TradingBase.AI Column | Why AI trading seems to be getting stronger, but fewer people are making money

If you only look at the data from the past few months, it’s easy to draw a conclusion: AI trading is becoming stronger and stronger.

Some institutions and strategies achieved extremely high returns at the beginning of 2026, and even some AI strategies delivered more than 200% performance in a single quarter—far beyond the return range of traditional strategies. At the same time, the share of AI quant in the market continues to rise, and automated systems have become the mainstream mode of execution.

From the surface, this looks like a phase of “a technological dividend explosion.”

But if you shift your perspective a little deeper, you’ll find a completely different reality: AI trading is getting stronger, but the systems that make money are becoming fewer.

An overlooked change: returns are becoming “concentrated”

In the past quant market, although there were differences, overall it was “dispersed profitability.” Different strategies, different styles—there were opportunities to make money in different stages.

But now, this structure is changing.

One direct result brought by AI strategies is a massive improvement in efficiency. But on the other side of efficiency gains is intensified competition. When a large number of systems enter the market at the same time and use similar data, similar training methods, and similar optimization objectives, the market’s “effective opportunities” get compressed quickly.

The result is: returns start to concentrate in a very small number of systems that execute optimally.

This is not an isolated phenomenon—it’s a structural change. AI hasn’t made opportunities more plentiful; it has made opportunities get “consumed” faster.

Why “the model gets stronger,” paradoxically makes it harder for most people to make money

Intuitively, the stronger the model, the easier it should be to make money in the market. But reality is exactly the opposite.

After AI capabilities become widespread, the model itself no longer constitutes a barrier. Most systems can achieve “roughly correct” performance. The problem is that when everyone is “roughly correct,” what ultimately determines returns is no longer judgment—it’s execution.

In today’s market, millisecond-level differences, path selection, and slippage control all directly determine the final outcome. In other words: it’s not whether you’re right in your analysis, but whether you execute fast enough and accurately enough.

That’s also why more and more strategies perform well in backtests but can’t deliver in live trading. Because backtests don’t reflect execution differences—real markets do.

AI doesn’t bring “more fairness,” it brings “more extremes”

AI trading also has another seriously underestimated impact: it makes the market more “extreme.”

On one hand, the advantages of excellent systems get amplified. They can obtain data faster, execute more precisely, and run more stably—allowing them to continuously accumulate an edge.

On the other hand, most systems fall into “ineffective competition.” Their strategy logic may be correct, but because their execution efficiency is insufficient or their system structure is not well designed, they ultimately can’t translate that logic into profits.

This kind of structure leads to a result: the strong get stronger, while the weak find it increasingly hard to survive.

From the perspective of market structure, this is closer to “top-end monopoly” rather than an “overall improvement.”

The real problem: are you competing with “systems of the same kind”?

If you look one layer deeper at this phenomenon, you’ll find an even more realistic problem: you are not competing with the market—you are competing with “systems that look a lot like yours.”

Because:

Similar data sources

Converging model architectures

Similar training methods

Same optimization objectives

This means most systems will eventually converge toward similar behaviors. And once behaviors are similar, the market shows a very typical phenomenon:

The same batch of opportunities is being snatched by the same batch of systems at the same time—not that “everyone makes money,” but that only the system with the best execution makes money, while the other systems passively pick up the results.

The next stage of AI quant trading is not “stronger,” but “more different”

If you summarize this round of changes in one sentence, it’s actually not that “AI is getting stronger,” but:

AI is making the market more crowded.

And in a crowded market, true advantage never comes from “being stronger”—it comes from being “more different.”

Future competition isn’t only about model capability, but also about:

Whether you have differentiated data

Whether you have independent execution paths

Whether you can avoid crowded trading

Whether you have structural advantages

These capabilities won’t necessarily show up in marketing, but they will show up directly in returns.

Conclusion

AI trading is indeed getting stronger, and there’s no question about that. But market changes are never one-way.

When technology spreads, advantages get diluted; when efficiency improves, competition gets amplified.

The real change isn’t “making money becomes easier,” but:

The threshold for making money is being redefined.

The future market doesn’t belong to the most intelligent systems—it belongs to:

Systems that can still find differences in a crowded environment.