AI Wants to Cure Cancer — But It Has a Surprisingly Basic Problem: Not Enough Data

 

Artificial intelligence has already changed software.

It can write code.

Generate images.

Translate languages.

Summarize research.

And increasingly, it is being asked to do something much harder:

help discover new medicines.

The promise is enormous.

If AI can understand how cells behave, predict how diseases develop, and simulate how drugs might work before expensive clinical trials begin, pharmaceutical research could become dramatically faster and cheaper.

Some researchers even hope AI could eventually help design treatments for diseases such as cancer, rare genetic disorders and neurodegenerative conditions.

But there is a problem.

Not enough biological data.

That sounds almost too basic.

AI companies have access to enormous computing power.

They have advanced models.

They have billions of dollars.

What they do not yet have is the biological equivalent of the internet.

And without that, AI cannot learn biology the way language models learned language.

Why ChatGPT Had an Advantage Biology Does Not

Large language models benefited from something extraordinary:

the internet.

Books.

Articles.

Websites.

Code.

Forums.

Reference material.

Billions upon billions of words.

The raw material already existed.

AI companies did not need to create language from scratch.

They scraped, licensed and collected enormous text datasets and trained models on them.

Biology is completely different.

There is no “internet of cells.”

There is no massive public archive containing every possible cell type responding to every possible drug, mutation, disease and environmental condition.

Much of that data simply does not exist yet.

That means AI researchers cannot just download biology.

They have to generate it experimentally.

And experiments are slow.

This Is Why Biohub Is Spending $1.8 Billion

A new effort backed by the U.S. government, Google, Meta and the nonprofit Biohub is trying to address exactly this problem.

The initiative plans to invest around $1.8 billion in creating large open biological datasets that can be used to train AI models. Reuters

The ambition is significant.

Researchers want to generate enough experimental data to build predictive models of how cells behave under different conditions.

In other words:

not just observing biology,

but eventually simulating it.

That is the idea behind what researchers often describe as a virtual cell.

What Is a “Virtual Cell”?

A virtual cell does not mean a cartoon cell on a computer screen.

The goal is much more ambitious.

Imagine a model that can predict:

how a cell reacts to a drug,

how a mutation changes its behavior,

how a disease alters a pathway,

or how two treatments might interact

before researchers run every physical experiment.

That could save enormous amounts of time.

Drug discovery currently depends on a huge number of laboratory experiments.

Most drug candidates fail.

Many fail late.

Late failure is extremely expensive.

A powerful biological model could help researchers eliminate weak candidates much earlier.

The Goal Is Prediction, Not Just Description

Traditional biological databases mostly tell us what researchers already observed.

A virtual biology model would aim to predict what happens next.

That is the key difference.

Instead of asking:

“What happened in this experiment?”

researchers could ask:

“What would happen if we changed this gene?”

or:

“What if we exposed this cell to this compound?”

That would make biology more computational.

And potentially much faster.

But AI Needs Enough Examples First

Here is the problem.

Cells are extraordinarily complicated.

A language model might learn that certain words tend to follow others.

A cell has:

genes,

proteins,

metabolic pathways,

signaling networks,

organelles,

epigenetic changes,

and environmental influences

all interacting at once.

Even two cells of the same type can behave differently.

That makes biological prediction much harder than text prediction.

AI needs huge numbers of carefully measured experiments before it can reliably generalize.

That is why data generation itself is becoming the bottleneck.

Biology Data Is Expensive

A text dataset can contain billions of examples gathered from existing documents.

Biological data often requires:

laboratories,

cell cultures,

reagents,

sequencing,

microscopy,

specialized equipment,

trained scientists,

and quality control.

Every new data point costs real money.

That changes the economics.

In AI software, compute is expensive.

In AI biology, both compute and experiments are expensive.

This Could Create a New Type of AI Lab

The next generation of AI research labs may look very different from the stereotypical room full of programmers.

They may contain:

robots,

pipetting systems,

microscopes,

sequencers,

automated incubators,

and AI software.

The idea is simple.

AI proposes an experiment.

Robots run it.

Sensors collect the results.

The model learns.

Then the AI proposes the next experiment.

That creates a feedback loop between software and the physical world.

Danaher Is Already Moving in This Direction

Life-sciences company Danaher has announced plans for an AI-powered autonomous research lab expected to begin operating in 2027. Reuters

The broader idea is important.

If experiments can be automated, AI does not have to wait for researchers to manually generate every new dataset.

It can create new data continuously.

That could dramatically accelerate biological learning.

Anthropic Is Building a Wet Lab Too

Anthropic, best known for Claude, has also established a biological wet lab in the San Francisco Bay Area as it expands into life sciences. Reuters

That is notable.

An AI company building a physical lab tells you how different biology is from ordinary software.

You cannot solve everything with more GPUs.

At some point, you need real cells.

Real molecules.

Real experiments.

AI biology is not purely digital.

That May Be the Most Important Shift

For years, AI companies competed through:

better models,

more parameters,

more compute,

and more data scraped from the internet.

In biology, the competitive advantage may become:

Who can generate the best proprietary experimental data?

That is a completely different race.

The strongest AI company may not be the one with the biggest model.

It may be the one with the best lab network.

Data Quality Matters More Than Data Volume

There is another problem.

More biological data does not automatically mean better AI.

Poorly controlled experiments can produce misleading signals.

Different labs may use different methods.

Samples can vary.

Measurements can be noisy.

Metadata can be incomplete.

If an AI model learns from inconsistent experiments, its predictions may be unreliable.

That means biology needs not just big data.

It needs standardized, high-quality data.

That is much harder.

The AI Industry Learned This Lesson in Language Too

Early AI development often followed a simple formula:

more data + more compute = better models.

Eventually, researchers realized the quality of the data mattered enormously.

Biology starts with an even more difficult version of that problem.

Bad biological data can have real consequences.

A wrong prediction about a drug is not merely an incorrect sentence.

It can waste years of research.

This Is Why Open Datasets Matter

The Biohub effort is especially interesting because the datasets are intended to become publicly available after an initial access period. Reuters

That could have a major effect on the industry.

If large biological datasets remain locked inside pharmaceutical companies, only a small number of organizations can build powerful models.

Open datasets allow:

universities,

startups,

research institutes,

and smaller biotech companies

to participate.

That could accelerate innovation.

But There Is Still an Incentive Problem

If a company spends hundreds of millions of dollars generating valuable biological data, why should it give that data away?

That tension will become increasingly important.

The AI industry grew quickly partly because enormous amounts of internet data were effectively available at low cost.

Biology is different.

Experimental datasets are expensive intellectual assets.

Companies may become reluctant to share them.

That could slow progress.

Biological Data Could Become More Valuable Than Models

This is worth thinking about.

AI models are increasingly reproducible.

A competitor may eventually build something similar.

But a proprietary dataset created through years of experiments is much harder to copy.

That means biotech companies could start treating biological datasets the way technology companies treat:

user data,

search indexes,

or mapping databases.

The data itself becomes the moat.

Pharma Already Has Huge Amounts of Hidden Data

Large pharmaceutical companies have decades of:

clinical trials,

failed drug programs,

laboratory experiments,

toxicity studies,

and patient data.

Much of that information has never been organized for modern AI.

That creates an enormous opportunity.

Some of the most valuable AI training material in healthcare may already exist.

It is simply fragmented across companies and formats.

Failed Experiments May Be Surprisingly Valuable

Science tends to publish success.

AI may benefit enormously from failure.

Imagine a drug candidate that failed.

Why did it fail?

Wrong target?

Poor absorption?

Unexpected toxicity?

No clinical benefit?

That information helps models understand what not to do.

Negative data is extremely valuable for prediction.

But historically, failed experiments have often been poorly shared.

That could become a major limitation.

AI Needs the Experiments Nobody Wanted to Publish

This is one of the strangest changes AI could bring to science.

A failed experiment may not make an exciting academic paper.

But for machine learning, thousands of failed experiments may be incredibly useful.

Models learn boundaries.

What works.

What does not.

Where predictions fail.

That could increase the value of previously ignored scientific data.

Virtual Drug Trials Are Already Happening

AI companies are already using simulated patients and predictive models to estimate how drugs might perform before or alongside traditional clinical trials. Reuters

The goal is not to replace human trials immediately.

It is to make them smarter.

For example, AI may help identify:

which patients are most likely to respond,

which dose is appropriate,

which side effects may appear,

or whether a drug is likely to fail entirely.

That could reduce unnecessary trials.

One Bad Phase 3 Trial Can Cost Enormous Amounts

Drug development is brutally expensive.

A company can spend years moving a drug through:

laboratory studies,

animal testing,

Phase 1,

Phase 2,

and finally Phase 3.

Then the drug fails.

Billions of dollars in potential value can disappear almost overnight.

If AI can identify even a fraction of those failures earlier, the economic value is enormous.

This Is Why Pharma Is So Interested

AI does not need to “cure cancer” to create value.

Even modest improvements could matter.

Suppose AI increases the probability of selecting a successful drug candidate by only a few percentage points.

Across hundreds of programs, that could save billions.

That is a much more realistic near-term business case than the idea of AI instantly discovering miracle cures.

AI Drug Discovery Has Been Overhyped Before

This is important.

The industry has already experienced waves of hype.

Companies promised that machine learning would:

dramatically shorten drug discovery,

find hidden targets,

and reduce clinical failure.

Progress has been real.

But not magical.

Biology remains difficult.

Many AI-designed drugs still need the same:

laboratory validation,

clinical testing,

regulatory review,

and manufacturing work

as any other medicine.

AI does not remove biology.

Models Can Be Confident and Wrong

This is especially dangerous in medicine.

An AI model can make a prediction with impressive confidence.

That does not mean the prediction is correct.

Biology contains hidden variables.

Models can miss important interactions.

Data may not represent certain patients.

Training datasets may contain biases.

That is why experimental validation remains essential.

The “Virtual Cell” Is Still a Research Goal

It is important not to oversell this.

A fully functioning general-purpose virtual cell that accurately predicts complex biological behavior does not yet exist.

The Biohub initiative aims to make major progress toward this kind of predictive biology over the next several years. Reuters

That is very different from saying scientists can already simulate a human cell perfectly.

They cannot.

Five Years Is an Aggressive Timeline

The Biohub initiative aims to compress what might otherwise take decades of biological data generation into roughly five years, with early datasets expected much sooner. Reuters

That is extraordinarily ambitious.

Whether the effort achieves that timeline remains to be seen.

But the investment itself shows how seriously researchers view the data problem.

This May Be Biology’s “ImageNet Moment”

In computer vision, large labeled datasets helped accelerate dramatic improvements in AI.

Researchers suddenly had common benchmarks.

Models could be compared.

Progress became measurable.

Biology may need something similar.

Huge standardized datasets could allow researchers around the world to train and test models on the same foundation.

That could accelerate progress dramatically.

Except Biology Is Much Harder Than Images

An image of a cat is still a cat.

A biological measurement depends heavily on context.

Cell type.

Age.

Disease state.

Environment.

Measurement technology.

Drug concentration.

Time.

Genetics.

That creates a massive number of possible combinations.

The biological state space is enormous.

No dataset will contain everything.

AI Will Need to Learn Which Experiments Matter Most

This is where AI could help generate its own training data.

Instead of randomly running millions of experiments, a model could identify the experiments that would reduce uncertainty the most.

This is sometimes called active learning.

The system essentially asks:

“What experiment should I run next to learn the most?”

That could make automated labs much more efficient.

The AI Scientist Loop

The ideal system might look like this:

AI develops a hypothesis.

↓

AI selects an experiment.

↓

Robotic lab runs the experiment.

↓

Sensors collect the result.

↓

Model updates.

↓

AI designs the next experiment.

That loop could run continuously.

If it works reliably, scientific discovery could speed up dramatically.

This Is Where Robotics Meets AI

Most people think of AI in biology as software analyzing medical data.

The next stage may be far more physical.

Laboratory robots are essential because AI needs fresh experimental feedback.

That means the AI biology boom could benefit companies involved in:

automation,

robotics,

sequencing,

microscopy,

diagnostics,

and laboratory equipment.

The economic opportunity extends far beyond software.

AI Could Change the Economics of Rare Diseases

Rare diseases are difficult to research because each one affects relatively few patients.

Traditional drug development may not always be economically attractive.

AI could reduce the cost of identifying:

targets,

drug candidates,

and patient groups.

If research becomes cheaper, more rare diseases may become viable targets.

That would be one of the most valuable social outcomes of AI drug discovery.

But Rare Diseases Also Have Less Data

There is a paradox.

The diseases that could benefit most from AI often have the least data.

Small patient populations produce limited datasets.

That makes prediction harder.

Synthetic data may help.

Transfer learning may help.

But the problem does not disappear.

Cancer Is Not One Disease

This is another reason “AI will cure cancer” is misleading.

Cancer is not one simple condition.

There are many cancer types.

And within each type, tumors can differ significantly based on:

genetics,

mutations,

microenvironment,

immune response,

and patient characteristics.

A treatment that works for one subgroup may fail in another.

That complexity makes AI useful.

It also makes the problem enormously difficult.

The Real Opportunity May Be Personalization

AI may not discover one universal cancer cure.

It may help identify:

which treatment,

for which patient,

at which time,

in which combination

has the highest probability of success.

That is a much more realistic vision.

Medicine becomes more individualized.

AI Could Help Match Patients to Treatments

Today, oncologists already use:

genomic testing,

biomarkers,

imaging,

and clinical history

to guide treatment.

AI could integrate far more variables simultaneously.

Potentially:

genomics,

proteomics,

pathology images,

medical records,

and drug-response data.

That could improve treatment selection.

Again, the limiting factor is data.

Healthcare Data Is Everywhere — But Hard to Use

This sounds contradictory.

Hospitals generate enormous amounts of data.

So why is biology data scarce?

Because healthcare data is messy.

Electronic medical records contain:

missing fields,

different coding systems,

unstructured notes,

inconsistent measurements,

and privacy restrictions.

Having lots of data is not the same as having useful machine-learning data.

Privacy Makes Medical Data Different

An AI model trained on internet text can use public information.

Patient data is different.

Medical records are highly sensitive.

Researchers must consider:

consent,

anonymization,

security,

regulation,

and data ownership.

That makes assembling large datasets much harder.

This Is Why Cell Data May Be So Valuable

One advantage of large-scale cell experiments is that researchers can generate standardized data without relying entirely on patient records.

Cells can be studied systematically.

Conditions can be controlled.

Experiments can be repeated.

That makes the data more useful for AI.

It also avoids some privacy issues.

But Cell Models Are Not Humans

This is another limitation.

A drug may work beautifully in isolated cells and fail in humans.

Human biology includes:

organs,

immune systems,

metabolism,

hormones,

microbiomes,

and countless interactions.

A virtual cell is not a virtual human.

That is why AI drug discovery still requires multiple layers of validation.

The Next Step Is Virtual Tissues and Organs

If cell prediction improves, researchers may eventually build increasingly complex models.

Cells.

Tissues.

Organs.

Eventually, perhaps partial models of entire patients.

But each level adds enormous complexity.

This is a long-term progression.

Not something likely to happen overnight.

Digital Twins Could Be Part of This Future

A digital twin is a computational model designed to represent a real system.

In medicine, the idea could eventually mean building a model of an individual patient.

Doctors might test treatment strategies on the model before applying them to the person.

That is still an emerging concept.

But better biological data makes it more plausible.

The Biggest Breakthrough May Be Fewer Failed Experiments

This is less exciting than “AI cures cancer.”

But economically and scientifically, it may matter more in the near term.

Imagine reducing the number of dead-end experiments by 20%.

That saves:

time,

money,

lab capacity,

and patient exposure.

Scientific productivity rises even without a dramatic single breakthrough.

AI Could Make Scientists More Efficient Before It Replaces Any Science

This is another important distinction.

The likely near-term use case is not:

AI replaces the biologist.

It is:

AI helps the biologist decide what to test next.

That is a much more credible transformation.

Human Expertise Still Matters

A model may identify a statistical relationship.

A scientist has to ask:

Does this make biological sense?

Could the experiment be flawed?

Is there another explanation?

What should we test next?

AI can accelerate scientific reasoning.

It does not eliminate scientific judgment.

This Could Change How Labs Are Designed

Traditional labs often operate project by project.

Future AI-native labs may prioritize standardized data production.

The goal becomes:

generate massive, reusable datasets

instead of merely answering one narrow scientific question.

That is a major change in research culture.

Biology Could Start Looking More Like Engineering

Engineering relies heavily on models.

Simulate before building.

Test before manufacturing.

Optimize before deploying.

Biology historically depends more heavily on physical experiments because predictive models are limited.

AI could move biology closer to engineering.

Simulate first.

Then test the most promising options.

That would fundamentally change research.

Pharma Companies Could Become Data Companies

A pharmaceutical company traditionally competes through:

scientific expertise,

drug pipelines,

patents,

and manufacturing.

In an AI-driven world, another asset becomes critical:

proprietary biological data.

Companies with decades of high-quality experimental data could gain a huge advantage.

That may change how pharmaceutical companies value old research archives.

Tech Companies Want In

This explains why companies associated with software are increasingly interested in biology.

Google DeepMind.

Isomorphic Labs.

Meta.

Anthropic.

OpenAI-related research efforts.

They see a familiar pattern.

Large datasets + powerful models + compute can transform a field.

But biology forces them to go one step further.

They need to help create the dataset too.

That Is Why $1.8 Billion Makes Sense

The amount sounds enormous.

But compared with the global cost of drug development, it may be relatively small.

A single successful blockbuster drug can generate billions of dollars annually.

A single failed late-stage clinical program can destroy enormous value.

If large biological datasets improve drug discovery even modestly, the return could be huge.

This Is Not Just a Healthcare Story

It is also an AI infrastructure story.

The previous generation of AI infrastructure meant:

GPUs,

data centers,

electricity,

and networking.

AI biology adds another category:

laboratories.

That means the physical infrastructure of AI keeps expanding.

The AI economy is moving deeper into the real world.

First It Needed Chips

Then power.

Now experiments.

That progression is revealing.

AI keeps discovering new bottlenecks as it expands into more complex domains.

Language needed text.

Robotics needs physical interaction.

Biology needs experimental data.

Every field has its own substrate.

The Next AI Winners May Look Unexpected

The companies that benefit may not all be famous AI model developers.

They could include businesses involved in:

lab automation,

sequencing,

microscopy,

scientific instruments,

data infrastructure,

bioinformatics,

and drug development.

AI creates demand throughout the scientific supply chain.

But Regulation Will Matter

Drug development is highly regulated.

An AI model cannot simply declare a treatment safe.

Regulators need evidence.

Clinical trials remain necessary.

Reproducibility matters.

Validation matters.

The more AI influences medical decisions, the more scrutiny it will receive.

That is appropriate.

The FDA Will Not Approve a Black Box Just Because It Is Smart

Medicine demands explanations and evidence.

If an AI model predicts that a treatment will work, regulators and scientists still need to understand:

how reliable the prediction is,

which data it used,

where it fails,

and whether the result can be reproduced.

AI confidence is not clinical evidence.

The Most Powerful Models May Still Fail on New Biology

Models are strongest when new situations resemble their training data.

What happens when a completely new disease appears?

Or a novel mutation?

Or a rare population?

The model may struggle.

That means biological AI needs mechanisms for uncertainty.

Sometimes the smartest answer should be:

We do not know. Run an experiment.

That may be one of the most important capabilities.

AI Needs to Know When It Is Wrong

This is true in every field.

It is especially important in healthcare.

A model that is 95% accurate sounds impressive.

But the remaining 5% may involve real patients.

Reliable uncertainty estimation will matter enormously.

This Is Why Human Trials Are Not Going Away

Virtual trials can help.

AI prediction can help.

Automated labs can help.

But human biology ultimately has to be tested in humans.

Clinical trials remain the strongest evidence that a treatment is:

safe

and

effective.

The future is likely hybrid.

AI filters candidates.

Experiments validate them.

Human trials confirm them.

The Goal Should Be Better Trials, Not No Trials

That is the more realistic vision.

Fewer hopeless candidates enter expensive studies.

Patient selection improves.

Dose selection improves.

Trial design improves.

More successful drugs reach patients.

That is transformative enough.

No science-fiction claim required.

There Is Also a Scientific Discovery Opportunity

AI may find patterns humans would never notice.

Biological datasets can contain millions of interacting variables.

Humans cannot manually inspect every relationship.

Models can identify hidden structure.

That could reveal:

new drug targets,

unexpected pathways,

or disease subtypes.

Some of the most important discoveries may come from correlations researchers did not know to look for.

But Correlation Is Not Causation

AI can find patterns.

Science still needs to determine whether those patterns are causal.

That requires experiments.

Again, we return to the same point.

AI cannot escape the laboratory.

This Is Why Autonomous Labs Are So Important

They close the loop.

AI predicts.

Experiment tests.

Result updates model.

Repeat.

That feedback loop is far more powerful than training once on a static dataset.

If autonomous labs become reliable, AI biology could improve continuously.

The Real “GPT Moment” for Biology May Not Be One Model

It may be a system.

Model.

Dataset.

Robotic lab.

Sequencing.

Simulation.

Human scientists.

All connected.

That system could be much more important than any individual algorithm.

This Is Also Why Healthcare AI Will Probably Advance Unevenly

Some areas have better data.

Imaging is a good example.

Radiology already generates standardized digital images.

That is why AI performed well there relatively early.

Molecular biology is harder.

Some diseases have huge datasets.

Others have very little.

Progress will vary by domain.

The Biggest Near-Term Winners May Be Narrow Models

A universal model of biology is extraordinarily difficult.

A model trained for one specific task may be much more practical.

For example:

predicting drug toxicity,

identifying protein interactions,

classifying cancer pathology,

or forecasting treatment response.

Specialized models can create value long before a full virtual cell exists.

Biology May Follow the Same Pattern as AI Software

Early AI hype focused on artificial general intelligence.

Actual economic value often came from narrow tools:

recommendation,

translation,

fraud detection,

advertising,

coding.

Biology may follow a similar pattern.

The grand vision is virtual cells.

The money may first come from dozens of smaller predictive tools.

This Could Still Transform Drug Development

Imagine the drug-development pipeline with AI at every stage.

Target identification.

Molecule design.

Toxicity prediction.

Patient selection.

Trial simulation.

Manufacturing optimization.

Post-market monitoring.

No single step needs to become perfect.

Small improvements across the entire pipeline could dramatically improve productivity.

Why This Matters for Patients

Drug discovery is slow.

It is expensive.

And many diseases still lack effective treatments.

If AI can reduce wasted experiments and failed trials, new medicines could potentially reach patients faster.

That is the real reason this matters.

Not because AI suddenly becomes a doctor.

Because research becomes more efficient.

What Should We Watch Next?

There are five things worth following.

1. How quickly Biohub releases usable datasets

The project becomes meaningful only if researchers can actually train models on the data.

2. Whether virtual-cell predictions reproduce in real experiments

Accuracy matters more than impressive demos.

3. Whether autonomous labs can operate reliably

Robotics and experimentation are critical.

4. Whether AI-predicted drug candidates perform better in clinical trials

This is the ultimate commercial test.

5. Whether data remains open

If biological datasets become closed competitive assets, innovation could become concentrated among a few giant companies.

My Take: The Biggest AI Biology Breakthrough May Be Boring

It may not be:

“AI discovers the cure for cancer.”

It may be:

“We generated 100 times more standardized biological data than before.”

That sounds far less exciting.

But it may be what makes everything else possible.

Artificial intelligence has always depended on data.

Biology is simply forcing the AI industry to confront a difficult reality:

you cannot train on information that humanity has never measured.

The Real Bottleneck Is Reality

Language models learned from the digital world.

Biological AI has to learn from the physical world.

That makes progress slower.

But it may also make the eventual impact much larger.

Because once AI becomes good at predicting physical biology, it moves beyond information.

It starts affecting medicine itself.

The Bigger Lesson

People often talk about AI as if better algorithms automatically solve problems.

They do not.

The model is only one piece.

AI also needs:

data,

infrastructure,

experiments,

validation,

and human expertise.

In biology, the weakest link right now may simply be the data.

That is why governments and technology companies are investing billions to create it.

The Bottom Line

AI researchers increasingly believe that powerful biological models could eventually help predict cellular behavior, improve drug discovery and make clinical development more efficient.

But the field faces a basic bottleneck:

there is not enough standardized, high-quality experimental biology data.

A new Biohub-led initiative backed by the U.S. government, Meta, Google DeepMind and Isomorphic Labs plans roughly $1.8 billion in investment to generate large-scale biological datasets and develop predictive models of cells. Reuters

At the same time, companies including Danaher and Anthropic are building more automated or AI-connected laboratory infrastructure, while startups are already experimenting with virtual drug trials. Reuters

None of this means AI is about to cure cancer next year.

The real shift is more fundamental.

The AI industry is learning that biology cannot be solved by simply scaling models.

First, it has to scale science itself.

And the next great AI race may not be about who builds the smartest model.

It may be about who can generate the best data from the living world.

Related Reading

AI Is Running Out of Power — And the Next Chip Shortage May Not Be About Chips at All

The $3 Trillion Question: What’s Actually Being Built for the AI Boom

China Is Quietly Building Its Own Nvidia — And Nvidia’s Moat Is Starting to Crack

The Person Looking at You Might Be Recording You — AI Glasses Are Becoming a Privacy Nightmare