AI investment theme evolution: from "tools" to services, platforms, and electricity

Event Background

In the latest CNBC report, Sundeep Gantori, Chief Investment Officer for Equities at Standard Chartered Bank, pointed out that AI investment themes are undergoing a profound evolution: the market is no longer solely focused on "gold rush tools" (i.e., chips, servers, and other infrastructure) but is shifting towards services, platforms, and energy. In the Asian market, this trend exhibits regional divergence—China remains dominated by "tools," India focuses on power infrastructure, while other regions are positioning around platforms.

This perspective reflects a critical shift in the AI industry from "hardware-driven" to "ecosystem-driven." Over the past two years, chipmakers led by Nvidia have been the biggest winners due to the surge in AI computing demand. However, as AI models move toward large-scale application, value is migrating to the higher layers of the industrial chain—services and platforms.

Digital Economy Analysis

User Growth and Traffic Changes

The rise of the platform layer means that the user reach of AI services will expand significantly. Taking cloud AI platforms (e.g., Microsoft Azure AI, Amazon Bedrock, Google Vertex AI) as examples, they lower development barriers through APIs and low-code tools, enabling small and medium-sized enterprises and even individual developers to leverage large model capabilities. This directly boosts the user base of AI services from tech companies to traditional industries. It is expected that by 2027, the number of global AI platform service users will exceed 1 billion, and the annual compound growth rate of enterprise-level AI API calls will surpass 80%.

Revaluation of Data Value

The core competitive advantage of the platform model lies in the data flywheel. AI service platforms can collect user feedback and model usage data to continuously optimize algorithms, forming a virtuous cycle of "better model → more users → better data." This means that platforms with the largest user bases will accumulate data advantages that are difficult to replicate, further reinforcing the Matthew effect.

Business Model Observations

From Selling "Shovels" to Selling "Solutions"

The traditional AI "tool" business model involves selling hardware or base models (e.g., chips, GPU computing power, pre-trained models) with one-time licenses or pay-per-compute-resource billing. In contrast, the service-oriented business model adopts on-demand API calls, subscriptions, or pay-per-result (e.g., per generation, per query). For example, OpenAI's ChatGPT Plus subscription and the API pricing of startups like Mistral AI both reflect the shift from "product" to "service."

Profit Margin Challenges and Solutions for the Platform Model

Although the platform layer has low marginal costs, competition is fierce. To acquire users, platforms need to continuously invest in R&D and marketing while facing pricing pressure. However, platforms can increase ARPU through value-added features (e.g., custom model fine-tuning, security and compliance services). Additionally, by integrating multiple AI capabilities (text, images, video), platforms can form a "one-stop AI workflow," enhancing user stickiness.

Electricity Becomes the New "Shovel"Gantori specifically highlighted the power theme in India. AI's energy consumption is becoming a bottleneck restricting the expansion of computing power. Training a large model on par with GPT-4 consumes electricity equivalent to the annual usage of thousands of households. Therefore, power infrastructure (including renewable energy, nuclear power, and efficient energy management) has become a new investment theme. This is not a "tool" in the traditional sense, but rather the underlying energy network supporting the entire AI economy.

Market Competition Analysis

China: Tool-Dominated Competition

In China, AI chips are affected by export controls. Domestic alternatives (such as Huawei Ascend, Cambricon, etc.) and policy support still offer significant opportunities in the "tool" sector. However, this also means that Chinese companies will focus on hardware and foundational models in the short to medium term, while the platform layer is not yet mature. Competition centers on computing efficiency and model breakthroughs.

India: Energy and Digital Infrastructure

India has a huge demand for data center construction while also pursuing green energy transition. AI's extremely high demand for electricity makes India a focal point for data center and renewable energy investment. Industry giants like Infosys and Reliance have begun to deploy AI cloud services, but platform expansion will take time.

Other Regions in Asia: Platform Competition

In Singapore, Southeast Asia, Japan, and South Korea, tech giants such as Sea Group, Grab, Naver, and Kakao are building localized AI platforms. They leverage regional data advantages (diverse languages, consumption habits) to develop differentiated services. Global cloud giants are competing for market share by establishing regional data centers and partnering with local entities.

Who benefits? Companies providing AI platform services (such as Microsoft, Google, Amazon) and data center operators with stable power cost advantages. Who faces challenges? Pure hardware vendors may face valuation corrections, and AI startups lacking platform capabilities will encounter growth bottlenecks.

Data and Regulatory Impact

Compliance Pressure on AI Service Platforms

As AI platforms penetrate sensitive fields such as finance, healthcare, and education, regulators are focusing on algorithm transparency, data privacy, and bias elimination. The EU AI Act has imposed strict compliance requirements on high-risk AI systems, requiring platforms to invest significant resources in auditing and governance. This increases operating costs but also builds entry barriers—platforms with strong compliance capabilities will gain a trust advantage.

Energy Regulation and Sustainable Development

Data center power consumption has drawn attention from multiple countries. Singapore has temporarily halted approvals for new data centers, and Europe plans to rate data center energy efficiency. In the future, AI energy consumption may be included in carbon trading markets, prompting companies to adopt more efficient hardware and green energy. This could drive AI computing power to migrate to regions rich in hydropower and nuclear power.

Global Trend Observation

Infrastructure Layering of the AI EconomyThis evolution marks the shift of the AI economy from a "computing power is everything" phase to a multi-layered structure of "computing power + platform + services + energy". Each layer has different investment logic and competitive dynamics. In the long term, the network effects and data flywheels of the platform layer will dominate value distribution, while the energy layer becomes a new geostrategic resource.

Super Apps and AI-Embedded Services

At the platform layer, super apps (such as WeChat, Gojek) are embedding AI capabilities, forming an "everything as a service" ecosystem. Users can complete tasks such as writing, translation, and image generation without leaving the app, which will further break the boundaries of traditional software.

Digital Sovereignty and Regional Deployment

Countries' emphasis on data sovereignty forces AI platforms to deploy infrastructure locally. This has given rise to the trend of "sovereign AI", which uses local data to train models and is subject to local laws. This creates opportunities for regional cloud service providers but also increases complexity for global platform companies.