JACE’S QUESTION #22 — WHO OWNS COMPUTE?
When Compute Becomes Capital, Who Owns the Future?
We usually talk about artificial intelligence as if intelligence itself were the scarce resource.
It isn’t.
At least, not anymore.
Increasingly, the scarce resource may be the infrastructure required to produce intelligence:
Compute.
GPUs.
Energy.
Data centers.
Networks.
Memory.
Models.
Capital.
As AI becomes embedded into finance, healthcare, insurance, government, commerce, and everyday life, compute stops being merely a technical resource.
It becomes economic infrastructure.
And once infrastructure becomes strategic, a different question appears:
Who controls it?
This question is larger than AI.
Because whoever controls the infrastructure that produces intelligence may eventually influence who can build it, who can access it, and who gets to participate in the next technological era.
1. COMPUTE IS NO LONGER JUST A TECHNICAL RESOURCE
Compute transforms energy and hardware into increasingly sophisticated forms of information processing.
The chain is roughly:
Energy → Hardware → Computation → Models → Applications → Economic Value
At the beginning of the chain are physical resources.
At the end are decisions, products, services, and economic activity.
This creates an important transition.
Compute is no longer simply something researchers use.
It is becoming something economies depend upon.
JACE’S QUESTION #22–1 — WHO OWNS COMPUTE?
When compute becomes capital, control over infrastructure can become control over capability.
2. WHY IS EVERYONE RACING FOR COMPUTE?
Every major technological era eventually identifies a new general-purpose bottleneck.
Steam engines transformed mechanical power.
Electricity transformed industrial production.
The Internet transformed information distribution.
Cloud computing transformed access to computation.
AI is increasingly transforming computation itself into a strategic resource.
The infrastructure underneath this transition is enormous:
- Semiconductor fabrication
- GPUs and accelerators
- Data centers
- Energy generation
- Cooling systems
- High-speed networking
- Memory infrastructure
- Software ecosystems
- Cloud platforms
The important point is that these layers are interconnected.
Owning one component does not necessarily mean controlling the entire system.
But concentration across multiple layers can create something more powerful:
structural advantage.
3. WHO OWNS COMPUTE TODAY?
The answer is not simply “AI companies.”
Compute is distributed across several institutional layers.
Big Technology Platforms
They operate massive infrastructure, build proprietary systems, and possess distribution networks.
AI Companies and Research Labs
They develop frontier models and compete for increasingly expensive computational resources.
Nations and Governments
They increasingly treat advanced compute as strategic infrastructure.
Semiconductor and Hardware Ecosystems
They determine what computational capabilities can physically exist at scale.
This produces an unusual economic structure.
AI companies need compute.
Compute providers need capital.
Capital increasingly follows AI capability.
AI capability creates new applications.
Applications create new demand for compute.
The system therefore becomes self-reinforcing.
4. THE HIDDEN BATTLE
The visible competition is usually:
Which model is smarter?
But underneath that question is another competition:
Who has access to enough compute to build the next model?
This distinction matters.
A researcher may have an excellent idea but insufficient compute.
A company may have enormous compute but mediocre ideas.
A government may possess infrastructure but lack frontier talent.
A platform may possess models, users, data, distribution, and capital simultaneously.
Therefore capability is not determined by intelligence alone.
It emerges from the interaction between:
Ideas × Compute × Capital × Data × Infrastructure × Distribution
This is where compute becomes more than hardware.
It becomes institutional power.
5. IF COMPUTE BECOMES ABUNDANT…
An interesting inversion occurs.
Today, much of the discussion assumes that compute scarcity is the primary bottleneck.
But what happens if compute eventually becomes abundant?
The bottleneck moves.
It may shift toward:
High-quality data
Meaningful objectives
Interpretation
Trust
Attention
Human judgment
Compute may become easier to obtain.
Wisdom may not.
And this produces the next transition in the series.
Because compute doesn’t merely produce intelligence.
It increasingly produces systems that observe and model human behavior.
6. WHEN FINANCE ALREADY KNOWS YOU
The second image changes the scale of the question.
Financial systems have spent decades transforming human behavior into structured information.
Transactions.
Credit.
Assets.
Loans.
Insurance.
Employment.
Purchasing.
Risk.
Now AI can sit on top of this infrastructure and make those systems more adaptive.
The result is not necessarily a machine that simply predicts markets.
It is a machine that can increasingly predict participants.
JACE’S QUESTION #22–2 — WHEN FINANCE ALREADY KNOWS YOU
Convenience can become control when prediction and personalization operate on the same behavioral data.
7. CONVENIENCE IS THE SWEETEST TRAP
Modern systems increasingly promise to remove friction.
Instant approvals.
Personalized recommendations.
Fraud detection.
Automated support.
Predictive services.
Everything becomes faster.
Everything becomes easier.
And that is precisely why the architecture can become difficult to see.
The user experiences:
Convenience.
The system experiences:
Information.
Every interaction can potentially become another signal.
Every transaction can become another data point.
Every choice can become another behavioral feature.
The system doesn’t necessarily need to know who you are philosophically.
It only needs to estimate:
What are you likely to do next?
That is a fundamentally different kind of knowledge.
8. FROM DATA TO PREDICTION
A simplified architecture looks like this:
Your Data → AI Model → Risk Profile → Decision
The important transformation occurs in the middle.
Raw events become features.
Features become probabilities.
Probabilities become scores.
Scores become decisions.
And decisions can determine:
- access to credit
- insurance conditions
- pricing
- recommendations
- eligibility
- visibility
- opportunities
At that point, prediction is no longer merely descriptive.
It becomes operational.
The system predicts behavior and then uses that prediction to determine what happens next.
9. THE FINE PRINT PROBLEM
There is another asymmetry.
The user usually sees the interface.
The system sees the model.
The user sees:
Approved
or
Denied
The system may have evaluated hundreds of variables.
The user may never know which ones mattered.
This creates what could be called an interpretability asymmetry:
The system knows more about the user than the user knows about the system.
That asymmetry becomes particularly important when automated decisions affect real opportunities.
10. THEY ARE NOT ON YOUR SIDE
Banks are not your friends.
Insurers are not your friends.
Platforms are not your friends.
Algorithms are not your friends.
But they don’t necessarily need to be.
They are systems optimized around particular objectives.
Profit.
Risk reduction.
Retention.
Prediction.
Efficiency.
Fraud prevention.
The important question is therefore not whether a system is “good” or “bad.”
The more technical question is:
What objective function is the system optimizing?
Because the same data can produce very different decisions depending on the objective.
11. WHEN PREDICTION BECOMES INFRASTRUCTURE
Once predictive systems become deeply embedded into institutions, something subtle can happen.
Get Kim, Jace (Jeong Hyeon)’s stories in your inbox
Join Medium for free to get updates from this writer.
Prediction becomes part of the environment itself.
You are not merely being predicted.
You are making decisions inside a system that predicts you.
That creates a feedback loop:
Behavior → Data → Prediction → Decision → New Environment → Behavior
The prediction influences the environment.
The environment influences the behavior.
The behavior generates new data.
And the cycle continues.
This is more powerful than prediction alone.
It is predictive feedback.
12. THE CREDIT SCORE OF REALITY
The third image pushes this logic one step further.
Financial institutions are not the only systems that score people.
Modern society increasingly contains scoring mechanisms everywhere.
Credit scores.
Risk scores.
Reputation systems.
Recommendation systems.
Performance metrics.
Social metrics.
Trust scores.
Health and insurance classifications.
Engagement rankings.
Even visibility itself can become a measurable variable.
JACE’S QUESTION #22–3 — THE CREDIT SCORE OF REALITY
When everything becomes measurable, measurement can begin to shape reality itself.
13. EVERYTHING IS SCORING YOU
Different systems may appear unrelated.
Banks evaluate financial risk.
Governments evaluate eligibility.
Insurers evaluate health and risk.
Retailers evaluate purchasing behavior.
Platforms evaluate engagement.
Employers evaluate performance.
But structurally, they can share the same pattern:
Observe → Extract Features → Assign Score → Predict → Decide
The domain changes.
The architecture remains surprisingly similar.
And once enough systems use scoring, something deeper happens.
The score can become more important than the underlying reality.
14. THE INVISIBLE FORMULA
Imagine a system evaluating a person using:
- financial transactions
- payment history
- employment
- purchasing behavior
- location patterns
- social interactions
- online activity
- behavioral consistency
No single variable necessarily defines the person.
But together they form a statistical representation.
The individual becomes:
a vector of probabilities.
This is where an important philosophical transition occurs.
A person is not reduced to a number because the number is necessarily inaccurate.
The deeper issue is that the number can become actionable.
If the score determines what you can access, then the score becomes part of your reality.
15. THE SYSTEM DOESN’T NEED TO KNOW EVERYTHING
This may be the most important technical point.
A predictive system doesn’t require a complete model of you.
It only needs enough information to produce useful predictions.
That means compression is not necessarily a weakness.
It can be the mechanism.
Thousands of behavioral observations can be compressed into a relatively small set of latent features.
Then:
High-dimensional life → Low-dimensional representation → Prediction
This is computationally efficient.
But it creates a dangerous conceptual possibility:
The map becomes easier to manipulate than the territory is to understand.
16. THE PREDICTIVE LOOP
Consider the following sequence:
1. Observe behavior
↓
2. Extract features
↓
3. Estimate probability
↓
4. Assign score
↓
5. Make decision
↓
6. Change the person’s environment
↓
7. Observe the resulting behavior
↓
8. Update the model
This is no longer simple classification.
It is a closed-loop socio-technical system.
The model does not merely observe reality.
Its decisions can alter the conditions under which future data are generated.
That distinction matters enormously.
17. THE REAL QUESTION
So we return to the beginning.
Who owns compute?
At first, it sounds like a hardware question.
Who owns the GPUs?
Who owns the data centers?
Who controls the energy?
Who operates the cloud?
Who builds the models?
But the question expands.
If compute produces intelligence…
and intelligence produces prediction…
and prediction produces scores…
and scores influence decisions…
then ownership of compute can eventually become connected to something much larger:
Who has the capacity to model reality at scale?
That is no longer merely an engineering question.
It is an infrastructure question.
An economic question.
A governance question.
And eventually:
a question about human autonomy.
18. COMPUTE → PREDICTION → CONTROL
The three images can therefore be read as one continuous chain.
22–1
Compute becomes capital.
↓
22–2
Capitalized compute becomes predictive infrastructure.
↓
22–3
Predictive infrastructure becomes decision infrastructure.
And the final transition is subtle.
Prediction does not automatically equal control.
But when predictions determine access, pricing, visibility, opportunity, or exclusion…
prediction becomes structurally consequential.
The difference between prediction and control begins to depend on what happens after the prediction.
19. WHAT SHOULD WE WATCH?
The future debate around AI should therefore not focus exclusively on model intelligence.
We should also examine the architecture surrounding it.
Compute concentration
Who can afford frontier-scale computation?
Data concentration
Who possesses the behavioral data?
Model concentration
Who builds the predictive systems?
Decision concentration
Who is allowed to act on those predictions?
Feedback concentration
Who controls the environment that generates the next round of data?
These five layers together may matter more than any individual model benchmark.
THE QUESTION BEHIND THE QUESTION
AI is often described as a race to build more intelligent machines.
But perhaps there are several races happening simultaneously.
A race for compute.
A race for data.
A race for prediction.
A race for infrastructure.
And eventually:
a race to define who gets to make decisions from all of it.
The most important resource may not be intelligence itself.
It may be the capacity to operationalize intelligence.
FINAL THOUGHT
“Compute is not just a machine.
It is the infrastructure that turns energy into intelligence — and intelligence into power.”
We began with a simple question:
Who owns compute?
But perhaps the deeper question is:
Who owns the systems that compute us?
Because once machines become capable of predicting our behavior, our financial lives, our preferences, our risks, and our futures…
the boundary between being measured and being shaped becomes increasingly difficult to see.
JACE’S NOTE
This is not an argument against computation, finance, AI, or predictive systems.
Those systems can create enormous value.
The question is about structural visibility.
We should know:
Who owns the infrastructure.
Who owns the data.
Who defines the objectives.
Who creates the scores.
Who makes decisions from them.
And who can challenge the result.
Because the most powerful system is not necessarily the one that predicts the future.
It may be the one that can change the conditions under which the future is produced.
THE REAL QUESTION
If compute becomes the foundation of intelligence…
who owns the future when they own the compute?
SCALING BUILDS CAPABILITY.
INFRASTRUCTURE DETERMINES ACCESS.
PREDICTION SHAPES DECISIONS.
AND DECISIONS SHAPE REALITY.
