In my previous article on edge AI, I discussed the business perspective of Total Cost of Ownership (TCO). That is an important angle, but not the whole picture.
International affairs have been turbulent lately. The developments involving Iran and the United States would have been shocking five years ago, but now feel almost routine. This has led me to ask another question: in an increasingly volatile world, what does AI really mean?
After thinking about it for a long time, my conclusion is this: AI is not just a business issue; it is a question of power.
Compute Is Power
AI is changing the fundamental formula of economic activity.
In the past: Capital + Labor = Productivity
Now: Capital → Compute + Electricity + Model = Productivity
The essence of this shift is that capital can be converted directly into productivity, bypassing human labor. And as embodied intelligence and robotics advance, this conversion efficiency will only increase.
And productivity has always been synonymous with power.
In feudal times, whoever owned the most land and slaves possessed the productivity, and thus ruled. In the industrial age, capital became the source of power, but still required human labor, so the concentration of power was limited. Now, if compute can be transformed directly into productivity, and robots can replace human labor—then an extreme minority monopolizing vast productive forces is no longer science fiction.
The Cloud Trap
Imagine a scenario.
Over the past few years, an ordinary individual or a small business has grown accustomed to accessing AI capabilities through cloud services. Their work, life, and production are entirely dependent on the intelligence services provided by a handful of large platforms.
What happens then?
The most immediate consequence is price hikes. Cloud providers can gradually raise prices, extracting surplus value from users. It is like a landlord raising rent—if you cannot pay, you must leave. But the problem is, you cannot leave. Your entire workflow is tied to the platform, and the cost of migration is unbearably high.
Furthermore, if tensions escalate, cloud providers can simply cut off your access. Or more subtly, they can limit your API call volume or degrade your service quality. You have no bargaining power because you hold no chips.
But these are not my deepest concerns.
In the past, when a group of people united, they could wield great power because human beings were the core of productivity. Capitalists needed workers; landlords needed peasants. But in this new era, if a small minority can command enormous productivity through AI and robots, do they really still need so many people?
I do not know the answer. But it is a question worth taking seriously.
The Meaning of Edge
This is where edge AI devices derive their value. They are not a cost-optimization scheme, but a check on power.
Edge devices have three characteristics: they are cheap, distributed, and difficult to control.
Nvidia's DGX Spark costs around 30,000 RMB, AMD's Ryzen AI Max+ 395 is under 20,000 RMB, and China's E300 module can run a 32B model in a volume of less than 10 cubic centimeters. There is no need to concentrate them in a single data center; they can be distributed across millions of households. And once you own the device, it is yours—no one can shut it down remotely.
It is like a situation where one side possesses tanks and heavy artillery, while the other side consists of civilians with rifles. The heavy weapons are undoubtedly powerful, but their numbers are limited. The rifles in civilian hands may be weak individually, but they win through sheer numbers and wide distribution. The two sides can form a balance of power. Moreover, from a legal perspective, edge devices are sold to you, while cloud services are rented to you—a distinction whose legal significance in the Agent era is far greater than most imagine.
Kevin Kelly, founder of Wired, said in 2025: over 70% of current investment is concentrated in centralized cloud computing, yet over 70% of actual computation is already happening on edge devices. The future AI architecture will likely be hybrid—cloud handles training, edge handles inference—and dominance will gradually shift toward the edge.
The Technology Is Already Here
Edge AI is not a distant future; it is a present reality.
30B-parameter models already run well on edge devices. WeChat Read runs a 30B model on the Snapdragon 8 Gen4 NPU at just 1.4 watts. Houmo AI's chips can run a 70B-parameter model at 10 watts. Edge devices capable of running 30B models have already dropped to around ten to twenty thousand RMB. Compared with annual cloud API fees ranging from tens of thousands to hundreds of thousands of RMB, the payback period on this investment is short.
Electricity is the main ongoing cost, but edge devices are vastly more energy-efficient than the cloud—ARM processors can use up to 10,000 times less power than the cloud for the same task.
The open-source community also provides a wealth of high-quality models: Llama, Qwen, DeepSeek... You do not need to depend on any single company.
The technical barriers are falling rapidly. In 2026, it is not at all far-fetched for an ordinary person to own an edge device capable of running a 30B model.
From Development to Survival
In recent years, we have grown accustomed to viewing problems through the lens of "development": how to grow, how to make money, how to scale up.
But the world is changing. The international order is crumbling; the classical Chinese phrase li beng yue huai (礼崩乐坏, the collapse of ritual and music) is not hyperbole for describing this breakdown of social norms. In this environment, the lens through which we view problems must shift from "development" to "survival."
Survival means you cannot rely entirely on external supply. You need your own backup plan. You need to possess things that others cannot take away.
Edge AI devices are such things. They give you options when cloud services raise prices, a fallback when platforms cut off supply, and autonomy when the external environment changes drastically.
This is not merely an individual or corporate issue. If a society's entire intelligence is concentrated on a handful of platforms, that society is extremely fragile. Distributed, decentralized edge AI can enhance the resilience of society as a whole.
Final Thoughts
I did not write this article to sing the praises of edge AI devices—the business case was covered in my previous article.
What I want to say is this: do not view AI merely as a business tool. AI is becoming the core of productivity. Whoever controls AI controls productivity, and thus holds power.
In this sense, the value of edge AI devices far exceeds their price tags. They allow ordinary people to possess their own intelligent production capabilities, rather than relying entirely on the "charity" of the cloud.
In 2026, the world is becoming increasingly turbulent. In this environment, owning things you can control is rational.
Edge AI is one such thing.