Investor Chamath Palihapitiya’s new research maps the six layers powering artificial intelligence—from rare minerals and semiconductor machinery to robots and software. The findings reveal an enormous technological transformation, unexpected vulnerabilities, and opportunities extending far beyond Silicon Valley.
Deep in the mountains of western North Carolina, near the small town of Spruce Pine, miners extract a material most Americans have never thought about.
It isn’t gold.
It isn’t lithium.
And it isn’t one of the rare-earth minerals frequently mentioned in discussions about technological competition with China.
It’s quartz.
But not ordinary quartz.
The region contains deposits of exceptionally pure quartz used in manufacturing components essential to producing semiconductor wafers—the foundation of the microchips powering smartphones, computers, artificial intelligence, and much of the modern economy.
Without sophisticated semiconductor manufacturing, the extraordinary advances we’re witnessing in artificial intelligence would not be possible.
That makes this quiet corner of North Carolina an important part of a global technological supply chain.
And according to billionaire investor Chamath Palihapitiya, who founded the investment firm Social Capital, understanding these often-overlooked pieces of infrastructure may be essential to understanding where the next generation of wealth and economic power will accumulate.
In May 2026, Palihapitiya published an introduction to a 138-page research project titled Where Value Accrues in the AI Stack.
His team examined the entire system required to create, operate, and distribute artificial intelligence.
Their central question was deceptively simple:
As AI transforms the global economy, which companies will capture the greatest economic value—and which will become increasingly replaceable?
The answer may surprise people who assume the biggest winners will simply be the companies producing the most impressive chatbots.
Artificial Intelligence Is More Than Software
When most people think about AI, they imagine typing a question into ChatGPT, generating an image, or asking an assistant to complete a task.
But the software interface is only the visible end of an enormously complex industrial system.
Behind every AI response are data centers, computer chips, electrical grids, specialized manufacturing equipment, raw materials, networking infrastructure, and sophisticated software.
Palihapitiya’s research organizes this system into six interconnected layers.
The Six Layers of the AI Economy
| Layer | What it provides |
|---|---|
| 2. Chips | Processors and the equipment needed to manufacture them |
| 3. Data | Information used to train and improve AI systems |
| 4. Models | The artificial intelligence systems themselves |
| 5. Execution | Agents and software that make AI perform tasks |
| 6. Applications | Products and services people actually use |
Each layer depends on capabilities supplied by others.
A sophisticated model is useless without sufficient computing power.
Computing power requires advanced chips.
Advanced chips require specialized manufacturing equipment, materials, and energy.
And an excellent AI model doesn’t automatically create a profitable business unless customers find its applications valuable.
The fundamental insight is that being technologically important and capturing economic profits are not necessarily the same thing.
Some industries may experience explosive demand while competitive pressure drives down profit margins.
Others may benefit from controlling scarce capabilities that remain difficult to replace.
Palihapitiya calls these strategically important bottlenecks fulcrum assets.
Understanding them provides a different way to evaluate the AI economy.
The Most Important AI Companies Might Not Make AI Models
Consider the Dutch company ASML.
Most consumers have never purchased an ASML product.
They may never have heard the name.
Yet the company occupies an extraordinary position in semiconductor manufacturing.
ASML is the world’s only commercial manufacturer of extreme ultraviolet lithography systems, often called EUV machines.
These exceptionally complex systems use light to help print intricate patterns onto silicon wafers, enabling production of many of the world’s most advanced chips.
Companies such as Taiwan Semiconductor Manufacturing Company depend on sophisticated lithography equipment and other manufacturing systems to produce leading-edge semiconductors.
ASML doesn’t need to develop a popular chatbot to benefit from demand for increasingly capable computing hardware.
It supplies technology that advanced semiconductor manufacturers depend upon.
Another example involves photoresists.
These are specialized light-sensitive chemical materials used during semiconductor manufacturing.
When exposed to precisely controlled light, a photoresist helps determine which patterns can be transferred onto a wafer.
Several Japanese companies—including Tokyo Ohka Kogyo, JSR, Shin-Etsu Chemical, and Fujifilm—are important suppliers in this specialized industry.
Their products are largely invisible to consumers.
But the manufacturing process cannot simply dispense with the necessary chemicals.
That is the kind of economic position Social Capital wants investors to examine.
A business may occupy an extremely valuable position not because customers recognize its brand, but because replacing its technology, manufacturing expertise, or production capacity would be extraordinarily difficult.
For investors, entrepreneurs, and policymakers, that raises a larger question:
In a world where intelligence becomes cheaper, what remains scarce?
Why North Carolina Matters More Than You Might Think
This brings us back to Spruce Pine.
According to industrial minerals producer Sibelco, the company’s IOTA high-purity quartz products are extracted from particularly pure mineral deposits in the Spruce Pine region.
The material is used to produce fused-quartz crucibles and other specialized components employed during silicon-wafer manufacturing.
Its exceptional purity makes it highly valuable.
Sibelco announced plans in 2023 for approximately $200 million in investment to expand its high-purity quartz operations in the region.
The location is not the world’s only possible source of quartz, nor is every semiconductor dependent on one individual mine, as some simplified descriptions suggest.
But the concentration of exceptionally high-quality deposits and established processing capability makes Spruce Pine unusually important.
That vulnerability became visible during Hurricane Helene in September 2024.
The storm devastated communities across western North Carolina and disrupted operations in the region’s quartz-mining industry.
Companies, suppliers, and technology analysts were forced to consider how a natural disaster in the Appalachian Mountains could affect semiconductor supply chains around the world.
The lesson extends far beyond quartz.
Advanced technology depends on physical resources, manufacturing expertise, transportation networks, and infrastructure concentrated in particular locations.
Those dependencies create opportunities.
They also create risks.
For North Carolina, this means that participation in the AI economy does not have to be limited to software startups in the Research Triangle or technology companies in Charlotte.
Industrial materials, engineering, specialized manufacturing, power systems, and infrastructure may also become increasingly important.
The Coming Electricity Challenge
Artificial intelligence consumes electricity.
A great deal of it.
Advanced AI processors perform enormous numbers of calculations, and the facilities housing them require power, cooling, and reliable connections to the electrical grid.
This is one of the most significant physical constraints on continued AI expansion.
The International Energy Agency projected in its 2025 Energy and AI report that global data-center electricity consumption could more than double to approximately 945 terawatt-hours annually by 2030 under its central scenario.
That is a projection, not a guaranteed outcome. Efficiency improvements, technology choices, grid constraints, and investment decisions could materially change the result.
Still, the scale is substantial.
And building new electricity infrastructure can take significantly longer than developing new software.
Power plants, transmission lines, substations, and major industrial facilities require capital, equipment, permits, specialized labor, and often lengthy construction schedules.
This creates potential demand for businesses supplying electrical equipment, backup generators, cooling systems, energy storage, industrial construction, and grid-management technology.
But it also creates costs and challenges for surrounding communities.
Large electricity users can intensify concerns about energy prices, land use, water consumption, and whether public infrastructure investments are being allocated fairly.
AI-related development should therefore be evaluated not merely by the size of a company’s investment announcement, but by the actual economic benefits and costs experienced locally.
The AI revolution is becoming an energy and infrastructure story as much as a software story.
What Happens When Intelligence Becomes Cheap?
While physical infrastructure remains expensive and difficult to expand, a very different trend is unfolding in AI software.
The cost of using capable AI models has declined rapidly.
According to Stanford University’s 2025 AI Index Report, the cost of running an AI model at roughly GPT-3.5-level performance fell from approximately $20 per million tokens in November 2022 to just seven cents by October 2024.
That represents a decline of more than 280 times in roughly 18 months.
Social Capital argues that the longer-term reduction in certain model-computing costs has been even more dramatic.
These figures refer to particular performance benchmarks and time periods. They do not mean that operating the newest, most capable AI models is literally free.
Training frontier models can still require enormous financial resources, and operating reliable AI services involves more than paying for individual tokens.
Nevertheless, the economic direction is significant.
Capabilities that once required expensive computing resources are becoming accessible to smaller companies and individual entrepreneurs.
Think about what happened to digital photography.
Once smartphones made taking photographs inexpensive and convenient, the ability to capture an image became commonplace.
But that didn’t eliminate every profitable photography-related business.
Value shifted toward different capabilities, including software, distribution, specialized equipment, services, and compelling creative work.
Artificial intelligence may follow a similar pattern.
When thousands of companies can access comparable language models, merely offering access to one may no longer be enough to sustain high prices.
The question becomes what a business can do with that intelligence.
Does it solve a difficult customer problem?
Does it improve productivity?
Does it possess valuable proprietary information?
Does it operate within an established customer relationship?
Can it deliver reliable outcomes better than competitors?
These questions matter more than whether the company’s marketing materials prominently feature the letters AI.
The Real Opportunity for Small Businesses
Imagine a family-owned HVAC company in Charlotte.
The company receives calls, schedules service appointments, prepares estimates, orders replacement components, tracks maintenance agreements, and communicates with customers.
Today, many of those activities require employees to repeatedly enter information, answer routine questions, or move data between disconnected software systems.
An AI-enabled business could potentially streamline parts of that work.
A system might summarize incoming service requests, prepare appointment confirmations, organize technician notes, flag overdue maintenance, and draft follow-up communications.
Qualified employees would still need to verify important details, approve technical recommendations, and handle safety-critical decisions.
But reducing repetitive administrative work could allow the company to serve more customers without proportionately increasing office overhead.
The real economic value wouldn’t necessarily belong to the company that created the underlying language model.
It might belong to the HVAC business that combines its customer relationships, industry experience, operational processes, and appropriate AI tools.
The same principle applies to accounting practices, insurance agencies, manufacturers, publishers, property managers, and other established businesses.
Artificial intelligence may be especially valuable when combined with expertise and systems that already solve problems people willingly pay to have solved.
For entrepreneurs, that suggests a different strategy from trying to invent the next ChatGPT.
Find a recurring, expensive, frustrating problem within an industry you understand.
Use reliable technology to solve it better.
Then build a business around the customer relationship and measurable results.
The Next Frontier: Artificial Intelligence Enters the Physical World
Social Capital’s research highlights another important distinction.
There is software AI, which processes information and performs digital tasks.
And there is physical AI, which must interact with the real world.
Think robotics, autonomous vehicles, industrial automation, and intelligent machines.
An AI assistant can produce a written response without understanding how much physical force is required to pick up a drinking glass.
A robot must handle entirely different problems.
It needs sensors, motors, control systems, batteries, mechanical components, and the ability to operate safely in changing environments.
Physical machines must deal with gravity, friction, imperfect surfaces, moving objects, and the unpredictable behavior of people.
That makes physical AI a substantially different engineering challenge.
Social Capital emphasizes two especially important constraints: energy storage and actuation.
Energy storage determines how long a machine can function without being recharged or connected to an external power supply.
Actuation refers to the components that convert commands into physical motion.
Better AI software is not enough if a robot cannot move reliably, manipulate objects precisely, or operate economically.
This creates potential opportunities in batteries, electric motors, precision manufacturing, industrial sensors, mechanical engineering, and the software that coordinates physical systems.
North Carolina already has manufacturing and engineering capabilities that could participate in parts of this expanding ecosystem.
But widespread commercial deployment will depend on whether these systems can perform valuable tasks at acceptable costs and safety standards—not simply whether they can produce impressive demonstration videos.
Where Could the Financial Rewards Accumulate?
The Social Capital framework suggests that the most durable economic advantages may arise where a company controls an essential capability that competitors cannot easily reproduce.
Those advantages can take several forms.
A manufacturer may possess unusually difficult technical expertise.
A supplier may control scarce production capacity.
A technology company may have software tools that developers have spent years learning to use.
A service provider may own deep customer relationships, specialized data, or an established distribution network.
An energy developer may control access to an appropriately located, permitted site with available electrical capacity.
But there is an important caution.
An essential industry is not automatically a good investment.
A company can operate in a rapidly growing market and still produce disappointing shareholder returns if investors pay too much for its stock, competitors drive down prices, or its expansion requires enormous capital expenditures.
The history of technology includes many examples of valuable infrastructure being built by companies that ultimately struggled financially.
Social Capital itself acknowledges this distinction in its investment writing, citing earlier telecommunications infrastructure projects that suffered when excessive construction collided with poor economics.
The current AI boom raises similar questions.
How much of the projected demand will materialize?
Which companies can earn attractive returns on invested capital?
How quickly will expensive chips or computing facilities become obsolete?
And how much growth is being financed with debt that must eventually be repaid?
In an October 2026 research note, Palihapitiya also examined the expanding role of private credit in financing AI infrastructure.
That is an important reminder that technological progress and financial risk can increase at the same time.
Investors should examine cash flow, balance sheets, competitive advantages, capital requirements, and valuation rather than treating exposure to AI as sufficient justification to buy a stock.
What This Means for Workers and Entrepreneurs
For people who do not own semiconductor factories or investment funds, the implications are still substantial.
The AI economy will need more than software engineers.
Electrical contractors, skilled tradespeople, industrial technicians, cooling specialists, construction professionals, cybersecurity experts, and people capable of integrating AI into real business operations may all find relevant opportunities.
The specific demand for individual occupations will vary, and AI may also automate tasks currently performed by workers.
That makes adaptability important.
One practical approach is to combine expertise in a real industry with the ability to use modern AI tools effectively.
An accountant who understands automation and financial controls.
A construction manager who can coordinate digital planning systems.
A healthcare administrator who can improve operations without compromising patient privacy.
A journalist who can use AI to accelerate research while independently verifying facts.
These are examples of combining technological capability with human judgment.
For business owners, the best starting point may be even simpler:
Identify a repetitive process that consumes substantial employee time.
Test whether an AI-assisted workflow can improve it.
Measure the result.
Does it reduce errors?
Does it save meaningful time?
Does it improve customer satisfaction?
Does it increase profitability after implementation and oversight costs?
If the answer is yes, expand deliberately.
If not, the mere presence of AI does not make the project worthwhile.
Five Questions Everyone Should Be Asking About AI
The next decade of AI development will almost certainly involve surprises.
Rather than attempting to predict every winning company, individuals and businesses can use five questions to evaluate opportunities:
1. What is becoming cheaper?
As useful AI capabilities become more affordable, more people and companies can access them. That creates opportunity while increasing competition.
2. What remains difficult to reproduce?
Scarce infrastructure, engineering capabilities, trusted relationships, proprietary information, and effective operations may retain value.
3. Who controls the essential inputs?
Electricity, advanced equipment, materials, and specialized suppliers can determine how quickly new technologies are deployed.
4. Who owns the customer relationship?
A company providing an indispensable service may capture value even when the underlying technology is available from multiple vendors.
5. Does the business actually make money?
Promising technology is not the same as sustainable economics. Look for durable customer demand, reasonable costs, and credible financial results.
Together, these questions provide a more useful framework than simply asking which AI company is attracting the most attention.
The Bigger Picture: The AI Revolution Is an Industrial Revolution
One of the most interesting conclusions from Social Capital’s research is that artificial intelligence should not be understood solely as another generation of computer software.
It is increasingly reshaping the relationship between computing, energy, manufacturing, automation, and the physical world.
The technology running inside a data center depends on global industrial supply chains.
A leading-edge chip requires extraordinarily complex manufacturing equipment and specialized materials.
A successful robotics system depends on mechanical engineering, batteries, sensors, and reliable physical components.
And the ultimate economic value depends on whether these technologies provide services that people genuinely need.
There will be extraordinary opportunities.
There will also be failed companies, excessive speculation, technological disappointments, and difficult choices about energy, employment, and infrastructure.
But the central lesson is worth remembering.
The history of great technological transformations is not only the history of the most famous inventions.
It is also the history of the businesses that made those inventions possible.
Railroads needed steel, land, locomotives, and financing.
Automobiles needed roads, fuel, factories, and repair networks.
The internet needed semiconductors, fiber-optic cables, servers, and electrical infrastructure.
Artificial intelligence is no different.
Its future will be shaped not only by the intelligence of its models, but by the enormous physical and economic systems supporting them.
And somewhere in the mountains of North Carolina, workers extracting exceptionally pure quartz are already participating in that future.
The next generation of wealth may not be created solely by the companies building artificial intelligence. Much of it may be created by the people and businesses making artificial intelligence possible—and those who discover how to put it to work.
Sources and Further Reading
- Social Capital — Deep Dive: Where Value Accrues in the AI Stack
- Social Capital — 2025 Annual Letter
- Social Capital — Is the AI Buildout Creating a Private Credit Bubble?
- ASML — EUV Lithography Systems
- Sibelco — High-Purity Quartz and Semiconductor Applications
- Sibelco — Spruce Pine Quartz Expansion
- International Energy Agency — Energy and AI
- Stanford University — 2025 AI Index
- U.S. Department of Energy — Data Center Electricity Demand
- PBS North Carolina — How Hurricane Helene Impacted the Technology Industry
Editor’s note: This article analyzes the publicly accessible findings from Social Capital’s May 2026 research. The complete 138-page report is available to paid subscribers and was not independently reviewed in full. Investment arguments are presented as research perspectives, not financial recommendations. AI demand, prices, and industry structures may change.