Billing reference

DCU unit, GPU/CPU/storage/registry pricing, CCI specs, and 2025-12-19 storage RMB switch

Alaya NeW Cloud uses prepaid, real-time deduction with DCU (Data Computing Unit) as the unit of compute. Every CCI, VKS, HyperTrain, ALab, and inference workload is metered against actual resource usage and billed in DCU.

1. The DCU unit

Conversion

1 DCU = 312 TFLOPS × 1 hour

One DCU represents the standard workload of a 312 TFLOPS (FP16) resource running for one hour.

Value composition

DCU value = V_compute + V_infra + V_software + V_ops

V_compute — accelerator (GPU/NPU) compute
V_infra — high-speed network, performance storage, physical infrastructure
V_software — Alaya NeW management plane and AI base software
V_ops — automated lifecycle ops, monitoring, expert support

Why packs

Pay as you go — no per-card pricing to reason about; pay for actual workload.
Low barrier — no infra knowledge required; just call the API.
Universally accessible — flexible quotas suited to small teams.

2. GPU billing

Different accelerators have different DCU rates determined by architecture, performance, memory bandwidth, and capacity. Effective rates are always shown on the resource provisioning page.

GPU	DCU per GPU-Hour
NVIDIA H800E (NVLink)	3.07
NVIDIA H800 (NVLink)	2.56
NVIDIA H800A (NVLink)	2.56
NVIDIA A100 (NVLink)	1
NVIDIA A800 (NVLink)	1
NVIDIA L40S	0.65

Rules: Charging starts when the workload occupies the GPU. Settlement runs hourly on the hour and again on release. Sub-hour usage is billed by the second.

3. CPU billing

Resource	DCU per core-hour
CPU	0.025

Rules:

Only CPU-only workloads are billed at this rate.
Workloads that also use a GPU do not pay separately for their attached CPUs.
Same real-time metering, hourly settlement, and per-second sub-hour billing as GPU.

4. CCI product specs

CCI offers three spec families — CPU-only, H800A, and L40S — covering everything from light development to large-model training.

Spec	CPU-only	H800A	L40S
GPU	/	NVIDIA-H800A-NV-80G ×1, ×2, ×4, ×8	NVIDIA-L40S-PCIE-48G ×1, ×2, ×4, ×8
CPU + RAM	2 core / 4GB	18/36/72/144 core + 200~1600GB	10/20/40/80 core + 80~640GB
Disk	50GB	50GB	50GB
DCU/hour	0.05	2.56 ~ 20.48	0.65 ~ 5.20
Best for	Light dev, debug, data prep	High-throughput inference, training, HPC	Large-model inference, online serving, vision, mid-size training

See CCI overview for product details.

5. Storage and image registry

Legacy DCU pricing (before 2025-12-19)

Resource	DCU/GB/hour
NAS bulk	0.0001
NAS performance	0.0002
Bulk	0.0001
Performance	0.0002
AI-accelerated	0.0002
Object	0.0001
Image registry	0.0001

Rules:

Charging starts at the configured maximum capacity from the moment the volume is provisioned. Hourly settlement on the hour, final settlement on release, sub-hour billed by the second.
Stopping a task or app does not release storage — the volume keeps occupying quota and accruing charges until explicitly deleted.

RMB pricing (effective 2025-12-19)

From 2025-12-19, storage products and the image registry switch from DCU metering to RMB pricing — either ¥/GB/hour or ¥/TB/month. Final pricing always defers to the order page.

Monthly RMB plans (sold in TB units, discounted):

Product	¥/TB/month
NAS bulk	120
NAS performance	1020
Bulk	120
Performance	1020
AI-accelerated	1020
Object	120
Image registry	120

RMB pay-as-you-go (provision on demand, scale freely):

Product	¥/GB/hour
NAS bulk	0.00027
NAS performance	0.002
Bulk	0.00027
Performance	0.002
AI-accelerated	0.002
Object	0.00027
Image registry	0.00027

Migration notes

DCU purchased before 2025-12-19 (including overdraft credit) keeps working for storage and the registry. Once that DCU pool runs dry, pricing automatically falls back to RMB.
DCU purchased after 2025-12-19 cannot pay for these products by default — confirm via the purchase page.
For clarifications or special pricing arrangements, submit a request via the Alaya NeW website's online inquiry.

Heads up: For data safety, stopping a task or app does not release storage volumes or registries. They keep accruing charges until explicitly deleted.