Once a buzzword, the "digital middle platform" is now mired in what Gartner calls the "trough of disillusionment" —data keeps piling up but rarely actively flows.The investment in its construction is huge,yet it can hardly develop on its own.

Currently,digital transformation among enterprises have entered a plateau, and companies are desperate for the next breakthrough.

According to a forecast by the International Data Corporation(IDC),the global data volume will grow at a rate of 26.9% in 2025, and it is expected to reach 527.47 ZB by 2029. Yet China’s data retention rate stands at just 5.1%, highlighting inefficiency in data utilization.

Amid this challenge, the "data flywheel" ,as an emerging concept, is attracting increasing attention.It regards data as a continuously circulating asset that creates value in motion, built on a closed-loop of data-insight-action-feedback. Like a physical flywheel, it requires strong initial momentum but, once spinning, can sustain itself through feedback loops and self-reinforcement.

Chart: 2024 China Data,Analytics,and AI Technology Maturity Curve as of August 2024.

Source: Gartner

So,what exactly is a data flywheel,who are the core players,and what are they doing?

Why the Data Flywheel Is Rising as the Digital Middle Platform Fades？

Once,the digital middle platform broke the stagnation of enterprise data silos through data servicization and sharing.However,as digital transformation moves toward practical implementation,enterprises have come to realize that data unification alone is insufficient to meet business needs.If massive amounts of data cannot form an effective flow,it will be difficult to release its actual value. With the development of artificial intelligence, the question arises: how can massive amounts of data be utilized effectively? The concept of the"data flywheel"has emerged as a systematic solution to this challenge.

The concept of the data flywheel draws on the "flywheel effect" in physic.This theory was proposed by management expert Jim Collins and later popularized in practice by Amazon founder Jeff Bezos. In 2001, Bezos’s team articulated the e-commerce flywheel model, which outlined a self-reinforcing cycle: low prices attract more customers, a growing customer base draws more third-party sellers, the increase in sellers drives down logistics and operational costs, and lower costs, in turn, enable even lower prices. With the deepening of digitalization and intelligence, that same principle is being applied to data. The data flywheel centers on data consumption — business activities generate new data that feed back into building stronger data assets. Those assets, in turn, enhance operations and drive new growth, creating a continuous, upward-spiraling cycle.

Schematic Diagram of the Application of the Flywheel Effect in Amazon's Business

Source: Amazon

The digital middle platform lays the foundation, and the data flywheel is the high-rise built on it. Compared with the traditional middle platform, the data flywheel has represented a conceptual upgrade. Middle platform mostly focuses on the centralized storage and management of data and is prone to be costly, while the data flywheel emphasizes the in-depth integration of data flow and business flow. It treats data as both the engine and the goal, proving its commercial value through continuous value output. This transformation from asset-oriented to application-oriented is what gives the data flywheel its real staying power in practice.

With the data flywheel in place, the way data and knowledge drive business decisions is evolving----from directly driving decisions to providing auxiliary support for decisions. A study by the School of Economics and Management at Tsinghua University, "How to Build a Data Flywheel in the AI Era" shows that in the past, business was relatively stable, allowing knowledge to remain applicable for long stretches of time, Enterprises had relatively lower demands for the decision-making capabilities of their employees. However, at present, business is changing rapidly, forcing companies to make an ever-growing number of real-time decisions to stay efficient and competitive. That means it’s no longer enough to rely on static, past knowledge. Instead, organizations need access to the underlying data that can recreate previous scenarios — data that helps them think through new conditions and craft decisions tailored to the moment.

Actual Combat Guide: How Do the Three Giants Drive Industry Growth with Data?

Leading industry players are actively putting the data flywheel concept into practice. In China, Volcano Engine and Alibaba Cloud are building new architectures around it, while AWS is driving similar innovation globally.

1. Volcano Engine: From Digital Intelligence to Data Flywheel 2.0

Volcano Engine has incorporated the"data flywheel"into its product philosophy. Placing data consumption at its core, the company has pushed beyond traditional data warehouse capabilities by integrating multimodal data — including text, images, audio, video, and event streams. Its end-to-end architecture spans from operators and heterogeneous computing to model training and deployment, encompassing products such as VeDI, its multimodal data lake, and a suite of full-link data tools. In its technical papers and product materials, Volcano Engine repeatedly mentioned the importance of connecting large-model training with enterprise business workflows, forming a dual-engine system that links data consumption, asset accumulation, and application — what it calls Data Flywheel 2.0.

2. Alibaba Cloud- Bridging Big Data and AI Platforms

Alibaba Cloud's technology stack has long centered on big data warehouse MaxCompute, real-time computing, data middle platform, and AI platform PAI. Together, these form a unified system that spans data storage, batch and stream computing, feature engineering, model training, and online deployment. The company’s approach focuses on turning enterprise data into scalable, intelligent services.

3. AWS-Modular Methodology

AWS promotes the data flywheel as a methodology, emphasizing that it is not a single product but a complete set of components: storage, cataloging, training, inference, monitoring, and governance work in synergy. Through practical implementations in MLOps and its own Flywheel mechanisms—for instance, within Amazon Comprehend—AWS demonstrates how data warehouses, versioned datasets, and automated training pipelines can form a closed-loop ecosystem that continuously improves itself.

Industry Application: The Fundamental Value of the Data Flywheel

Ultimately, the value of technical products is reflected in real-world scenarios. While each of the three tech giants brings a distinct approach to the data flywheel, they share a common outcome: creating a self-reinforcing cycle where data value and business growth drive each other forward.

1. Volcano Engine: Rapid Experimentation of Flywheel in E-commerce and Brand Operations

Volcano Engine connects VeCDP, growth analysis DataFinder, A/B testing DataTester, and intelligent insight DataWind into a closed loop. First, it links global behaviors and builds tags, imports data into the data lake, and then uses growth analysis to discover high-potential users and trending products. A/B experiments are conducted to verify operational strategies, and successful ones are scaled across more touchpoints. This process continuously generates cleaner training and statistical data, allowing the flywheel to spin more steadily.

2. Alibaba Cloud: Flywheel in Supply Chain, Large-scale Retail, and Logistics Scenarios

Alibaba Cloud builds an integrated data warehouse-and-lake architecture using MaxCompute, Hologres, real-time computing (Flink), and PAI for machine learning. Real-time inflows of waybills, vehicle status, and warehouse status drive the model to generate scheduling and route suggestions. The scheduling results and service performance are then fed back into the system as new training data, governance indicators, and business rules. This forms a closed loop of continuous optimization that improves on-time delivery rates while reducing costs and inventory levels. Moreover, Alibaba Cloud provides this full suite of implementation tools and practices for major customers in the supply chain and retail sectors.

3.AWS: Multifunction Products, Media, and Mixed Flywheel Methodology

Nowadays, streaming media, international e-commerce, or multi-regional service providers need to turn massive user behaviors and content performance into replicable personalized recommendation engines and continuously iterate models in different markets. A AWS positions the data flywheel not as a single product but as a complete, modular methodology, combining data lakes, Glue for data cataloging, SageMaker for training, and managed services such as Amazon Personalize and the Flywheel mechanism in Amazon Comprehend. Taking the"Flywheel"function of Amazon Comprehend as an example, it automates the entire process of model training, evaluation, deployment, and feedback collection, shortening the cycle from"learning to application"to"learning new things"for the model.

To clearly compare the differences among various players, we have created the following analysis table:

A Bright Future, but a Tortuous Path

Like many emerging technologies, the data flywheel has a broad prospect, but the path to widespread adoption remains complex and challenging.

At the technical level, key hurdles persist. The hallucination problem in large language models has not been completely solved, which affects the credibility of analysis results—a problem that plagues many manufacturers. In addition, it is difficult to balance the fusion of multi-source data, real-time performance, and consistency.

The data flywheel emphasizes on promoting data production through data consumption, but many employees still lack the awareness or capability to make data-driven decisions. Business teams often depend heavily on technical departments, creating silos that hinder collaboration and highlight the absence of a true "data business partner (Data BP)" role. The data flywheel needs continuous iteration, and the traditional project management method needs to be updated and transformed.

The issue of cost and investment is also an important obstacle for enterprises, especially small and medium-sized ones. Building a data flywheel requires significant upfront spending, while short-term returns are difficult to quantify. The technical learning curve and implementation threshold remain steep for smaller organizations.

Looking forward to the future, the data flywheel will continue to evolve along several clear trajectories:

·AI interaction with lower thresholds will become the key to the popularization of the data flywheel.

·Smarter feedback loops will drive continuous optimization. With the development of AI and machine learning technologies, data analysis will become more intelligent, automatically generating insights and action strategies.

·Wider industry adaptation will promote the implementation of the data flywheel in more scenarios. From retail and manufacturing to medical care and finance, the concept and method of the data flywheel are being verified and promoted in different industries.

Ultimately, the data flywheel represents a paradigm shift - from "data engineering" to "cognitive engineering." When the speed of data flow surpasses the business iteration cycle, it unlocks an exponential amplification of value. In the future, AI native will become the core feature of the data flywheel.

If Data Flywheel 1.0 was about integration and 2.0 focused on empowerment, then the 3.0 will mark the era of symbiosis—where AI is no longer just a tool but also become the core engine driving the data flywheel from within.

Now that the amount of data has surged, the feedback cycle has shortened, and enterprises have begun to focus on how to make the system learn on its own. The rise of the data flywheel is a natural transition—from a centralized governance to a more dynamic circulation.

It is not a replacement but a relay. The middle platform helps enterprises understand the past, while the flywheel helps them adapt to the future. The former builds stability, and the latter pursues speed.

The real dividing line is not in the concept but in the organization. Whoever can embed data into every decision - and integrate AI directly into execution—will unlock a model of growth that runs almost on autopilot.

Technology rarely repeats the past. Instead, it pushes the same idea forward in new ways: building systems that are smarter, faster and more useful.