In early 2025, Silicon Valley collectively lost sleep.
Not because of a stock market crash, nor because another tech titan derailed. It was because a Chinese company delivered 90% of the performance using one-tenth the capital of US giants.
Silicon Valley’s logic has always been simple and brute-force: cash is king. Whoever has the most GPUs and pays the highest electricity bills builds the strongest AI. This paradigm dominated the entire industry for years. Until DeepSeek showed up and kicked over the table.
Everyone said, “Don’t fight them; you can’t win.” And then, they won anyway.
What makes this so fascinating?
Round 1: Your Stance Dictates Your Battle
Everything in Kendo begins with the stance, or Kamae. The posture you choose determines the kind of war you fight.
OpenAI’s stance: Jodan (Upper Stance). The bamboo sword is raised high above the head. Overwhelming downward pressure. Intimidating by design. Its logic is the Scaling Law: more data, more GPUs, more power—and the model gets stronger. Simple, violent, direct. In 2024 alone, OpenAI spent over $7 billion on compute.
Seven billion. Think about that scale. You could probably buy all the cod in Iceland with that.
DeepSeek’s stance: Gedan (Lower Stance). The sword tip points down, seemingly powerless. You might think they forgot their gear. But it hides an invisible blade. What was the training cost for DeepSeek V3? $5.57 million. Not even a fraction of OpenAI’s rounding error.
Jodan relies on power. Gedan relies on efficiency. Normally, power crushes efficiency. But when efficiency is pushed to a critical threshold—the rules of the game flip.
Round 2: Your Imagined Moat Might Just Be a Ditch
OpenAI always believed its moat was impenetrable: “I have 10,000 H100s, and you don’t. Therefore, I win.” DeepSeek employed an extreme execution of MoE (Mixture of Experts). In plain English: you don’t mobilize the entire army every time. Out of 671B parameters, it only activates 37B per inference. Like a shrewd general, it avoids massive brawls, deploying only the most suitable specialized squad for each strike.
The result? It costs merely $2.19 per million output tokens. OpenAI’s GPT-4o? $15.
A nearly 7x difference. Silicon Valley’s compute moat turned into a ditch overnight.
In Kendo, this is called Irimi—entering, changing the angle, and slipping directly into the opponent’s face. You thought I would charge head-on? Naive. I slipped in from the side. By the time you react, my blade is already at your throat.
Lean vs. Fat: This Is Not Just an AI Company Problem
There is a specific type of person in the dojo.
Not young. Not fast. Not particularly strong. But when you face off against them, your hair stands on end. Because they have zero excess. Every movement is precise. Not an ounce of energy is wasted.
Japanese uses a single word to describe this state: Kare (枯).
It doesn’t mean withered. It means stripped to the absolute essentials. Like a thousand-year-old pine tree—sparse in foliage, but every root is wedged deep into the bedrock, unshakeable by the wind.
OpenAI’s models are “fat.” The aesthetics of brute force. More parameters, more data, more compute. DeepSeek’s models are Kare. Lean. Not a single gram of excess fat.
“Fat” means high costs and high barriers to entry—you are forced to find equally wealthy clients. “Lean” means low costs and rapid deployment—your client is the entire world.
What makes the lean approach so terrifying? You cannot fight it by getting “fatter.” You must also become lean—but you are already used to being fat.
This is exactly how business works. Those companies in the industry shouting, “We need to get bigger, stronger, and more comprehensive”—have they ever considered that perhaps their competitor isn’t trying to outgrow them, but to out-lighten them?
So, What Does This Battle Have to Do with You?
Some people might say, “I don’t work in AI. What does DeepSeek have to do with me?”
Everything.
DeepSeek proved something that allows everyone with “no resources” to stand tall: Money is not the ultimate barrier. Efficiency is.
Are you a small company fighting industry giants for market share? Don’t compete on “scale”—compete on “precision.” Are you a professional competing with elite-school grads for a job? Don’t compete on credentials—compete on the dirty, grueling work they are too lazy to do because of their path dependency. Are you a regular parent watching others spend $30,000 a year on private tutoring for their kids? Don’t compete on capital input—compete on methodology.
Poverty is never your excuse. Laziness is.
In this era where everything is being repriced— Whoever pushes efficiency to its absolute limit holds the power to launch a devastating asymmetric strike. And whoever is still flaunting how “fat” they are—is the one being disrupted without even realizing it.
Your Practical Exercise for the Week: Take a task you are currently working on that you believe “absolutely requires X amount of money or time” to complete. Forcefully slash the resources or time by 90%. Then, force yourself to spend ten minutes brainstorming three extreme solutions to the question: “How else can I pull this off without relying on money or pulling all-nighters?” Even if only one is viable, your moat just grew an inch deeper.
