This analysis organizes three Lessons from the 2025 AI Platform Race into a practical comparison of evidence, decisions and current limits.
Read the evidence below as a decision trail: what changed, why it matters, which trade-offs shaped the result, and where the conclusion still depends on context.
The impact of the new “industrial revolution” brought about by the AI revolution
Since OpenAI launched ChatGPT, the development of generative AI has entered a new stage. Major technology giants such as Microsoft, Google, Meta, etc., have all increased their investment in AI R&D and services. This situation is like the major industrial revolutions in history - companies that master new technologies can often rise rapidly in the next few years or even ten years; while those who fail to catch up with the trend may be quickly eliminated from the market.
In this wave of AI revolution, "data autonomy" and "computing power" have become the core driving factors. If the industrial revolution is compared to a leap in machine production efficiency, the AI revolution uses massive data learning and inference to greatly improve human efficiency in manufacturing, production, management, marketing and various service models. Because of this, AI competition among global technology giants has become extremely fierce. Companies must not only "grab the market" but also "grab data." After all, being able to control huge amounts of data and transform it into precise and diverse applications is the real key to success.
The “IDM model” and competitive landscape of global technology giants
1. Extension of IDM thinking
In the traditional semiconductor industry, the so-called IDM (Integrated Device Manufacturer) refers to the "vertical integration" model that controls the entire process from wafer manufacturing, IC design to packaging and testing. By analogy to the current field of AI services, many large technology companies such as Microsoft (Azure), Amazon (AWS) and Google (Google Cloud) also have "one-stop" supply capabilities from basic computing resources to AI algorithm services.
However, in the AI era, IDM thinking is not limited to hardware or cloud services, but also extends to the control of "data acquisition" and "generative AI models". For closed AI service providers, all user data and parameters are trained and inferred in their cloud systems; users can only choose existing models or APIs provided by them for development or deployment, which is equivalent to the past state where they could only purchase IDM chips.
Under such a situation, if an enterprise wants to own its own proprietary data or models, it must rent or purchase computing power services from these "IDM-type" technology giants, and even face the risk of potential data leakage or competitive conflicts.
2. Possibility of open competition
Just as the semiconductor industry was dominated by IDM and eventually the market was significantly reshaped by the rise of the "foundry model", a similar new "division of labor" situation may also occur in the AI field.
In the future, if more open source models (such as Meta’s LLAMA etc.) and the emergence of independent training service providers. Those companies that want to maintain data autonomy will be able to cooperate with "AI foundries" to customize exclusive AI models while ensuring data security, without having to rely entirely on the closed solutions of large technology companies. Such an industrial model deserves great attention from technology companies and policymakers in various countries.
Data-driven core values
1. “Brute force computing” and data autonomy
Currently, many large-scale language models (LLM) adopt a "brute force calculation" strategy and are trained on massive amounts of public or private text. For example, ChatGPT continuously optimizes model parameters through huge text data on the Internet. However, if the company's internal confidential information or highly specialized domain knowledge is directly handed over to third-party giants, significant risks will arise:
- Competing risks: The company’s proprietary know-how was leaked, causing the company to lose its differentiation advantage.
- Security and privacy concerns: Lack of control over data flow, which may violate privacy or compliance regulations.
Therefore, data autonomy will be particularly important in the future. Only by having a controllable and secure data pipeline can we avoid handing critical enterprise information directly to external service providers, and can also train dedicated AI models that best meet enterprise goals as needed.
2. Data applications in multiple fields
The essence of AI is to extend "perception and decision-making" capabilities to a wider range. From smart manufacturing (factory automation and production optimization) to marketing (text and image generation), finance (risk control and investment analysis), we can all see the potential for large-scale application of AI.
More importantly, the data types and application requirements in each field are different: automotive electronics, medical imaging, bank credit, retail supply chain... all need to combine "Domain Knowledge" and "Generative AI" to truly realize value. This also implies that the more expertise a company can master in vertical fields and the more flexible it can use AI tools, the more advantages a company can gain in the market.
The Dilemma of AI Competitors and Industry Transformation
1. The Innovator’s Dilemma
“The Innovator’s Dilemma” refers to the fact that when leading companies face new technologies, they are often unable to abandon their original businesses and rush into new areas with all their strength, resulting in latecomers having the opportunity to subvert the market. Currently, in the field of AI, all major manufacturers are facing this dilemma:
- If we continue to invest heavily in "universal large-scale models", we will certainly be able to control a huge market, but at the same time we will need to pay high computing costs and face the challenges of user data leakage or supervision by various countries.
- If we switch to a "small but precise" customized model, we are worried about the spread of resources and the inability to charge on a large scale.
In the foreseeable future, more specialized AI (such as small models or private models) may emerge to provide more customized solutions for specific enterprise application scenarios, leaving room for growth for small and medium-sized manufacturers or new startups.
2. Imagination of industrial transformation
Once the "AI OEM model" gains enough trust and mature technology, there will be a large-scale division of labor in the market:
- core layer：Manufacturers that provide chips, servers and basic hardware resources.
- Cloud and computing layer: Similar to the large-scale computing power and basic platform provided by AWS, Azure, and Google Cloud.
- model layer: A service provider specializing in customizing and training AI models.
- Application layer: Software service provider that embeds AI into various industries and applications.
In this multi-level ecosystem, open cooperation is bound to gradually replace the "one-stop" monopoly of large manufacturers in the past.
Conclusion: Master data autonomy and embrace the AI revolution
From the industrial revolution to the Internet revolution, and then from the Internet revolution to the AI revolution, technological leaps have not only rewritten the industrial landscape, but also reversed human work patterns and competition logic. Under this trend, business leaders need to consider the following key strategies:
- Gain data autonomy: Establish a complete data governance and security mechanism internally, and carefully evaluate the cooperation model with cloud suppliers or AI model service providers externally.
- Develop software and hardware integration capabilities: In a multi-level ecosystem, a team that can understand both hardware computing characteristics and software algorithm development will become the company's core competitiveness.
- Embrace open collaboration: Utilize open source models and open platforms as much as possible to leverage their strengths; at the same time, maintain their own technology and data barriers to avoid being "locked in" by large platform providers.
- Focus on the vertical field: Instead of pursuing a "big and comprehensive" general model, it is better to find key solutions to pain points in the industries or customer groups you are most familiar with and establish differentiated advantages.
The impact of AI is far greater than a single application or a single industry, but has completely reshaped the ecological chain of "data-algorithm-application". Looking back at history, countries and companies that mastered steam engine and electric power technology during the Industrial Revolution established their world status; now, those who can discern trends in the new AI revolution and actively invest in R&D and market development will be the next leaders to "define the future."
What are your thoughts on the AI revolution? Welcome to share in the message area!
Articles you may be interested in
TSMC Semiconductor’s key decision: Global layout strategy in response to the AI Manhattan Project
US Chip Act - Profound Impact on Taiwan and China's Semiconductor Industry
What to take away
The article's value is in the evidence and trade-offs behind three Lessons from the 2025 AI Platform Race, not in treating the conclusion as universal.