Elon Musk, Jensen Huang, and Anthropic co-founder closed-door roundtable: Superintelligence (SI), orbital computing power, and White House security declaration
Elon Musk
Founder & CEO, Tesla & SpaceX
Original Statement
1. Core Concept Upgrade: Transitioning from AI to "Super Intelligence (SI)" and Restructuring the Computing Stack
• From Retrieval Computing to Generative Super Intelligence (Generative SI):
• Jensen Huang pointed out that "Super Intelligence (SI)" represents a complete rewrite of the entire computing architecture technology stack—from the design of underlying processor chip architecture, to model pre-training methods, and to the execution logic of end applications.
• The essence of past computing was to "retrieve existing information" from existing databases, while current computing is purely driven by energy to "instantaneously generate cognition and value."
• Energy is the absolute physical foundation of all computing power:
• The computing stack unfolds from the bottom layer in sequence: energy generation and delivery $\rightarrow$ land and data center shell $\rightarrow$ computing power infrastructure $\rightarrow$ fundamental algorithms $\rightarrow$ terminal industry applications. Without sufficient and cheap electricity supply, even the most advanced chips and models cannot operate at scale.
2. Hardcore Calculation of Computing Power and Economics: Musk's "1% Power = 1% GDP" Law
• National electricity consumption baseline and marginal pull:
• The current normalized average electricity consumption in the U.S. is about 500 GW (gigawatts).
• Musk proposed a core inference: every increase of 5 GW in sustained steady-state electricity supply corresponds to a 1% increase in total electricity consumption across the U.S., which, under the premise of continuous improvement in hardware architecture and large model efficiency, will directly translate to about a 1% increase in U.S. GDP.
• Quantifying economic output: If 1% GDP corresponds to about $300 billion in value, it means that every 1 GW of computing power electricity can directly drive $50-60 billion in actual economic output annually.
• Global geopolitical electricity production capacity competition pressure:
• Musk bluntly stated that although the U.S. maintains absolute leadership in software, large model algorithms, and digitalization, in terms of long-term physical energy dimensions, China's total power generation has reached about three times that of the U.S.
• The key to ensuring long-term strategic advantage lies in the unprecedented speed of expanding electricity installations across the U.S., while ensuring the supply capacity of advanced process logic chips and high-bandwidth storage chips in domestic or absolutely secure allied regions.
3. Orbital Compute and the "Starship" Space Revolution
• The geometric efficiency gap between ground solar energy and space-based solar energy:
• Ground-deployed solar energy is affected by day-night alternation, weather fluctuations, and atmospheric refraction, with actual average power generation efficiency only being 1/8 to 1/5 of the rated nameplate capacity, and requiring extremely expensive and heavy energy storage battery matrices.
• The space orbital environment is in an "always sunny" condition, allowing solar panels to generate power at 100% rated peak load year-round without the need for large-scale energy storage transitions.
• Starship fully supports the deployment of orbital computing:
• With Starship successfully completing its first orbital mission and successfully launching the Starlink V3 satellite, which has a wingspan comparable to a Boeing 737 (the largest single payload to reach orbit since Skylab), the launch frequency of Starship is expected to accelerate to once a week or even twice a week.
• Target scale: SpaceX, in collaboration with Tesla, plans to deploy a solar power array and AI computing cluster of 200 GW in space annually (equivalent to 40% of the current total electricity consumption in the U.S.), and eventually aim for a 1 TW (terawatt) level of orbital computing, directly transporting cutting-edge computing chips like those from NVIDIA into orbit.
4. "Super Intelligence Factories" Reshaping the U.S. Real Economy and Reindustrialization
• Fundamental evolution of data center nature:
• The core function of traditional "data centers" was merely passive storage and archiving of data.
• Today's clusters are referred to as "Super Intelligence Factories"—akin to heavy industrial production workshops driven by generators, integrating ultra-high voltage power conversion, high-density liquid cooling pipelines, ultra-low latency network topologies, and thousands of high-end GPUs, continuously outputting high-value intelligence to society through input power and data.
• The first reindustrialization and blue-collar employment boom in 50 years:
• The U.S. is expected to advance the construction of super intelligence factories by 10-20 GW annually, directly driving the construction of power plants, grid upgrades, steel structure construction, heavy pipeline laying, and other cross-industry supply chains, creating over 1 million high-quality blue-collar and skilled jobs.
• This aims to reverse the severely distorted structure of the U.S. economy that has leaned towards pure white-collar service industries over the past few decades, promoting the return of physical construction and engineering craftsmen to the core of society.
• Community win-win practices (taking the Colossus computing base as an example):
• Giant data centers often settle in small towns in the Midwest or South, facing public resistance due to high energy consumption;
• xAI has invested $250 million to build an industrial-grade wastewater recycling plant on-site, provided local residents with half-price Starlink access, and doubled the local tax revenue budget, reversing the local unemployment crisis and creating a model of collaborative prosperity with local communities.
5. NVIDIA Agent Safe Open Source Architecture: Open Shell and Bluefield Physical Monitoring
• Refuting the false proposition that "technological progress must sacrifice safety":
• Jensen Huang emphasized that the early internet solved the credit card encryption theft problem, which gave rise to trillion-dollar digital businesses like Google and Amazon; advanced technology inherently includes higher-dimensional security, and the two are complementary sides of the same coin.
• Two core mechanisms for Agent operational safety: isolation and hardware external monitoring:
• Open Shell (Agent Isolation Sandbox): Like a browser designed specifically for Agent operation isolation, it delineates non-crossable boundaries for file systems, network ports, and operational permissions for agents, preventing unauthorized contamination.
• Bluefield chip independent real-time out-of-band monitoring: Never trust the Agent's self-examination; instead, a physically independent Bluefield dedicated chip captures, measures, and intercepts all boundary-crossing calls in real-time outside the sandbox environment. Once an abnormal action violating established policies is detected, it is immediately blocked at the hardware level and reported.
6. Cutting-edge Model Iteration and the White House Super Intelligence Security Declaration
• Anthropic Opus 5.5: Autonomous delegation of long-horizon tasks and explosive growth of software assets:
• Co-founder Tom Brown explained the core breakthrough of the latest cutting-edge model—supporting the autonomous delegation of complex tasks over longer time spans.
• All digital tools in human history were previously handwritten line by line by human programmers; now we have entered a new era of "human and super intelligence collaborative programming," where industrial software and consumer tools are being reconstructed at dramatically low costs and exponential iteration speeds.
• Profound reflection on tool values and ethical neutrality:
• Jensen Huang raised a warning: SI should first be defined as a precision tool; excessive a priori moral judgment may undermine its core effectiveness as a tool (analogous to a steak knife refusing to cut due to insufficient meat doneness is absurd; in cybersecurity defense, the defensive countermeasures often resemble hacker attacks in form, and it must be ensured that tools do not refuse to respond due to rigid preset judgments).
• Consensus from the White House closed-door meeting: Signing the "Super Intelligence Security Joint Declaration":
• Tech leaders and the White House formally reached and signed a nationally binding security cooperation framework;
• Three core requirements:
• Industry cross-review (Grading each other's homework): Abandoning self-endorsement behind closed doors, establishing a red-blue confrontation and joint evaluation mechanism across giants;
• Establishing an independent security special committee at the board level to implement mandatory internal controls and traceable audits on key internal R&D nodes;
• Introducing authorized authoritative third parties to conduct periodic external audits and establishing an immediate sharing mechanism for major systemic threats and engineering best practices in the industry.
ABAB AI Insight
This material is valuable, but if I were to follow your request for "highest cognition, real substance" standards, I wouldn't speak directly from it.
Because the most important thing here is not "Elon Musk talked about space again, Jensen Huang talked about chips again," but rather: the United States is officially upgrading AI from a software industry to a new industrial system driven by energy, industry, finance, national security, and space infrastructure.
Let me give you a conclusion first:
In the next phase of AI, the competition is no longer about whose large model is smarter, but about who can control the entire intelligent production chain of "energy → chips → data centers → models → agents → industrial applications."
From this perspective, what is happening today is more like the combination of the 19th-century railway, 20th-century electrification, and the internet revolution, rather than a simple "AI technology upgrade."
However, there are a few points in your original text that need correction. I will first separate the facts from the opinions and then provide you with the deeper logic that is truly worth discussing.
────────────────
1. First, let's correct a few key facts: this will directly determine the credibility of the article.
1. The White House has indeed established a very important "superintelligence" industry safety framework.
On September 29, 2026, Trump held a meeting with tech company leaders including Elon Musk, Jensen Huang, Dario Amodei, Mark Zuckerberg, and Sundar Pichai, and announced a set of voluntary safety commitments surrounding the so-called "Super Intelligence."
But it must be noted here:
It is currently not a legally binding regulatory framework.
Trump himself described it as having a "morally binding" nature, which is closer to:
Industry joint commitment + self-regulatory framework.
Mainly includes:
• Internal safety controls within companies;
• Independent third-party audits;
• Board-level supervision;
• Sharing best safety practices between companies.
So the original text states:
"A nationally binding security cooperation framework with substantial constraints"
This expression is a bit exaggerated.
A more rigorous statement should be:
Top AI companies in the U.S. have formed a politically binding but not yet legally enforceable joint safety framework at the White House level for the first time.
This distinction is very important. Reuters
────────────────
2. The real big change: AI is transitioning from a "software industry" to a "capital-intensive heavy industry."
This is the most important thing to understand today.
Many people still think of AI as:
ChatGPT → model → software.
This understanding is outdated.
The real AI industry chain now is:
Natural gas / nuclear power / photovoltaics → power grid → transformers → land → data centers → GPUs → HBM → networks → models → agents → software → robots.
This signifies a historic industrial reversal:
In the past internet era,
the core resource for startups was:
programmers.
You could even:
With 10 people + a few million dollars,
create a global software company.
In the AI era,
if you want to train the world's most advanced model:
you might need:
tens of billions, hundreds of billions, or even over a trillion dollars in capital.
Thus, AI is bringing the tech industry back to a very old economic structure:
capital-intensive industry.
This is increasingly similar to:
railways,
steel,
electricity,
oil,
telecommunications.
────────────────
3. Therefore, what Jensen Huang is really talking about is not GPUs, but rather "intelligence becoming an industrial product."
The term "AI Factory" is very important.
Many people translate it as:
AI factory.
But it actually expresses a new economics.
Traditional factories:
Steel + energy
↓
Cars
Refineries:
Crude oil + energy
↓
Gasoline
Power plants:
Coal / natural gas / nuclear fuel
↓
Electricity
AI Factory:
Electricity + data + chips
↓
Tokens / Intelligence
This actually means:
Intelligence is for the first time becoming a commodity that can be industrially mass-produced.
This is an extremely significant historical change.
In the past, human society's "intelligence" was mainly stored in:
human brains.
Thus, the productive capacity of society was subject to a natural limitation:
The number of high-skilled talents is limited.
Lawyers only have 24 hours a day.
Programmers only have 24 hours a day.
Doctors only have 24 hours a day.
Analysts also only have 24 hours a day.
But after the emergence of the AI Factory:
Intelligent production can expand by increasing:
GPUs,
electricity,
data centers.
In other words:
For the first time,
the supply of intelligence begins to have "capital expansion attributes."
In the past, if you wanted to increase the number of programmers by 1 million, it required decades of education systems.
In the future, theoretically, it only requires building more computing infrastructure.
This is one of the deepest changes AI brings to economic growth.
────────────────
4. The statement "energy equals AI" is much more important than chips.
Musk repeatedly emphasizes that the average electricity usage in the U.S. is about 500GW.
This scale is basically reasonable; he publicly cited similar figures at the 2026 Davos. World Economic Forum
What is truly worth understanding here is not:
500GW.
But rather:
AI has for the first time directly linked electricity and "intellectual production."
Traditional industry:
Electricity → factories → products.
AI:
Electricity → GPUs → tokens → intelligence.
So in the future, electricity may acquire a new economic attribute:
Every kilowatt-hour can be converted into a certain amount of intelligence.
You can understand this as:
Electricity is becoming the "raw material for digital labor."
This is why:
Meta,
Google,
Microsoft,
Amazon,
xAI,
OpenAI
have suddenly all started talking about:
nuclear power,
gas turbines,
grid,
storage,
data centers.
Because ultimately, the constraint on AI development may not be algorithms.
But rather:
Power.
────────────────
5. However, "1% electricity = 1% GDP" must not be treated as an economic law.
This is the point in your draft that requires the most caution.
Equating:
an increase of 1% in electricity
directly with:
an increase of 1% in GDP,
can be understood as a macro deduction or industrial vision from Musk, but it cannot be treated as a verified economic law.
The reason is simple:
GDP is not a linear function of electricity.
Increasing 5GW of electricity, if used for:
digging low-value assets,
and if used for:
operating the world's most advanced AI,
the resulting GDP would be completely different.
What truly determines the outcome is:
Electricity × capital efficiency × chip efficiency × software efficiency × AI commercialization efficiency.
So a more accurate formula should be:
AI economic output ≈ Electricity × Compute Efficiency × Model Efficiency × Utilization × Economic Value per Token.
Each of these multipliers is very important.
────────────────
6. This explains a very important phenomenon in the capital market: why all companies are now studying "power per watt."
One of the most critical indicators for AI in the future may not be:
the number of GPUs.
But rather:
Intelligence per Watt.
How much effective intelligence can be produced per watt.
Because assuming:
GPU efficiency doubles,
model efficiency doubles,
inference algorithm efficiency doubles,
with the same 1GW of electricity:
intelligence output could become:
8 times.
This is also why the AI industry in the future will not just be about "building more power plants."
More importantly:
improving the overall system efficiency.
Winners are likely to emerge in:
chips,
HBM,
liquid cooling,
networks,
power management,
optical communication,
model compression,
inference optimization,
data center scheduling,
these seemingly "unsexy" areas.
────────────────
7. This is actually the "Jevons Paradox": the more efficient AI becomes, the greater the energy demand may actually be.
This is a very worthwhile economic knowledge for readers to learn.
19th-century British economist William Stanley Jevons discovered that:
After the efficiency of steam engines improved,
people originally thought that coal consumption should decrease.
The result was exactly the opposite.
Because once steam power became cheaper:
the usage scenarios increased massively,
and ultimately, the total demand for coal actually rose.
This is called:
Jevons Paradox.
AI may very well experience the same thing.
Many people believe:
Improved chip efficiency → decreased AI power consumption.
In reality, it may be:
Decreased AI costs
↓
Explosive AI demand
↓
Explosive growth in the number of agents
↓
Explosive growth in the number of robots
↓
Total computing power actually increases.
So the more efficient AI chips become:
Electricity demand may not decrease, but rather may increase.
This is a very important logic for understanding energy investments in the next decade.
────────────────
8. Why the U.S. has suddenly re-entered the "industrialization era."
You captured a very important direction in your original text.
The U.S. has experienced:
deindustrialization + service-oriented + financialization over the past 40 years.
Many manufacturing processes have shifted to:
China,
Mexico,
Southeast Asia.
But the AI industry has a special point:
Data centers cannot be moved.
Power grids cannot be moved.
Nuclear power plants cannot be moved.
Natural gas pipelines cannot be moved.
Substations cannot be moved.
Construction cannot be moved.
Therefore, AI may bring about a very unique:
"non-offshorable industrial chain."
A 1GW-level AI campus requires:
power engineers,
electricians,
welders,
HVAC technicians,
plumbers,
construction workers,
server maintenance personnel,
transformers,
switching equipment,
cooling systems.
All of these must be done locally.
So AI may simultaneously produce two completely opposite labor force outcomes:
In the white-collar world,
AI replaces some:
programmers,
customer service,
analysts,
clerks,
junior lawyers.
In the physical world,
it may require more:
electricians,
welders,
mechanical engineers,
construction personnel,
energy engineers.
This may become one of the most important labor structure changes in the U.S. over the next decade.
────────────────
9. This is also why the economic value of "blue-collar skills" may rise again in the future.
The path for young people in the U.S. over the past 30 years has been:
University
↓
Finance / Consulting / Technology
↓
White-collar
The AI era may see a reversal:
Electricians,
HVAC,
plumbing,
data center engineering,
robot maintenance,
these professions may see their incomes continuously rise.
Because of a very realistic problem:
AI can quickly increase the supply of programmers,
but AI is difficult to immediately increase:
certified high-voltage electricians.
So many people's mistake in the future may be:
everyone rushing to learn AI.
What is truly scarce is:
the real-world skills that AI cannot replicate in the short term.
────────────────
10. The real terrifying competition is not Nvidia vs AMD, but "U.S. industrial capability vs Chinese industrial capability"
Musk has always emphasized the scale of China's electricity.
This cannot simply be summarized as "U.S. technology, Chinese manufacturing."
The real strategic competition is turning into:
U.S. advantages
AI models
GPU design
software
capital markets
cloud computing
entrepreneurial ecosystem
Chinese advantages
power generation capacity
grid construction
photovoltaics
batteries
power equipment
manufacturing
industrial supply chain
China's overall power system scale has significantly surpassed that of the U.S., which is a reality; the specific multiple should be distinguished by power generation, installed capacity, or stable usable power, and cannot simply be written as "3 times." IEA
This precisely indicates that:
the AI competition may ultimately become:
Silicon Valley's software efficiency × U.S. capital markets
against
China's manufacturing efficiency × energy infrastructure.
────────────────
11. So the real anxiety in the U.S. now is not about "whether there is the best model"
The U.S. already has:
OpenAI,
Anthropic,
Google,
Meta,
xAI.
The real issue is:
if all these companies expand simultaneously:
Where will the electricity come from?
Where will the transformers come from?
Where will the gas turbines come from?
Where will the HBM come from?
Where will the advanced processes come from?
Where will the cooling equipment come from?
A very typical case reported by Reuters today is:
Due to the long delivery cycle of large gas turbines, many data center developers are turning to smaller gas turbines that can be deployed faster, and it is expected that by 2030, the U.S. may add nearly 30GW of behind-the-meter gas generation, most of which is related to data center demand. Reuters
This indicates that:
AI has already begun to directly change the U.S. energy equipment industry.
────────────────
12. This is the biggest "gold rush era" of the future
The California Gold Rush of the 1850s.
In the end, those who truly make stable profits may not be the gold miners.
But possibly:
Levi's.
Railroads.
Banks.
Tool suppliers.
AI is the same.
Everyone is focusing on:
ChatGPT,
Claude,
Gemini.
But the real long-term beneficiaries of AI infrastructure may be:
power generation
gas turbines
nuclear power
grid equipment
transformers
data center construction
cooling
GPU
HBM
optical communication
power supply equipment
network equipment
data center REITs
────────────────
13. Nvidia's truly dangerous ambition: it is transforming from a GPU company into an "AI industrial standard setter"
This is much more important than OpenShell itself.
Nvidia's recently released Open Agent Safety Platform mainly includes:
OpenShell
Control:
What files the agent can access;
What APIs it can call;
What networks it can access;
What code it can execute.
Sentry
Runs on:
BlueField-4 DPU
as a software/hardware supervision layer independent of the agent.
If the agent attempts to breach permissions:
Sentry can isolate or block it. NVIDIA Developer
The real important business logic behind this is:
Nvidia is trying to control AI's:
computing layer
network layer
agent runtime
security layer
data center infrastructure layer
In other words:
it does not just want to sell:
GPUs.
It wants to become:
the Intel + Cisco + VMware + Oracle infrastructure layer of the AI era.
────────────────
14. The most important idea of OpenShell is actually: do not trust AI, but limit AI
This is actually a very mature security concept.
There is a core principle in traditional network security:
Zero Trust
It means:
Never trust, always verify.
Do not trust just because:
"This is an internal employee."
The same goes for the agent era.
In the future, companies will not ask:
"Will Claude do bad things?"
but should ask:
"Even if Claude goes out of control, what can it do at most?"
This is a completely different design philosophy.
For example:
a financial agent can:
read invoices;
organize payments.
But cannot:
directly transfer funds.
Unless:
human approval.
This is the way enterprise AI can truly be deployed at scale.
────────────────
15. So one of the biggest business opportunities for agents in the future may not be the agents themselves, but "agent permission management"
This is something many entrepreneurs tend to overlook.
In the future, a company may have:
100,000 human employees
but have:
1 million agents.
So the company must solve:
Who is this agent?
What can it access?
How much money can it spend?
Who can it email?
What databases can it modify?
What APIs can it call?
Who approves?
Who is responsible in case of an accident?
This will actually create a whole new:
Machine Identity Economy.
In the past, we managed:
Employee Identity.
In the future, we also need to manage:
Agent Identity.
This is a very large market for future network security.
────────────────
16. The truly important aspect of Anthropic Opus 5.5 is not the benchmark
Opus 5.5 is indeed set to launch on September 22, 2026.
Anthropic emphasizes:
stronger long-term agent task capabilities,
lower token costs,
complex coding capabilities. Anthropic
But what really changes the economy is:
"Delegation"
Previously with AI:
You ask a question.
It answers a question.
Now:
You say:
"Complete this project."
AI may:
research information
write code
test
find errors
modify
retest
submit results.
This means the AI economy is transitioning from:
tool economy
to:
digital labor economy.
────────────────
17. After the emergence of digital labor, the logic of company scale may change completely
In the past:
A company with $10 billion in revenue,
might need:
20,000 employees.
In the future, it may be:
$10 billion in revenue,
only needing:
2,000 human employees
• 200,000 agents.
This will create a very frightening financial change:
Revenue per Employee
may rise significantly.
In other words, the characteristics of the best companies in the future may be:
huge revenue,
very few human employees.
This can already be seen in many software companies.
And agents will amplify this trend.
────────────────
18. The truly crazy part comes: why Musk must move computing power to space
Many people's first reaction is:
Isn't this crazy?
In fact, it has a very clear physical logic.
Earth's AI data centers have four core limitations:
power
land
cooling
community resistance
Space theoretically provides:
solar energy
ample space
no local residents protesting
no traditional land costs.
SpaceX has clearly included orbital AI computing in its long-term strategy and plans to start testing from a smaller scale and gradually expand; Reuters has also reported its plans to validate orbital AI computing capabilities in the coming years. Reuters
────────────────
19. But "100% utilization of solar energy in space" should not be understood too simply
This part of the original text also needs correction.
Orbital solar energy indeed has:
no cloud cover,
no atmospheric obstruction,
higher solar radiation utilization rate,
and other advantages.
But not all orbits:
are in sunlight 24 hours a day.
It also depends on:
orbital height,
inclination,
earth obstruction,
orbital design.
More importantly:
Space data centers also have several huge engineering challenges:
heat dissipation
Space is a vacuum.
There is no air.
It cannot dissipate heat like Earth data centers:
using air to carry away heat.
It can only mainly rely on:
radiative cooling.
This may require a huge radiator area.
Radiation
High-energy particles may damage chips.
Maintenance
If an Earth server breaks down:
engineers replace it.
If a space server breaks down:
it is very troublesome.
Launch costs
must be low enough.
The Financial Times also pointed out that heat dissipation, radiation, maintenance, and launch economics remain core issues faced by orbital data centers. Financial Times
────────────────
20. So what really determines whether "Orbital Compute" can be established is not AI, but Starship
This is key.
If launch costs remain at traditional levels:
orbital data centers are almost meaningless.
But if Starship ultimately achieves:
high-frequency launches
complete reusability
100-ton payloads,
the economic model may change.
SpaceX's public documents have already linked future orbital AI computing power with Starship. SEC
So you can understand:
Starship is essentially not just a rocket
If this strategy succeeds:
it will become:
a truck connecting the Earth economy with the space industrial economy.
Just like:
railroads connected the American continent.
Container ships connected global trade.
Starship may connect:
the Earth and orbital industry.
This significance is more realistic than "going to Mars."
────────────────
21. But the "deployment of 200GW per year" should currently be seen as ambition, not a fait accompli
You must tone down this part of the original text.
SpaceX-related public documents have indeed proposed very aggressive orbital AI computing power goals, including long-term plans to reach deployment levels of tens or even 100GW per year. SEC
However:
this does not mean that such capabilities exist today.
From:
1GW
to:
100GW,
is not as simple as 100 times.
Because you simultaneously need:
100 times satellite production,
100 times launch frequency,
100 times solar panels,
100 times chips,
100 times cooling equipment,
100 times communication infrastructure.
So the correct understanding should be:
This is the industrial scale SpaceX is trying to establish, not the capacity that has already been realized.
────────────────
Twenty-two, here you can truly learn a "billionaire mindset": Don't look at the industry from the product perspective, but from the bottleneck perspective.
This is the most valuable lesson for entrepreneurs from this incident.
Ordinary people see:
AI is very popular.
So they think:
Let's make an AI App.
Experts will ask:
What is the biggest bottleneck in the entire industry?
Currently, it might be in order:
2023:
GPU.
2024:
HBM.
2025:
Data centers.
2026:
Electricity.
Next, it might be:
Transformers,
Gas turbines,
Nuclear power equipment,
Cooling,
Optical communication,
Land,
Grid connections.
Real big money often does not come from:
Chasing the hottest things.
But rather from:
Solving the most painful bottlenecks in the entire hot industry.
────────────────
Twenty-three, almost all super industrial revolutions in history follow the same pattern:
Railway era:
Railway prosperity
→ Steel prosperity
→ Banking prosperity
→ Land prosperity.
Automobile era:
Automobile prosperity
→ Oil
→ Roads
→ Suburban real estate
→ Fast food
→ Motels.
Internet era:
Internet
→ Fiber optics
→ Data centers
→ Cloud computing
→ E-commerce
→ Digital advertising.
AI will similarly emerge:
AI models
↓
GPU
↓
Data centers
↓
Energy
↓
Robots
↓
Industrial automation
↓
New business models.
So the places where true wealth creation occurs are often not at the first level.
But rather:
The second and third-order effects of the entire ecosystem.
────────────────
Twenty-four, from a higher perspective, this White House meeting actually revealed the U.S. national strategy for the next 10 years:
It is not simply:
"Develop AI."
But rather four things happening simultaneously:
First
Superintelligent industrialization.
AI is no longer a laboratory technology.
Second
Energy expansion.
The U.S. must rebuild its energy infrastructure on a large scale.
Third
Re-industrialization.
AI data centers drive local engineering industries.
Fourth
Safety institutionalization.
As agents begin to truly operate in the real world, safety will move from the model layer into:
Software layer,
Operating system layer,
Chip layer,
Board level,
Government level.
────────────────
Twenty-five, finally, here’s a framework that is truly worth taking away:
In the future, when studying AI, do not just ask:
"Which model is the best?"
Instead, ask seven questions:
First, where is the energy?
Second, where is the computing power?
Third, who manufactures the chips?
Fourth, who provides the capital?
Fifth, who controls the models?
Sixth, what real powers can agents execute?
Seventh, where does the final economic value flow?
Clarifying these seven layers:
What you see will no longer be:
What new features ChatGPT has updated.
But rather:
A new global industrial system is forming.
This is the real point of attention in this "Superintelligence Conference."
And I believe the most important sentence this time is not even "superintelligence."
But rather:
The intelligence of the future may essentially be an industrial product produced in bulk by electricity, chips, and capital.
Once you accept this logic, AI, nuclear power, natural gas, data centers, Nvidia, SpaceX, Starship, robots suddenly all become part of the same story.
E