DeepMind CEO: AI will bring abundance, but there will be a decade or so of reshuffling first.

The lights in the office have just gone out, and the lights at home are on again.

As the day ends, it is the moment when his real work begins.

Demis Hassabis revealed his work schedule in the latest video interview by Fortune:

“Around 10 p.m., I'll start my second round of work and keep going until 4 a.m.”

During the day, he has back - to - back meetings with hardly any breaks; at night, he sets aside six hours to do just one thing: think.

He has maintained this work schedule for ten years. And in today's AI industry, this rhythm seems even more necessary.

The Google DeepMind he leads is at a critical moment:

• The monthly active users of the Gemini App have reached 650 million.
• The AI Overview of Search reaches 2 billion people a day.
• The most powerful model, Gemini 3, ranks among the top in multiple key rankings.

“We're making rapid progress.”

His tone is calm, but his words conceal the anxiety of the entire industry: the closer the technology gets to the critical point, the faster the reshuffle will be.

In this interview, Hassabis talked about competition, bubbles, computing power, and talent, and also speculated on how AI will reshape science, medicine, and the form of future devices.

Section 1 | The reshuffle has begun: competition is accelerating

Let's start with competition.

There is a very crucial sentence in the interview:

“The lead may only last for a few months.”

This is the current state of the AI industry. The gap between top - tier laboratories is getting smaller, and the leading advantage can be broken at any time.

Model competition: the update speed determines the position

Hassabis is very satisfied with the performance of Gemini 3, but he also admitted that the competition has never been as fierce as it is now.

Because everyone is in a sprint: model updates have changed from “once a year” to “once every few months”, and new capabilities have evolved from single - point breakthroughs to all - round expansions, with code, multi - modality, video, and voice being iterated simultaneously.

He didn't directly say “we must accelerate”, but his work schedule of working until 4 a.m. every night says it all.

In this rhythm, being one step slower in model capabilities will relegate you to the second - tier.

Shortage of computing power: chips have become a new threshold

In the interview, he repeatedly mentioned:

The demand is unprecedented. Even Google's chips are far from enough.

This is the biggest bottleneck the entire industry is facing. To build stronger models and make products truly viable, computing power cannot be avoided.

What does this mean for enterprises?

The focus of the budget has shifted: from buying servers to scrambling for computing resources.

Large companies can lock in supplies in advance, while small companies have to wait in line.

Whether you can do it depends on whether your system can handle it.

Without sufficient computing power, even the best ideas can't be put into practice. This is another battlefield: whoever secures the computing power entrance gets the qualification to continue competing.

Talent competition: money is just the foundation, mission is the bargaining chip

The industry has reported that a researcher received an offer of $100 million. This is the first time such a figure has appeared in the AI industry.

But Hassabis believes that what can truly retain top - tier talent is a sense of mission and work that can have an impact.

Money is of course important. However, at this level, the attraction between individuals and teams comes more from the ability to participate in cutting - edge research, turn research into products used by hundreds of millions of users, and solve real problems in medicine and materials.

Top - tier talent values value and influence, and this standard is changing the employment rules of the entire industry.

For ordinary people, future competition depends on whether you can get closer to higher - value scenarios. The job title no longer matters.

During the reshuffle period, old jobs will disappear, but the requirements for each person's core competencies will be higher; teams will be reorganized, and a large number of new opportunities will emerge.

Section 2 | Abundance is taking shape: three technological paths

While the reshuffle is taking place, opportunities are also emerging.

In the interview, Hassabis pointed out three directions that have turned from concepts into reality.

Multi - modality assistant: a new gateway to understanding the world

When asked what excited him the most, Hassabis was very clear: multi - modality. This has been their goal from the start, and it will become a portable assistant.

What's the significance of multi - modality?

It enables AI to move from answering questions to understanding the environment. It can see, understand, and respond to the real world.

Specifically:

AI has evolved from a search box to smart glasses and become a portable device.

It has shifted from passively waiting for instructions to actively understanding your situation.

It has transformed from a software tool into a portable thinking partner.

Why now? Google experimented with smart glasses more than a decade ago, but it was too far ahead of its time and lacked killer applications. Now, the time is right, and the AI assistant is that application.

Google's collaborations with Warby Parker and Gentle Monster are turning this ability into physical products.

This means that future tools can handle more trivial tasks for you, allowing you to focus your time on more valuable things. Your personal productivity will multiply, and efficiency will skyrocket.

AI drug design: calculation replaces trial - and - error

In addition to device entry points, Hassabis also sees breakthroughs in the medical field. He listed a series of specific progress:

• Isomorphic Labs has entered the pre - clinical stage of multiple drugs.
• It is collaborating with Johnson & Johnson, Eli Lilly, and Novartis simultaneously.
• There are approximately 17 drug projects in progress.

AI can now directly design drug molecules in the computer. Traditional drug R & D can take over 10 years to find a molecule from a target; AI can compress this process into a few months.

More importantly, AI can detect molecular structure features and drug design paths that humans can't see. This is a brand - new scientific research method.

AI fundamentally shortens the treatment R & D cycle, enabling humans to conquer more diseases.

New material breakthrough: an automated scientific research closed - loop

Near the end of the interview, Hassabis mentioned:

“We will establish an automated materials laboratory in the UK.”

AI is no longer just predicting proteins; it's starting to design materials.

What can new materials change? Battery life, chip conductivity, hydrogen storage, superconductors, and new energy materials. Breakthroughs in these areas will have a chain reaction, benefiting multiple industries simultaneously.

The role of the automated laboratory is to form a closed - loop:

AI design → robot synthesis → equipment measurement → data feedback to AI

A traditional experiment cycle can take weeks or even months. This closed - loop can operate 24/7, constantly iterating and optimizing, and the research speed will be much faster.

Hassabis described three futures in the interview:

An assistant that can understand the world

An ability to bring drug R & D back within a controllable range

A scientific research production line that can continuously produce new materials

These three directions are what he calls “abundance”. The changes may not be obvious in the first three years, but after a decade, the accumulated quantitative changes will trigger irreversible qualitative changes.

Section 3 | Why must there be a reshuffle before abundance?

The reshuffle is happening, and abundance is approaching, but why is this the sequence?

Hassabis gave a time estimate:

“As early as 2030, there may be a 50% chance of reaching AGI.”

That is to say, we still have to wait a few years for abundance. However, technology is evolving rapidly, market demand is booming, everyone wants to seize a position, but resources are limited.

What will happen before the arrival of AGI?

The technological critical point is approaching but won't be achieved immediately

It's about 4 - 8 years from now to AGI.

Hassabis's prediction is relatively conservative: there won't be a sudden leap.

In the short term, AI won't replace all jobs at once, but there are new changes every year. The content of a certain job changes, a certain product needs to be redesigned, and a certain team finds that the original process is no longer applicable. The changes seem small each year, but the gap widens over several years.

This process is the reshuffle. Some companies will stand firm, while others will be eliminated.

Bubbles coexist: the overall valuation is reasonable, but individual projects are overheated

Regarding bubbles, Hassabis's judgment is sharp: the overall demand in the AI industry is real, but the valuations of some early - stage projects are indeed too high.

Why is there such a disparity?

Because two trends are happening simultaneously. On the one hand, model call volumes, user scales, and corporate purchases are all surging, indicating real market demand. On the other hand, a large number of early - stage companies that haven't completed technical verification have received tens of millions of dollars in financing just because they fit the concept.

Capital is eager to seize entry points, often leading to ineffective pricing.

The result is that the industry as a whole is growing, but projects with inflated valuations will be eliminated. The teams that survive are those that can prove their commercial value. Money follows value.

Route differentiation: applications yield quick returns, while cutting - edge research determines the landscape

In the interview, Hassabis mentioned that Chinese teams are more focused on application implementation, while Western teams are more focused on cutting - edge breakthroughs.

• Application path: quick cash flow, clear scenarios, and easy to scale.
• Cutting - edge path: high technical barriers, long payback periods, and can reshape the industry once successful.

Both paths have their value. Focusing on applications can keep a company alive in the short term, but long - term competitiveness depends on cutting - edge breakthroughs.

There are risks in taking only one path: companies that only focus on applications may be left behind in technology; teams that only do research may run out of money before finding a commercialization path.

Teams that can truly weather the cycles need to implement applications quickly and also have technological accumulation.

In the next decade, it's not about quantity but precision. Technological breakthroughs may only take a few years. Capital is also becoming more patient, more concerned about whether you can prove your concept rather than just listening to it. For individuals, the opportunities lie in new technologies, new sciences, and new materials. The key is whether you can seize them.

The choice of technological route determines who can reach the end.

Abundance will come, but it won't be evenly distributed to everyone.

Before that, this years - long reshuffle will first conduct a brutal screening of the entire industry.

Conclusion | The reshuffle is here, and abundance is ahead

Before the end of the interview, Hassabis said:

“I hope I still have time to think seriously.”

The closer we get to abundance, the more we need to stay calm.

In the past two or three years, model iterations have accelerated, application implementation has sped up, and capital has been re - allocating. The industry seems bustling, but the underlying logic is changing: chips have become a bottleneck, AI is driving scientific progress, and inflated valuations are being adjusted.

The inflection point is right now.

The future won't arrive overnight, but the path will become clearer and clearer.

Original links:

https://www.youtube.com/watch?v=BhfTQXMtoZw&t=977s

https://www.youtube.com/watch?v=-RPbxvz6sB8

https://x.com/demishassabis/status/2021223548744822972?referrer=grok - com

https://www.businessinsider.com/google - deepmind - ceo - demis - hassabis - work - routine - sleep - six - hours - 2026 - 2

Source: Official media/Online news

This article is from the WeChat official account “AI Deep Researcher”. Author: AI Deep Researcher. Editor: Shen Si. Republished by 36Kr with permission.