Company/Organization Profile

Founded in 2016, Zhejiang DataCenter Data & Technology Co., Ltd. is a green intelligent computing infrastructure and service provider for the core AI industry, and is among China's first batch of national-level "Little Giant" specialized and sophisticated enterprises as well as a National High-Tech Enterprise. Actively implementing the national "dual carbon" strategy, the Company has been deeply engaged in green data center and energy-saving & carbon-reduction technologies, leveraging AI to empower cooling optimization, on-site green electricity consumption and carbon emission reduction in intelligent computing centers. It has launched low-carbon products such as the integrated modular micro data center, obtained energy management system certification and won multiple green energy-saving honors. Its products are widely applied across more than ten industries including finance, transportation, energy, healthcare, education, government and telecom operators, continuously building a new space for green computing power.

机构简介

浙江德塔森特数据技术有限公司成立于2016年，是面向人工智能核心产业的绿色智能算力基础设施及服务提供商，是国家第一批重点专精特新“小巨人”企业、国家高新技术企业。公司积极践行国家“双碳”战略，深耕绿色数据中心与节能降碳技术，以人工智能赋能智算中心制冷优化、绿电就地消纳与碳减排，推出模块化微型数据机房一体机等低碳产品，通过能源管理体系等认证、荣获多项绿色节能荣誉，产品广泛应用于金融、交通、能源、医疗、教育、政府、运营商等十多个行业，持续构建绿色算力新空间。

Project Overview

Data centers are key infrastructure of the digital economy, yet they are also major sources of high energy consumption and carbon emissions.

Traditional intelligent computing centers generally face the following pain points:

High energy consumption: PUE has long remained in the 1.5–2.0 range, with low heat-dissipation efficiency and severe energy waste, making it difficult to adapt to the green and low-carbon development trend;

2. Design and energy-saving optimization rely heavily on the experience of senior engineers and repeated CFD simulation iterations, with a single design cycle lasting 3–6 weeks; manual tuning tends to be conservative — "hesitant to adjust and unable to adjust precisely" — resulting in prominent energy waste;

3. Safety risks are discovered too late: traditional power and environment monitoring can only alarm after a limit is exceeded, responding passively;

4. The output of green electricity such as photovoltaic and wind power fluctuates strongly on a minute-level scale, conflicting with the strict continuity requirements of intelligent computing workloads for power supply and heat dissipation.

To address these issues, the Company has used its own computer room as a testing ground to explore using artificial intelligence (world models and digital twins) to empower energy saving and carbon reduction in intelligent computing centers. Through AI-based forward-looking prediction and coordinated scheduling of temperature fields, airflow organization, energy consumption and equipment status, it has achieved a shift from "passive energy saving and after-the-fact alarming" to "proactive carbon reduction and pre-emptive simulation", building a new space for green computing power.

项目背景

数据中心是数字经济的关键基础设施，也是高耗能、高碳排的重要场景。

传统智算中心普遍面临以下痛点：

能耗高，PUE值长期处于1.5—2.0区间，散热效率低、能源浪费严重，难以适应绿色低碳发展趋势；

设计与节能优化高度依赖资深工程师经验与CFD仿真反复迭代，单方案周期长达3—6周，人工调优偏保守、“不敢调、调不准”，能源浪费突出；

安全风险发现滞后，传统动环监控只能在越限后告警、被动响应；

光伏、风电等绿电出力呈分钟级强波动，与智算负载对供电散热的连续性要求之间存在矛盾。

针对上述问题，公司以自有机房为试验场，探索以人工智能（世界模型、数字孪生）赋能智算中心节能降碳，通过AI对机房温度场、气流组织、能耗与设备状态进行前瞻预测与协同调度，实现从“被动节能、事后告警”向“主动降碳、事前推演”的转变，构建绿色算力新空间。

Project Implementation

In 2025, the Company established the DTCT.AI Artificial Intelligence R&D Center, forming a dedicated R&D team of over 30 professionals (including 5 PhDs and 14 masters) composed of algorithm scientists, large-model training engineers, data engineers and industrial automation experts. Leveraging its academician workstation and postdoctoral workstation, it also brought in external expertise in HVAC and fluid dynamics to fully undertake the R&D and deployment of this project. The project invested more than RMB 22 million cumulatively in 2025, all dedicated to AI computing power construction.

To address the four major pain points of "slow design, difficult energy saving, lagging risk response, and fluctuating green electricity", the project developed the "DTCT Lingjing AIDC World Model" and the "DTCT-EvoAgent Yaoling Self-Evolving Agent Framework", building a technical architecture of "forward-looking simulation by the world model and closed-loop execution by the agent cluster" that covers the full lifecycle of computer room planning–design–construction–operation–maintenance. The specific implementation is as follows:

Built a hybrid physics-information digital twin foundation, integrating high-fidelity physical simulation (CFD fluid dynamics, heat conduction and power distribution flow) with real-time sensor data to fundamentally eliminate physical hallucinations such as "non-conservation of heat" and meet industrial-grade safety and energy-saving requirements.

2. Trained the action-conditioned world model, continuously pre-training (equivalent to approximately 320 billion tokens) on data from computer room thermodynamics, fluid dynamics, and power distribution/HVAC domains to output multi-step predictions of future temperature fields, airflow organization, energy consumption curves and equipment status, and trained an FNO (Fourier Neural Operator) physics surrogate model to accelerate conventional CFD simulation from hour-level to second-level.

3. Built a reinforcement-learning optimization layer in the "imagination space", using safety, energy saving and reliability as a compound reward to simulate cooling optimization, workload scheduling and emergency plans in the world model's "imagination space" — verifying first in simulation and then rolling out gradually — to achieve a safe, energy-saving closed loop of "think it through before acting".

4. Developed the "Lvyao" (Green Radiance) green electricity–cooling capacity–computing power co-dispatch sub-model, incorporating photovoltaic and wind power output prediction and the "virtual cold storage" of the liquid cooling system into unified simulation, storing cold during green-power surplus periods and releasing it during shortage periods, thereby increasing on-site green electricity consumption and further reducing PUE.

5. Built the self-evolving agent framework, coordinating five agents — planning, design, construction supervision, operation and maintenance — with the closed loop of "diagnosis–data generation–self-training–verification–release" to achieve continuous self-evolution under private deployment.

In terms of resource assurance and computing power construction, the project adopted a two-tier "rental + self-built" computing architecture: elastic computing power was rented during training peaks for large-scale pre-training and fine-tuning, and from December 2025 a private computing cluster comprising 13 high-performance computing servers and a supporting computing management platform was established, meeting the "data does not leave the network" security and compliance requirements of finance and government customers.

The project was advanced in an orderly manner through six stages — "data engineering–pre-training–fine-tuning alignment–green-power sub-model–self-evolution closed loop–edge deployment" — starting in May 2025 and completing in April 2026. Using the Company's own Ningbo intelligent computing room to carry out an "unmanned operation transformation" as the first deployment scenario, energy-saving and safety strategies were directly issued to the power and environment monitoring control layer through digital-twin offline verification and gradual rollout mechanisms, achieving unmanned operation.

项目实施

2025年，公司成立DTCT.AI人工智能研发中心，组建由算法科学家、大模型训练工程师、数据工程师与工业自动化专家构成的专职研发团队30余人（其中博士5人、硕士14人），并依托院士工作站、博士后工作站引入暖通空调与流体力学外部智力，全面承担本项目研发与落地。项目2025年度累计投入2200余万元，全部用于人工智能算力建设。

针对“设计慢、节能难、风险滞后、绿电波动”四大痛点，项目研发了“DTCT灵境AIDC世界模型”与“DTCT-EvoAgent曜灵自进化智能体框架”，构建“以世界模型做前瞻推演、智能体集群做闭环执行”的技术架构，覆盖机房规划—设计—建设—运维—运营全生命周期。具体实施如下：

1、构建物理信息混合数字孪生底座，融合高保真物理仿真（CFD流体力学、热传导、配电潮流）与实时传感器数据，从机理上杜绝“热量不守恒”等物理幻觉，满足工业级安全与节能要求。

2、训练动作条件世界模型，基于机房热力学、流体力学、配电暖通领域数据持续预训练（约3200亿Token等效），输出未来温度场、气流组织、能耗曲线与设备状态的多步预测，并训练FNO物理代理模型将传统CFD小时级仿真加速至秒级。

3、建设想象空间强化学习优化层，以安全、节能、可靠为复合奖励，在世界模型“想象空间”中推演冷却调优、负载调度与应急预案，先仿真验证、再灰度上线，实现“先想清楚、再动手”的安全节能闭环。

4、研发“绿曜”绿电-冷量-算力协同调度子模型，将光伏、风电出力预测与液冷系统“虚拟蓄冷”纳入统一推演，绿电富余时段蓄冷、匮乏时段释冷托底，提升绿电就地消纳并进一步压低PUE。

5、搭建自进化智能体框架，以规划、设计、监造、运维、运营五大智能体配合“诊断—造数—自训—验证—发布”闭环，实现私有化部署下的持续自进化。

在资源保障与算力建设上，项目采取“租赁+自建”两级算力架构：训练高峰期租用弹性算力承担大规模预训练与微调，2025年12月起建成含13台高性能算力服务器的私有化算力集群及配套算力管理平台，满足金融、政务等客户“数据不出网”的安全合规要求。

项目实施按“数据工程—预训练—微调对齐—绿电子模型—自进化闭环—边缘部署”六个阶段有序推进，2025年5月启动、2026年4月完成，并以公司自有宁波智算机房开展“无人化运行改造”作为首要落地场景，通过数字孪生离线验证与灰度发布机制，将节能与安全策略直接下发至动环控制层，实现无人化运行。

Project Outcome

Significant before-and-after comparison:

1) Improved energy efficiency: the Company's own Ningbo intelligent computing room was already a highly efficient modular computer room with a Power Usage Effectiveness (PUE = total facility energy / IT equipment energy) of 1.42, better than the industry benchmark (1.5–2.0); after AI optimization, PUE was further reduced to 1.21 and cooling system energy consumption dropped by about 31%. With further innovation, the on-site green electricity consumption rate of the pilot edge computer room equipped with photovoltaic and wind power rose from about 62% to 91%, with PUE further reduced to 1.18.

2) Improved safety capability: hot-spot risk was upgraded from "alarm after exceeding the limit" to forward-looking warning averaging 37 minutes in advance; equipment fault prediction accuracy (72 hours in advance) reached 92.5%; unmanned night duty was achieved, with zero business interruptions caused by thermal faults throughout the year.

3) Improved efficiency: the design cycle for new computer room solutions was shortened from 3–6 weeks to within 2 days; single-solution simulation was accelerated from hour-level (CFD) to within 0.8 seconds; and computer room O&M manpower demand dropped by about 60%.

2. Outstanding technological innovation highlights:

First, the Company was the first in China to engineer the hybrid architecture of "physics-information digital twin + action-conditioned world model" into data center scenarios, using physical anchors to suppress the physical hallucinations of generative models and achieving industrial-grade safety and usability. Second, it pioneered a self-evolving agent cluster covering the full lifecycle of AI infrastructure, keeping the recursive drift rate within 1%. Third, it opened up the end-to-end chain of "world model simulation–strategy verification–power & environment execution", achieving the leap from "advisory suggestions" to "directly driving unmanned operation". Fourth, it was the first to incorporate the three elements of "green electricity–cooling capacity–computing power" into unified simulation, aligning with the policy direction of "direct green-power connection" and "computing-power coordination".

The project has significant sustainability and replicability:

The single-round iteration cycle of the self-evolving framework was compressed from 21 days to 6 days, and the recursive drift rate dropped from 8.7% to 0.9%. Nine rounds of closed-loop iterations have been stably run, forming a long-term operating mechanism that "gets more accurate with use". After verification matured, the solution was replicated to two external pilot customers — a financial industry data center and a government cloud intelligent computing center — and is offered to existing modular computer room customers as a "world model + digital twin" subscription service for energy-saving optimization and safety warning.

成果亮点

1、项目实施前后对比显著：

1）能源利用效率提高：公司自有宁波智算机房本身已是模块化高效机房，电源使用效率（PUE=数据中心总设备能耗/IT设备能耗）为1.42，优于行业标准（1.5—2.0）；经AI优化PUE进一步降至1.21，制冷系统能耗降低约31%；进一步革新，配置光伏、风电的试点边缘机房绿电就地消纳率由约62%提升至91%，PUE进一步降至1.18。

2）安全能力提高：热点风险由“越限后告警”升级为平均提前37分钟前瞻预警，设备故障预测准确率（提前72小时）达92.5%，实现夜间无人值守，全年未发生一起因热故障导致的业务中断。

3）效率提高：新建机房方案设计周期由3—6周缩短至2天以内，单方案仿真推演由小时级（CFD）加速至0.8秒以内，机房运维人力需求下降约60%。

2、技术创新亮点突出：

一是国内率先将“物理信息数字孪生+动作条件世界模型”混合架构工程化落地于数据中心场景，以物理锚点抑制生成式模型的物理幻觉，达到工业级安全可用水平；二是首创面向AI基础设施全生命周期的自进化智能体集群，递归漂移率压制到1%以内；三是打通“世界模型推演—策略验证—动环执行”端到端链路，实现从“辅助建议”到“直接驱动无人化运行”的跨越；四是率先将“绿电-冷量-算力”三要素纳入统一推演，契合“绿电直连”“算电协同”政策导向。

3、项目具备显著的持续性与可复制性：

自进化框架单轮迭代周期由21天压缩至6天、递归漂移率由8.7%降至0.9%，已稳定运行9轮闭环迭代，形成“越用越准”的长期运行机制；方案验证成熟后已复制推广至金融行业数据中心、政务云智算中心等2家外部客户试点，并以“世界模型+数字孪生”订阅服务方式向存量模块化机房客户输出节能优化与安全预警服务。

Project Highlights

Ecological and environmental benefits: after AI optimization, the Company's own Ningbo intelligent computing room has significantly reduced electricity consumption and carbon emissions. Based on the promotion and application estimate, if the relevant green energy-saving technologies were extended to 10% of data centers nationwide, they would save about 4.6–5.0 billion kWh of electricity per year, equivalent to reducing carbon emissions by 2.93–3.35 million tons, strongly supporting the national "dual carbon" goal.

Social impact: the project has formed a replicable "AI + green computing power" demonstration paradigm that has been promoted to external customers in the finance and government sectors; the Company's R&D team has deeply participated in the drafting of national and industry standards in the modular data center field, providing standard support for the industry's green and low-carbon transformation; the project, located in the Ningbo High-Tech Zone, has linked up with the academician workstation, postdoctoral workstation and the Ningbo University Institute of Intelligent Edge Computing to drive the clustering of regional AI infrastructure talent and industry, delivering outstanding industry demonstration and regional spillover effects. The Company's exhibition hall has received nearly a thousand visitors, including trainees of the Municipal Party School, government personnel and high-tech enterprise talent, spreading advanced concepts of green computing power and AI-based energy saving and carbon reduction, winning broad recognition from both industry and government circles and creating a favorable cognitive foundation and industry atmosphere for the subsequent deployment of similar technologies.

成果影响力

生态环境效益方面：公司自有宁波智算机房在AI优化后已显著减少电力消耗与碳排放，按产品推广应用测算，相关绿色节能技术若推广至全国10%的数据中心，年节电量约46—50亿度，相当于减少碳排放293—335万吨，有力支撑国家“双碳”目标。

社会影响方面：项目形成可复制的“AI+绿色算力”示范范式，已向金融、政务等外部客户复制推广；公司研发团队深度参与模块化数据中心领域国家标准与行业标准编制，为行业绿色低碳转型提供标准支撑；项目落地宁波高新区，联动院士工作站、博士后工作站与宁波大学智能边缘计算研究院，带动区域AI基础设施人才与产业聚集，具有突出的行业示范与区域带动效应；公司展厅接待市委党校学员、政府部门人员、高新企业人才等参观人数近千名，普及了绿色算力、AI节能降碳的先进理念，获得了产业界与政务领域的广泛认可，为后续同类技术落地应用营造了良好的认知基础与行业氛围。

Reviews & Honors

User and industry recognition: after trial use, users including the First Affiliated Hospital of Ningbo University, Ningbo Airport Group and Guoneng Zhejiang Beilun No. 3 Power Generation Co., Ltd. reported positive feedback, recognizing the outstanding performance in safety and reliability, energy saving and consumption reduction, and O&M efficiency. The DTCT-AIDC intelligent computing all-in-one machine was appraised by five experts including Academician Tan Jianrong and Academician Peter Sachsenmeier as reaching an internationally leading level overall; the Company's products have been promoted and applied in more than 30 provinces and cities across China.

2. Qualifications and certifications: the Company is one of China's first batch of national-level "Little Giant" specialized and sophisticated enterprises, a National High-Tech Enterprise and a Zhejiang Provincial Technology "Little Giant" Enterprise. It has passed ISO 9001, ISO 14001, ISO 45001, ISO 50001, ISO 20000 and ISO 27001 system certifications, as well as CQC energy-saving certification.

3. Awards and honors: the Company has won honors including the China Data Center Green Energy-Saving Award and the China Intelligent Building Industry Innovative Product Award; it has been selected onto the Ningbo Municipal Green Manufacturing List and rated as a Zhejiang Provincial Manufacturing Single-Item Champion Enterprise. The integrated modular micro data center was included in the 2024 National MIIT Recommended Catalog of Energy-Saving and Carbon-Reduction Technology and Equipment; the DTCT-AIDC intelligent computing all-in-one machine was approved as a 2025 domestically first-of-its-kind major technical equipment, a Provincial Outstanding New Industrial Product, and a first-edition software product.

评价与荣誉

1、用户与行业认可方面：宁波大学附属第一医院、宁波机场集团、国能浙江北仑第三发电有限公司等用户试用后反馈良好，认可其在安全可靠、节能降耗与运维效率方面的突出表现；DTCT-AIDC智算一体机经谭建荣院士、Peter Sachsenmeier院士等五位专家鉴定，整体技术达到国际领先水平，公司产品已在全国30多个省市推广应用。

2、资质认证方面：公司为国家第一批重点专精特新“小巨人”企业、国家高新技术企业、浙江省科技小巨人企业，通过ISO9001、ISO14001、ISO45001、ISO50001、ISO20000、ISO27001等体系认证及CQC节能认证。

3、奖项荣誉方面：公司荣获中国数据中心绿色节能奖、中国智能建筑行业创新产品奖等荣誉，入选宁波市绿色制造名单，获评浙江省制造业单项冠军企业；模块化微型数据机房一体机列入2024年《国家工信领域节能降碳技术装备推荐目录》；DTCT-AIDC智算一体机获批2025年度国内首台套、省优秀工业新产品及首版次产品。