Summary of Key Points
AI investment is shifting from chips and data centers (the "brain") to the network infrastructure of telecommunications operators (the "nervous system"). As AI evolves from virtual intelligence to physical applications such as robots, autonomous vehicles, and smart glasses, it requires a network to connect the large models (the brain) with the devices (the body), addressing issues like real-time reasoning and offloading computational tasks. Operators have the potential to become winners in this new phase of AI development by offering token-based pricing for computing power and differentiated connection services tailored to specific use cases.
Detailed Explanation
1. The "Nervous System" of Physical AI: Why is the Network Essential?
You can think of AI as a person, where large language models represent the brain, robots, drones, and autonomous vehicles represent the body, and the network serves as the nervous system. Without a network, even the most intelligent brain cannot control its limbs effectively. For example, an autonomous vehicle needs to make instant decisions; if data must be transmitted over long distances to a data center (even if it’s just for a few seconds), accidents could occur. Smart glasses need to be lightweight and cannot rely on large batteries, so some computations must be offloaded to nearby servers (known as edge computing). Ericsson has collaborated with robotics companies to move the computational load of robots onto the network, enabling them to launch products at lower costs.
2. Network Bandwidth Becomes a New Bottleneck for AI: Will the Market Double in Size?
Previously, the focus was on chips and data centers, but as AI clusters grow larger (e.g., training large models requires extensive data transfer), network bandwidth is becoming insufficient. A report by Bank of America suggests that the demand for AI networks has been significantly underestimated, with the market expected to reach $316 billion by 2030 (an increase of $76 billion from previous forecasts). This is similar to your home WiFi: it was sufficient for watching videos, but now, when everyone uses AI-generated content simultaneously, the bandwidth becomes bottlenecked. The data volume generated by AI is much larger than that of regular video, so networks need to be upgraded.
3. Operators: Moving from Selling Data Traffic to "AI Computing Packages" (Token-Based Services)
Previously, operators sold unlimited data traffic, and as more users consumed it, their costs increased without a corresponding rise in revenue. However, with the introduction of tokens (which represent the basic units of AI processing, e.g., the number of tokens required to generate a sentence), this has changed. China’s three major telecom operators have already launched token-based services, offering the computing power of large AI models as data packages. For example, users pay based on the number of tokens used. This is seen as promising by experts like Lan Shangli, who believes that revenue from these services will increase as wearable devices and drones become more widespread.
4. Differentiated Connection Services: Operators’ Competitive Advantage
Current networks provide a one-size-fits-all service. However, future applications will require different levels of performance and security. Autonomous vehicles need low latency, smart cities require high data security, and the low-altitude economy (e.g., drone delivery) needs stable connections. By offering customized services that meet these requirements, operators can command higher prices. For instance, providing a "green channel" for autonomous vehicles with a data transmission delay of less than 0.1 seconds could result in higher fees. Operators, with their control over networks, customer relationships, and AI assets, are well-positioned to capitalize on these trends.
5. Overcoming the Barrier of Incompatible Devices
Different devices use various communication protocols, similar to the language barriers between different cultures (e.g., Chinese and French speakers). This can lead to compatibility issues, such as gardening robots and security drones failing to communicate effectively. Ericsson’s acquisition of Vonage aims to create an API platform (Aduna) that enables devices with different protocols to interact seamlessly. This is a critical issue that must be resolved before the widespread adoption of physical AI.
In Summary
For AI to become a reality, a robust network is essential. Operators that can address network bottlenecks, offer effective AI computing services, and provide differentiated solutions will have a significant advantage in the future of AI development.