CUSTOMER STORIES
Data Stays On-Premises
Compute Arrives Instantly
From university training to industrial inspection, from medical imaging to financial risk control — see how WeCalc delivers 48-hour deployment, 40–62% TCO reduction, and full on-premises data processing across industries.
CASE 01
·WeCalc-BEducation & ResearchAI Training Platform for 100+ Students — Built in 72 Hours
Beijing Information Science & Technology University · Beijing
Challenge
The university needed an on-premises teaching and training platform for its AI program, supporting 100+ concurrent model training sessions while keeping all student data within the campus network. Traditional solutions quoted over RMB 800K with a 3-month minimum timeline — exceeding both budget and schedule constraints.
Solution
Deployed 1 WeCalc-B Basic unit, completing the full process from delivery, racking, networking, to platform readiness within 72 hours. The disaggregated storage-compute architecture enables elastic multi-user resource isolation. A one-click management console lets instructors manage compute allocation without dedicated IT staff.
Previously, our students could only run demos on free cloud credits. Now every student can independently train their own models on campus with full data security. One semester in, competition award rates increased by 30%.
— Director, Computer Science Lab Center, Beijing Information Science & Technology University
Highlights
CASE 02
·WeCalc-BSmart ManufacturingAI Visual Inspection: Defect Miss Rate from 2.3% to 0.15%
Auto Parts Manufacturer · Yangtze River Delta
Challenge
The company produces 8 million precision parts annually. Manual inspection had a 2.3% miss rate, causing over RMB 2M in annual return and claim losses. An attempt to upload inspection images to the cloud was blocked by the security compliance team, as production data contains customer drawings and process parameters.
Solution
Deployed 1 WeCalc-B at the production edge, running a custom visual inspection model. Industrial cameras capture and infer in real-time on-site — data from capture to verdict never leaves the factory floor. Deployed in 48 hours; model iteration continues through local incremental training.
Two months after going live, our customer return rate dropped by 80%, and the inspection team went from 12 people to just 3 for verification. Most critically, not a single production image ever left the factory.
— Director of Quality, Auto Parts Manufacturer
Highlights
CASE 03
·WeCalc-BHealthcareAI-Assisted Imaging: Average Report Time Reduced by 65%
Provincial Capital Tier-3A Hospital · Central China
Challenge
The radiology department processes 400+ CT/MRI scans daily, with each physician reading 80+ cases per day under extreme workload. The hospital explored cloud-based AI diagnostic platforms, but the imaging data contains extensive patient privacy information. The health commission and IT security department explicitly required all data to remain within the hospital network.
Solution
Deployed 1 WeCalc-B in the hospital data center running NMPA-registered lung nodule screening and fracture detection models. Integrated with the PACS system, images are automatically pushed for local inference, with AI annotations returned to the diagnostic workstation within 30 seconds — all data stays within the hospital.
Since deploying WeCalc's local AI diagnostics, our average report turnaround dropped from 45 minutes to 16 minutes. Physicians can now focus more energy on complex cases. Most importantly, not a single byte of patient data has ever left the hospital.
— Chief of Radiology, Provincial Capital Tier-3A Hospital
Highlights
CASE 04
·WeCalc-PFinancial TechnologyOn-Premises Intelligent Risk Control: Real-Time Anti-Fraud ≤80ms
East China City Commercial Bank · East China
Challenge
The bank processes 500K+ daily credit card transactions. Its legacy rule engine had a 5% false positive rate, blocking legitimate transactions and generating customer complaints. The banking regulator requires core transaction data and customer information to remain off-cloud, with risk control inference responding within 100ms.
Solution
Deployed 1 WeCalc-P Professional cluster hosting three AI models: real-time risk control inference, anti-fraud detection, and customer profiling. A 100G RDMA low-latency network ensures high-throughput real-time data ingestion and model inference — all data and models operate within the bank's closed network.
After deploying WeCalc, our risk control false positive rate dropped from 5% to 1.2%, and customer complaints were cut in half. During regulatory audits, our IT department can confidently say — all customer data stays in-house, not a single record has left.
— GM, Information Technology Department, East China City Commercial Bank
Highlights
CASE 05
·WeCalc-PAutonomous DrivingLocal Closed-Loop Data Processing: Model Iteration Cycle Cut by 40%
L4 Autonomous Driving Technology Company · Beijing
Challenge
The company generates over 2TB of daily road-test data (video, point clouds, IMU). Previously, uploading data to public cloud for training was slow, bandwidth-expensive, and involved city road information and pedestrian privacy — creating growing compliance risks.
Solution
Deployed 2 WeCalc-P Professional units at the R&D center, building an integrated local platform for data annotation, model training, and simulation validation. Road-test data is transmitted via a dedicated link and processed locally — cleaning, annotation, and training happen without cloud upload. EBOF all-flash storage ensures high-speed write and random read for 2TB/day throughput.
Data upload alone used to take half a day. Now data starts training the moment it arrives — our model iteration pace is nearly twice as fast. And we no longer need to go back and forth with legal over data security.
— Head of Algorithm Platform, L4 Autonomous Driving Technology Company
Highlights
CASE 06
·WeCalc-BSmart ParkShared AI Compute for 30+ SMEs in the Park
National Hi-Tech Zone Management Committee · Central China
Challenge
The hi-tech zone hosts 200+ tech SMEs, with 30+ having clear AI needs (visual inspection, intelligent customer service, data analytics), but individual hardware investment was prohibitively expensive. The management committee aimed to provide public compute services to lower the AI adoption barrier while keeping data within the park.
Solution
The committee deployed 3 WeCalc-B units in the park data center, using WeCalc's management platform for multi-tenant resource isolation and on-demand allocation. 30+ enterprises access shared compute via the park intranet — each company's data and models are fully isolated, billed by actual usage. The committee offers compute as a public service, dramatically lowering AI adoption costs.
Previously, park enterprises had to either use the cloud or buy their own hardware for AI. Now with shared compute services, a company can run its own models for just a few hundred RMB per month. This is truly universal computing.
— Director, Digital Economy Development Center, National Hi-Tech Zone Management Committee
Highlights
TCO COMPARISON
WeCalc TCO Comparison Overview
Based on appendix H calculations from the WeCalc Business Plan and actual deployment data
| Scenario | Traditional / Public Cloud | WeCalc Solution | TCO Savings | Timeline Comparison |
|---|---|---|---|---|
| 1E Compute Build-Out | RMB 355–380M | RMB 140–185M | 58–62% | 6–18 months → 2–4 weeks |
| 1P Video Gen AI (Purchase) | RMB 960K–1.44M | RMB 122K (purchase) | 87–92% | 2–4 weeks → 48–72 hours |
| 1P Video Gen AI (Lease) | ~RMB 1.05M (cloud) | RMB 72K (lease) | 93% | Instant → 48–72 hours |
WHY IT SAVES
Why WeCalc Delivers Significant Cost Savings
Cost reduction isn't just about lower purchase prices — it comes from systematic optimization across architecture, delivery, utilization, and operational efficiency
Disaggregated Architecture
Compute and storage scale independently on demand. Adding nodes delivers linear growth in both compute and storage — no more full-rack stacking.
EBOF All-Flash Storage
Hardware-accelerated NVMe-oF all-flash storage with EC erasure coding cuts storage costs by 40%+, with 20% redundancy overhead far below traditional RAID.
80%+ Resource Utilization
Traditional solutions average only 40%. WeCalc's elastic scheduling and disaggregated architecture push utilization above 80%, halving hardware investment for the same workload.
Financing Lease Option
WeCalc-B financing lease starts at just RMB 2,000/month — 3-year TCO of approximately RMB 72K lets SMEs access local AI compute with near-zero upfront cost.
Get Your Custom Cost-Saving Plan
Whether you're a university, manufacturer, hospital, or financial institution, we can provide a practical cost analysis and deployment plan based on your specific business scenario, data scale, and compliance requirements.