Pricing 價格方案
Every deployment is sized to your workload. Get a quote from our sales team, or estimate your on-prem vs cloud TCO below.每套部署都依您的工作負載規劃。向業務索取正式報價，或在下方試算地端與雲端的總持有成本。
Price breakdown 價格明細
Hardware 硬體 One-time purchase 一次性採購
Software 軟體 Annual subscription 年訂閱
TCO Explorer TCO 試算
What will on-prem vs cloud cost you? 想知道地端與雲端的成本差異？
Compare the total cost of running the same AI inference workload on-prem with aiDAPTIV, on rented cloud GPUs, or through a per-token cloud API. 比較同一個 AI 推論工作負載，在地端 aiDAPTIV、雲端 GPU 租用、雲端 API 按 token 計費三種方案下的總持有成本。
- Three deployment routes side by side 三種部署方案並排比較
- Break-even month and cost multiple 地端回本月數與成本倍數
- Email the estimate to yourself 估算結果可寄到您的信箱
TCO Explorer · Three-way comparison TCO Explorer · 三方案比較
On-prem 地端 aiDAPTIV vs Cloud GPU雲端機器 vs Cloud API雲端 API
Benchmarked on Qwen3.5-397B FP4: for the same workload, compare the total cost over the analysis period of on-prem PRO6000×8 with aiDAPTIV (purchased outright), AWS cloud GPU rental (g7e.48xlarge), and per-token cloud API pricing. Drag the parameters to find the most economical route. 以 Qwen3.5-397B FP4 實測效能為基準：同一工作負載下，比較地端 PRO6000×8 + aiDAPTIV 買斷、AWS 雲端 GPU 機器租用（g7e.48xlarge）、與雲端 API 按 token 計費三種方案在分析期間的總成本。拖曳以下參數，找到最划算的部署路線。
≈ 200 employees (about 1 in 4 active at the same time) 約對應 200 位員工（以約 1/4 同時使用估算）
On-prem break-even 地端回本時點
On-prem aiDAPTIV 地端 aiDAPTIV
Cost breakdown 計算明細
| Option方案 | Scale規模 | Basis計費方式 | vs On-prem相對地端 | |
|---|---|---|---|---|
Cloud API is billed per token on daily volume. Cloud machines bill by daily operating hours. On-prem is a one-time purchase plus electricity; maintenance is excluded. 雲端 API 依每日 token 用量計費；雲端機器以每日運轉時數計費；地端為一次性買斷 + 電費，未含維運。
Estimates are based on internal assumptions and are not a price quote. 以上為依內部假設的估算，非正式報價。
Need numbers for your exact SKU? 需要依實際 SKU 的報價？
Get exact price quote 取得正式報價Assumptions 假設參數
Workload 工作負載
- Benchmark model 效能基準模型
- Qwen3.5-397B FP4
- Input tokens / request Input tokens / 請求
- 32K
- Output tokens / request Output tokens / 請求
- 1K
Per-machine capacity = 3600 ÷ time per request × concurrent sessions × 2. Machines needed = the larger of the session-based and request-based counts. 單機處理能力 = 3600 ÷ 每請求耗時 × 併發數 × 2。所需台數取「依連線數」與「依請求數」兩者較大值。
Performance 效能假設
- Concurrent sessions per machine w/ aiDAPTIV 每台可支援 Concurrent sessions（aiDAPTIV）
- 36
- TTFT (time to first token) w/ aiDAPTIV TTFT（首個 token 時間，aiDAPTIV）
- 8.415 s
- TPS (tokens per second) w/ aiDAPTIV TPS（每秒 tokens，aiDAPTIV）
- 20.353 t/s
- Power per machine 單機功耗
- 5 kW
One on-prem aiDAPTIV machine delivers the throughput of two rented cloud machines, so the cloud unit count is derived from half the per-machine capacity. 一台「地端 aiDAPTIV」可發揮兩台「雲端機器租用」的效能，雲端所需台數以單機能力的一半推算。