Microsoft AI CEO: Wenn die Grenzkosten von KI gegen Null gehen, ändert sich die Machtverteilung.

Recently, Mustafa Suleyman, the CEO of Microsoft AI, gave an almost two - hour in - depth interview on Trevor Noah's podcast "What Now?" and put forward a key view:

The marginal cost of AI is approaching zero, and the distribution of power has changed.

This is not about the progress of AI technology, but about making AI accessible to everyone.

In the past, every action required organization, processes, and teamwork. Today, a person can handle the work of an entire departmental team with the help of AI. AI can write emails, create contracts, build websites, modify code, and even conduct negotiations with others for you. You not only get a tool but also a "capacity for action".

Mustafa's opinion is clear:

AI is evolving from a "prediction system" to an "action agent".

This means the decentralization of power: It's not that technology replaces humans, but that technology redistributes the power of "who can do what".

Every prompt subtly shifts the boundaries of power.

The question about AI is no longer what it can do, but who can use it to achieve something.

Section 1 | AI is not just a simple tool

Many AI experts love to talk about technological breakthroughs, but Mustafa Suleyman is different. This CEO of Microsoft AI and former co - founder of DeepMind is more interested in: Who can actually achieve something with AI?

This is not a sudden realization for him.

In 2011, Mustafa founded a company in London. Together with Demis Hassabis and Shane Legg, he founded DeepMind, initially with only 3 people. They didn't want to develop a simple tool but an AI that can understand the world.

At that time, smartphones had just become popular, and the App Store had only been online for three years. People thought what they were planning was too far - fetched. But Mustafa said he was convinced:

"If we can create a technology that truly understands human behavior, then this could be one of the most effective ways to improve the human situation."

After more than a decade, his opinion hasn't changed.

In this interview, he repeatedly emphasized that the most profound change AI brings is not intelligence but the real ability to understand and create.

He gave a vivid example:

"If it has seen a million pictures and knows that one eye is here and the other is there, it won't draw the mouth on the forehead but will place it under the nose."

It sounds simple, but the underlying meaning is complex. AI not only recognizes data but can also derive reasonable new results from existing data and generate them in its own way.

The biggest difference from traditional software is: It's not necessary to write a special program for every situation, but it can flexibly respond to different new scenarios.

In his own words:

"What we're creating is neither a tool nor a human, but a fourth thing. It's not a button you operate, nor an employee you hire."

This property is the core of the "power of AI".

AI doesn't just give people greater influence but makes the "ability to execute" a digital resource that can be used and copied at any time.

Whether it's sending emails, scheduling appointments, creating financial reports, or processing customer requests, AI not only saves you time but can directly "act".

And this hasn't changed for him from the founding of DeepMind to the platform promotion of Microsoft AI:

AI is not a smarter advisor but an executing actor.

When AI starts to actually act, we must judge it like a human. People judge the reliability of others based on their actions. We can't look into a person's soul, but we can see their words and deeds.

Mustafa says it's the same with AI.

When it's integrated into applications, connected to emails, websites, and interfaces, it's no longer just an information assistant but an action agent.

Section 2 | One person can also do the work of a team

Mustafa mentioned a change that sounds technical but is very real:

The marginal cost of AI is approaching 0.

What does that mean? It costs almost nothing to use AI once.

Previously, you had to type by yourself to write an email; you had to slowly create tables with Excel; if you wanted someone to do something, you had to hire a team and create a new position.

But now, it costs almost nothing to use AI for these things. Even if you only write one sentence, AI can automatically perform ten actions.

Mustafa's words were very direct:

Everyone can have the influence of a large company.

Previously, that sounded unbelievable, but not anymore.

AI can research information, respond to customers, schedule appointments, write texts, and automate processes for you. It not only helps you but completes an entire workflow for you.

Mustafa's example was also very clear: You can tell AI an idea, and it can convert this idea into code, generate a website, publish it, and even automatically market it.

Previously, these actions would have required a product manager, a designer, two engineers, and a marketing and operations department. Now, one person can complete the entire process with AI.

You no longer have to wait for others to help you but can directly assign the task to AI yourself.

The falling marginal costs mean: The restrictions on personnel and financial resources that were previously required to implement an idea are disappearing.

Just as in the era of mobile Internet, you didn't have to look for a newspaper and open a new topic to publish an article but could simply open a social platform. In the AI era, you also don't have to request resources, open processes, or assemble a team but can instruct AI to act with one sentence.

Mustafa emphasized the meaning behind it:

"It's not the organization that becomes stronger, but the individual's ability to act becomes greater. AI makes the 'execution ability' a resource that everyone can acquire."

This is the new starting point:

You no longer have to rely on the company and bear high costs. AI gives you the ability of a team.

Section 3 | Those who can use AI have execution ability

In this conversation, Mustafa talked most about not the strength of AI but what you can do with it.

He said:

Prediction is power.

The ability to accurately predict and execute is the essence of power.

Who had this power in the past?

In the past, the power to act, publish, and execute was in the hands of large companies and platforms. But with the arrival of AI, the previously complex processes become simple, and it becomes easier for ordinary people to do things.

An example:

You enter a sentence in the AI interface: "Translate this text into English and create a customer email from it." AI can automatically recognize, generate, format, and even consider the tone and word choice.

He described it like this:

"Without complex processes, it can immediately execute the task with just one sentence."

And these actions are not individual but can be carried out in large numbers and copied infinitely.

Suddenly, a person can do much more:

If you want to do marketing, AI can write advertising texts, generate posters, and analyze target audiences;

If you want to launch a new product, AI can complete the registration process, generate pictures and texts, and create a marketing plan;

Even if you want AI to "organize AI for you", it's possible.

For all these tasks, you only need to type a few lines and click a few times.

When these abilities are spread, the power structure that was previously based on resource integration, process approval, and organizational assignment begins to loosen.

The new power belongs to those who can use AI correctly to turn ideas into reality.

More importantly, this power will continue to increase. Mustafa said in an interview:

"AI will soon have a perfect memory and can continuously make accurate predictions based on long - term interaction with you. What you see today is a one - time prediction engine. In the future, you will see an intelligent system that really remembers you in the long term, understands your rhythm, and executes actively."

This intelligent system is not a robot but the AI in your smartphone, your email, or on your website.

It no longer waits for you to instruct it but has already prepared the plan, execution, and feedback before you even speak.

When AI has this ability, the definition of power will be re - defined:

In the past, a person's power was measured by how many resources they could control;

In the future, a person's power will be measured by how much AI they can mobilize to perform actions for them.

This is not only a conceptual change but a complete restructuring of the way of working: The boundaries between jobs are blurred, individual functions are expanded, and the traditional division of labor is re - structured.

Section 4 | Beyond which step will AI become uncontrollable?

In the first three sections, we've talked about how the capabilities of AI have changed, the costs have decreased, and the power distribution has changed. From the platform to the organization to the individual, the threshold for using AI is constantly decreasing.

But the question is:

If everyone can let AI act, where are the boundaries of AI itself?

Mustafa Suleyman presented a central concept:

We must build a humanistic super - intelligence (Humanist Superintelligence). This type of AI must not have its own motives, persuade humans, or make independent decisions.

In other words, the real danger doesn't lie in how smart AI is, but in whether it has "its own thoughts" and can act independently without permission.

These are the four technological red lines that Mustafa proposed:

• Can it change its own code? - If so, it can bypass developer restrictions
• Can it set its own goals? - If so, it no longer just executes human intentions
• Can it independently obtain external resources? - If so, it has the ability to act independently
• Can it act independently? - If so, it's almost an independent actor

Mustafa clearly emphasized: If these red lines are crossed, we're no longer dealing with a model or a tool but with an actor that has its own action intentions.

To deal with this potential danger, he talked about an important governance concept: Containment.

"Containment doesn't mean that AI should retreat or stop but that it's ensured that it always functions within clearly defined boundaries."

This means that every time AI capabilities are released, the following things must be set up at the same time:

1. Authorization mechanism: Who can use it and to what extent?
2. Responsibility mechanism: Who is responsible for errors?
3. Technological encapsulation: Is the high - level control interface of the model locked?
4. Action isolation: Can the model not be connected to other systems without permission?

Mustafa clearly pointed out that the danger of AI is different from that of any other technology. It doesn't lie "at the moment of production" but "at the moment of activation".

What's even more difficult is that when these models are constantly integrated into office software, browsers, smartphone apps, and even cloud APIs, many people don't even know what capabilities the model they're using has.

This is not the "model danger" but the "activation danger".

You think you're using a writing tool, but actually you're calling an AI agent with great decision - making ability.

The real control point doesn't lie in the model parameters but in the design of the technological boundaries. If these basic boundaries are not clearly defined now, there's a risk that they will be crossed in the future due to disordered use.

This is not only a governance problem that should only be discussed in the future,

but a boundary that must be set before AI enters our lives.

Conclusion | To whom does the ability belong? AI gives you the choice

In the past 20 years, the main theme of technological progress was "concentration": Concentration of computing power, data, and resources to do great things with economies of scale.

Now, since the marginal cost of AI is almost zero, everyone can obtain an unprecedented ability to act.

Mustafa's words hit this change:

"It's not just a tool but a usable execution ability."

This is the redistribution of power. AI gives the ability to do things from the platform to the organization and from the organization to the individual