The Strongest AI Will Not Necessarily Rule the World

The Strongest AI Will Not Necessarily Rule the World

Rethinking the Nature of AI Competition Through Google

Imagine two swords.

One is an enormous, heavy greatsword. A proper swing could cut through almost anything.

The other is a moderately sized sword, less powerful than the greatsword.

Compare the weapons alone, and there is no contest. The greatsword wins.

Now change the conditions slightly.

The greatsword’s wielder has a narrow field of vision. They do not know where the enemy is. Their movement is restricted. Each swing takes enormous effort, and they need time before attacking again.

The other side has a slightly weaker weapon but can see almost the entire battlefield. They know who is where, where the gaps are, and can approach through countless routes. Their sword is light and cheap, so they can keep swinging.

Who is stronger across the whole war?

I would choose the latter.

Overwhelmingly.

AI is similar.

For the past few years, we have largely viewed AI competition through a single number.

Who is smartest?

Who codes better?

Who solves more math problems?

Who tops the benchmarks?

Those things matter, of course.

But that measures only the sharpness of the blade.

The more important real-world questions lie elsewhere.

What can that intelligence see?

How far can it reach?

How often does it appear in front of users?

How cheaply can it act?

And ultimately,

how much territory does it occupy between the moment a person forms an intention and the moment the action is complete?

Look at companies through these questions, and Google looks entirely different.

The Model Is a Weapon. Google Owns the Battlefield.

Google’s AI model need not be the world’s best at every moment.

Judging Google as a company through the Gemini model alone is, in fact, strange.

Google has Search.

YouTube.

Android.

Chrome.

Gmail.

Google Maps.

Photos.

Calendar.

Workspace.

Cloud.

And its own AI accelerators, TPUs.

These products do not exist in isolation.

Alphabet states that, as of 2026, it has 13 products with more than a billion monthly users, five of which exceed three billion. AI Overviews reaches over 2.5 billion monthly users, and Search’s AI Mode has already surpassed one billion. The Gemini app itself crossed one billion monthly users in August 2026.

Viewing Gemini solely as an app-versus-app fight with ChatGPT misses the structure.

From Google’s perspective, users do not even need to open the Gemini app.

They can encounter Gemini while searching.

Reading email.

Using a browser.

Creating a document.

Watching YouTube.

Using Android.

Google technology will even underpin Apple’s AI going forward.

In January 2026, Apple and Google officially announced that the next generation of Apple Foundation Models would be built on Google’s Gemini models and cloud technology, supporting future Apple Intelligence features including a personalized Siri.

This is a significant point.

Google is not merely a company with its own operating system.

It can also enter the AI foundation layer of a competing operating system.

At this point, the question changes.

“Where does Gemini rank on benchmarks right now?”

Is that really the most important question?

AI’s Real Power Is Intelligence × Surface Area

I think of AI’s practical power roughly this way.

AI Power ≈ Intelligence × Perception × Reach × Frequency × Economics

Intelligence.

What does it judge, and how well?

Perception.

What can it see about the world and its users?

Reach.

Where can it actually enter?

Frequency.

How often does it intervene in human activity?

Economics.

How cheaply can it repeat those judgments and actions?

The important point is that these factors behave more like multiplication than addition.

Even with intelligence of 100, information of 1 limits what can be done.

Even knowing the world well is of little use without a channel to reach users.

Conversely, once model intelligence passes a sufficient threshold—say 80 or 90—overwhelmingly greater perception and reach change the result.

Suppose the greatsword’s attack power is 120 and the smaller sword’s is 90.

But the greatsword sees only 5% of the battlefield, while the smaller sword sees 80%.

The comparison already looks different.

The latter can also reach far more places.

This is why, in AI competition, the area intelligence reaches can matter more than intelligence itself.

Google’s Real Asset Is Having Observed the Internet for So Long

Calling Search merely a “search service” understates Google’s asset.

For more than twenty years, Google Search has been a vast system observing what people wonder about, click, find satisfying, and search for again.

The U.S. Google search antitrust litigation made this difference concrete.

The court found that Google receives roughly nine times as many search queries as all competing search engines combined, and roughly nineteen times as many on mobile.

The court explained that the thirteen months of Google click-and-query data used by NavBoost, one of its core ranking systems, would take Bing approximately 17.5 years to accumulate at the same scale.

There is an even more important point.

This data is not merely logs sitting on a server.

According to the court, Google uses user data to decide which additional websites to crawl, expand its search index, rerank results, improve freshness, and train models that improve search quality.

A crucial structure exists here.

Many people use Google.

↓

Google gains more data.

↓

Search quality improves.

↓

More people use Google.

↓

More data is generated again.

This is not mere data ownership.

It is a closed reinforcing loop.

And AI is entering that loop.

Search Is Formidable Because It Is Where Human Intent First Appears

Search is special even among Google’s many products.

People type their thoughts into the search box.

“Tokyo hotels”

may mean someone is considering a trip.

“MacBook battery replacement”

may signal a problem.

“Dentist near Seoul National University Station”

may mean someone is about to act.

“How to incorporate a company”

may indicate preparations to start a business.

These are more than records of past behavior.

They reveal what people intend to do next.

This is the moment human intent surfaces.

For AI, that data has enormous significance.

A good AI is more than a system that answers questions accurately.

Ultimately,

it is likely to become a system that understands what the user currently wants and carries out the next action for them.

That is the likely direction.

And the search box is one of the interfaces that has received human intentions at scale for the longest time.

It Has Begun Connecting Personal Context Too

Google launched Personal Intelligence in 2026.

When users opt in, Gemini can connect information from Google services such as Gmail, Google Photos, YouTube, and Search to create personalized answers. It began as a U.S. beta in January and later expanded to other markets, including Korea.

This structure also entered Search’s AI Mode.

Users can connect Gmail and Photos, and Google has announced future Calendar integration. At I/O 2026, it announced expansion to roughly 200 countries and territories and 98 languages.

Another change occurs here.

If traditional Search

was a system for understanding the world,

that was its role;

AI combined with Personal Intelligence

begins seeing the world + me

at the same time.

The web contains hotel information.

Gmail may contain my flight ticket.

Photos may contain destinations I enjoyed before.

Search may show what I have looked up recently.

Calendar adds my schedule.

Maps adds places.

YouTube adds interests.

Even with identical intelligence, a system with this much context and one starting from an empty chat window with no information are unlikely to deliver the same results.

This is what I mean by differences in information surface area.

A Model Being 5% Smarter Is Not the Same as Seeing Ten Times More of the World

The AI industry likes benchmarks.

They are easy to measure.

Model A scores 87.

Model B scores 84.

It is then easy to say A is better.

Reality is not that simple.

Suppose an AI is deciding where I should have lunch tomorrow.

Model A has the world’s best reasoning ability.

But it does not know where I am.

Or my schedule.

Or what I like.

Or which restaurants are nearby.

Or whether reservations are available.

Model B reasons slightly less well.

But it is connected to location, schedule, maps, restaurant information, and past preferences.

Which is more likely to give a better answer?

A few points of model IQ are not what matters here.

What matters more is how accurately the problem was observed before attempting to solve it.

Management is similar.

Even a brilliant CEO will struggle to keep making good decisions without accounting records, customer data, or knowledge of what employees are doing.

A sufficiently capable person with real-time visibility into the entire company, however, becomes powerful.

Intelligence alone does not move the world.

It needs the ability to observe.

And After Observing, It Must Be Able to Act

Go one step further, and AI competition does not end with information either.

It is about action.

Far stronger than an AI that merely searches well

is one that searches,

compares,

checks schedules,

books,

pays,

sends email,

creates documents,

and continues into the next action.

Google is already connecting its products in this direction.

Universal Cart, unveiled at Google I/O 2026, demonstrated a direction in which people collect products across Search, Gemini, YouTube, Gmail, and other Google surfaces, track prices and inventory, and connect with payment infrastructure such as Google Pay.

This is not about one shopping feature.

It reveals Google’s vision for AI.

Moving from a search engine that finds information to a system inside the process of human action itself.

Search ends in action, not a link click.

This is likely to be AI’s ultimate battlefield.

Google Even Makes Money When Competitors Grow

There is another interesting part.

Google does not only operate Gemini.

It also sells infrastructure to rival AI companies.

In 2025, Anthropic announced a major expansion of Google Cloud usage, planning to use up to one million Google TPUs, with capacity well above 1 GW and an agreement worth tens of billions of dollars. Anthropic also explicitly described a multicloud structure retaining AWS as its primary cloud and training partner.

In 2026, Anthropic further expanded next-generation computing capacity with Google and Broadcom.

An interesting structure.

If Gemini succeeds, Google earns.

If Claude grows and uses more Google TPUs and Cloud, Google earns again.

Of course, Anthropic uses multiple suppliers, including AWS and NVIDIA, so this does not mean Google controls everything.

The important point is different.

Google is both a player and a supplier of part of the stadium.

That is structurally very different from a model company alone.

Why the U.S. Government Ordered Data Sharing

If this information surface is such a strong competitive advantage, a question follows.

How do the U.S. government and courts see it?

Interestingly, an answer has already arrived.

In the Google search monopoly case, a U.S. court ordered remedies in 2025 requiring certain search-index and user-interaction data to be provided to qualified competitors.

The Department of Justice explained that this was necessary to help search and AI competitors narrow Google’s data advantage.

I find this highly symbolic.

Even regulators are treating Google’s search data as an asset affecting competition, rather than a mere by-product.

If model performance alone mattered, there would be little reason for an order to share search-index and user-interaction data to matter so much.

The antitrust case ironically demonstrates the importance of information surface area in the AI era.

This Does Not Mean Model Performance Is Unimportant

At this point, someone could object:

“But if the model is stupid, can it do anything?”

Correct.

Model performance matters.

Frontier performance differences can be crucial in coding, scientific research, long-running agent tasks, mathematics, and demanding corporate knowledge work, where a single error is costly.

Some battles definitely require a greatsword.

The mistake is assuming that whether its attack power is 100 or 105 decides every war.

Once an AI model passes a certain performance threshold, other variables rapidly gain importance.

Price.

Speed.

Personalization.

Distribution.

Tool access.

Context.

Reliability.

And the scope of possible action.

Their influence is even greater in consumer AI used by billions.

Developing the world’s best model does not suddenly give you Android, Chrome, Search, Gmail, Maps, and YouTube.

Model rankings can reverse within months.

Information surfaces and distribution networks cannot be built that quickly.

Google’s Real Weakness May Not Be Its Model Either

Reverse the reasoning, and Google’s greatest risk becomes visible too.

Its greatest danger is not a rival model scoring five points higher on a benchmark.

It is people beginning to express their intentions without going through Google.

Previously, curiosity led to search.

In the future, people may simply tell an AI agent:

“Arrange everything for my business trip to Japan next week.”

Imagine that agent finding information, comparing flights, booking hotels, making an itinerary, and paying on its own.

The user need not see a Google Search results page.

If Search was the gateway between humans and the internet,

agents could become the new gateway between humans and the digital world.

That threatens one of Google’s most important assets: its intent surface.

The real war, then, is not whether Gemini, Claude, or GPT answers more exam questions correctly.

Who captures the moment human intention appears?

And

who connects it all the way to the final moment when intention becomes action?

That is the more important war.

Something More Formidable Than a Greatsword

Return to the swords at the beginning.

A company may build an extraordinarily powerful greatsword.

One swing produces astonishing results.

That is remarkable technology in itself.

But across the whole world, sword size alone does not decide the contest.

Eyes that see what is where.

Routes that determine where you can move.

Supplies that let you keep fighting.

Distribution networks that place swords in countless locations.

And information that tells you when to use them.

Together, these create actual power.

Google is not a perfect company.

It faces substantial regulatory risk and the innovator’s dilemma of protecting existing businesses. Nor does it always lead the best-model race.

But seeing Google merely as “the company that makes Gemini” keeps producing strange conclusions.

Much of Google’s real power lies outside the model.

It sees the world through Search,

accesses user context through Gmail, Photos, and other services,

distributes through Android, Chrome, and Workspace,

computes through Cloud and TPUs,

and now places an intelligence called Gemini on top.

So my conclusion from looking at Google is simple.

The strongest AI does not necessarily win.

If a sufficiently strong AI sees the widest world, reaches the most places, and acts most frequently, the story changes.

A greatsword is frightening.

But a wielder who sees only one side of the battlefield has limits.

Conversely, even with a slightly smaller sword,

if you can see the whole battlefield,

go anywhere,

know where the opponent is,

and move again and again,

then from the perspective of global dominance, the comparison becomes difficult even to make.

Power in the AI era may not reside only in model parameters.

The surface area intelligence can reach.

I increasingly think that is the real core.