The AI era is becoming increasingly polarized: the rich are getting richer, and the poor are getting poorer.

Writing by: jiayi

AI has changed our habits of living, and that’s a fact.

Using AI to write emails, create PowerPoint presentations, search for information, even craft social media posts—we’ve grown accustomed to AI’s presence, just like WiFi, it feels natural.

But few stop to ask: Are the AI tools you use the same as others are using?

The illusion of “fairness” in the AI era

Silicon Valley loves to tell a story: AI gives everyone a super assistant, knowledge is no longer a privilege of the few, and everyone is equal.

It sounds beautiful. But the truth is—AI is inherently unfair at its core; it’s a competition of financial resources.

From chips to computing power, from model training to token consumption, every step of AI development costs money.

A single NVIDIA H100 chip costs over $25,000. Training a GPT-4 level model costs over a hundred million dollars. Every question you ask AI involves burning tokens—and tokens have a price.

Claude Opus costs $5 per million tokens for input and $25 for output. ChatGPT Pro is $200/month. Plus Perplexity, Cursor, Midjourney… a heavy AI user easily spends over $500 monthly on tools.

Some spend $5,000 a month to build competitive barriers with AI; others think using the free ChatGPT makes them up-to-date.

They’re not even on the same track. Not even in the same game.

At the national level: structural gaps are irreversible

This logic is even more brutal when applied to countries.

AI arms races require three things: chips, computing power, and talent—all of which demand huge capital.

The US controls over 70% of global AI computing power. China is catching up, but chip bans are choking it. As for most developing countries—out of 46 emerging markets, the cost of entry-level broadband accounts for 40% of monthly income.

When a young person in Nigeria can’t even afford stable internet, what “AI equality” are we talking about?

94% of high-income countries have internet access, only 23% of low-income countries do. 84% of high-income countries have 5G coverage, only 4% of low-income countries.

For third-world countries, the starting line in the AI era isn’t just a step behind—it’s a complete exclusion.

This structural gap cannot be bridged by effort alone.

On an individual level: your ceiling is being redefined by AI

The same logic applies to each person.

I wrote in my Twitter bio: Personal ceiling = worldview + cognition + practical ability.

What has AI done to these three?

▶️ First, AI has solved many efficiency problems.

What used to take a week to produce a report now takes a day. Coding from scratch used to be hard; now AI helps you set up frameworks. In terms of efficiency, AI is indeed leveling the playing field.

▶️ But second, AI greatly amplifies cognitive gaps.

The same AI tool—what you ask, how you ask, whether you can judge if the answer is right or wrong—depends entirely on your existing level of cognition.

A person with deep understanding uses Claude for research, knows what questions to ask, how to follow up, which answers need verification. AI saves him 80% of execution time, which he then spends on deeper thinking.

What about someone with shallow cognition? They toss questions to AI, accept whatever it gives, and move on. No thinking involved. Over time, they stop thinking altogether. AI doesn’t make them smarter; it makes them lazier and dumber.

▶️ Third, the gap in output quality widens exponentially.

Based on your existing cognition, asking AI questions results in outputs that vary hugely in depth, accuracy, and real-time relevance. Using Claude Opus, one person produces deep insights; another produces superficial nonsense that looks convincing.

A study at Aalto University in Finland found that the more people use AI, the more they tend to overestimate their abilities. AI makes you “feel” stronger—outputs seem professional and smooth. But if you lack the ability to judge quality, you’re just producing “refined mediocrity.”

Thus, worldview, cognition, and practical skills—these three dimensions are being amplified infinitely in the AI era.

The smarter you are, the deeper your understanding, the wealthier you are—using better tools, you can widen the gap even more. Conversely, under AI “assistance,” others become lazier, shallower, and poorer.

Cost × cognition: a double divide stacking up

Here’s a logical chain many haven’t fully grasped:

Money determines what level of AI you can access → The level of AI determines the quality and depth of information you get → Information quality defines your cognitive boundary → Cognitive boundary influences your decision-making quality → Decision-making quality affects how much money you can make.

It’s a closed loop. The rich get richer, the poor get poorer.

The illusion rate of free ChatGPT is nearly 40%. That means, out of 10 questions, 4 answers are fabricated. Paid GPT-4 has a hallucination rate of 28%, and the latest version has dropped by 45%.

Decisions made with free versions versus those with Opus—over time, they lead to two completely different life trajectories.

There’s always a huge information gap in this world. AI hasn’t eliminated it; it has turned it into a paywall.

Those who can bypass restrictions and those who cannot are living in two different worlds

A personal observation I find quite poignant:

The reason you’re reading this article is probably because you can bypass the Great Firewall and browse Twitter.

But think about it—how many people around you can’t bypass it? When you talk to them, do you already feel that your cognition is on a different level?

This isn’t about IQ. It’s a long-term cognitive divergence caused by different information environments.

One person is exposed daily to cutting-edge information, in-depth discussions, and top content creators worldwide. The other sees algorithm-fed short videos and filtered feeds.

Over five or ten years, their ways of thinking, judgment, and worldview become completely different.

The AI era further amplifies this gap. Those who can bypass the firewall use Claude, Perplexity, and the best global AI tools. Those who can’t—ChatGPT is blocked in China, Claude is blocked in China—they can only use localized substitutes or buy through middlemen at higher prices.

The “walls” of the AI era aren’t just physical firewalls. There are language barriers—advanced AI models optimize far better in English than in other languages. There are paywalls. Algorithmic echo chambers. Every wall pushes people into different worlds.

Research from Stanford shows that non-English speakers need five times more tokens to process the same content with AI. That means, for the same amount of money, they get less information, and of lower quality.

The scariest part: you’re already falling behind, but you don’t realize it

This is the point I most want to emphasize.

Even with the free AI, you can ask questions, get help writing, and search. So you might think—“I’m using AI too, I’m not falling behind.”

But the free version has shallower reasoning, more hallucinations, and older information. The answers you get “look” correct but are often riddled with fallacies and errors.

It’s like two people running. One is genuinely moving forward; the other is jogging in place on a treadmill. Both think they’re running, but only one is actually progressing.

In psychology, there’s a concept called the Dunning-Kruger effect: the less you know, the more you think you know. AI amplifies this tenfold—you rely more on AI, the more confident you become. But you’re losing the ability to think independently; you just don’t realize it.

This is the cruelest part of the AI era.

It’s not that AI will replace you. It’s that those with better AI and deeper understanding will leave you far behind. And you might not even realize how you fell behind until the day you’re eliminated.