud API regardless of the Hybrid Mix setting.
  • Projected Cloud API usage growth compounds cloud token volumes year over year but does not affect AMD hardware sizing or selection. (Currently set to 0% — this feature will be re-enabled in a future update.)
  • Break-even is shown for two scenarios: (1) Local AI vs Cloud Only — first month cumulative local AMD spend falls below cloud-only spend; (2) Hybrid vs Cloud Only — first month cumulative hybrid spend falls below cloud-only spend. In Requires Hybrid scenarios, the Local break-even is not applicable.
  • Max potential savings is the larger of (Cloud Only − Local AI) or (Cloud Only − Hybrid). In Requires Hybrid scenarios, it reflects the Hybrid savings.
  • Excluded from this calculator: model inference quality differences, software licensing, IT management costs, migration effort, taxes, financing fees, network/egress costs (unless entered as API uplift), and provider-specific volume discounts. AI model performance will vary by platform — ensure models you want to run locally meet your performance needs before deployment.
  • Input/Output throughput test parameters: Qwen 3.6 35B A3B, Q4 K M, 262000 context window, pre-filled context to 128,000, sustained throughput measurement, Vulkan llama.cpp.

All values are illustrative estimates based on publicly available data · no live pricing calls are made · validate against current AMD product specifications and cloud provider pricing before making purchasing decisions.

Disclaimer: All claims generated by this calculator are estimates. The information provided here is for information purposes only. AMD DISCLAIMS ALL WARRANTIES, implied warranties or guarantees of any kind, including but not limited to the suitability or fitness of any product mentioned here for any purpose. This tool is intended to illustrate the estimated amounts and comparable differences. Prices are sourced in US dollars. The tool does not constitute an offer to buy nor sell any product shown. IN NO EVENT SHALL AMD BE LIABLE FOR ANY DAMAGES, WHETHER THOSE DAMAGES ARE DIRECT, CONSEQUENTIAL, INCIDENTAL, OR SPECIAL, FLOWING FROM THE USE OF OR INABILITY TO USE THE TOOL OR INFORMATION PROVIDED HEREWITH OR RESULTS OF THE TOOL'S USE EVEN IF AMD HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.

—

Cost Calculation Detail

AMD Tokenomics Calculator — illustrative cost breakdown.

Illustrative defaults. Token pricing per publicly available data as of July 2026. Prices subject to change. Power: $0.15/kWh assumed. Validate against current AMD product specs and live API pricing before making decisions.