​

We are entering an era in which intelligence itself is becoming a luxury good. Not human intelligence in the biological sense. Artificially augmented intelligence, meaning the ability to analyze more information, test more scenarios, automate decisions, detect risks and act faster than an unaided person or organization ever could. As artificial intelligence becomes more powerful, the people, companies and countries with access to the best models, proprietary data, reliable electricity and abundant computing infrastructure will not just become more productive.

​

• They will make decisions faster.

• They will identify opportunities earlier.

• They will learn from mistakes more quickly.

• They will see risks before everyone else sees them coming.

• The rich will not just have more money.

• They will have better brains.

​

That statement is deliberately provocative, but the underlying problem is already real. AI may eventually become one of history’s greatest equalizers. It could give a rural physician access to world-class diagnostic support, help a small business compete with a multinational corporation and place an expert tutor in the hands of nearly every student. But that outcome is not guaranteed. If access to advanced AI becomes determined by wealth, geography and infrastructure ownership, the technology could produce an inequality divide more consequential than the digital divide that preceded it.

​

​

​

 

 

 

 

 

 

 

 

 

​

 

 

​

​

 

​

The Access Divide

People love to speak about AI access as though having a free chatbot account means the playing field has been leveled.

It has not and there is an enormous difference between occasionally using a consumer AI assistant and operating an integrated intelligence system connected to proprietary data, specialized models, secure infrastructure and continuously running autonomous agents. One person may use AI to improve an email, and another organization may use it to monitor thousands of market signals, model competitors, review contracts, optimize prices, screen investments and deploy capital, all before the first person finishes breakfast.

​​

Technically, both have “access to AI.” Economically, they are not participating in the same reality. The good news is that the cost of using capable models has fallen dramatically. Stanford’s 2025 AI Index found that the cost of querying a system performing around the level of GPT-3.5 declined more than 280-fold between November 2022 and October 2024. The unseating part is that control remains heavily concentrated. The same report found that private-sector companies, not universities or governments, developed nearly nine out of every ten notable AI models released in 2024. Advanced AI may be getting cheaper to use, but the infrastructure, capital and talent required to build and control it remain concentrated among a relatively small number of institutions. Stanford Institute for Human-Centered AI.

​​

In short, affordable access to someone else’s model is not the same as owning the infrastructure, governing the data or controlling the terms under which the intelligence is available. A company that depends entirely on an external AI provider can face changing prices, usage limits, discontinued models, restricted capabilities or policies established by a platform it does not control. Renting access may be practical, but total dependency is not a long-term strategy.

​

The emerging divide will therefore not be between people who use AI and people who do not. It will be between:

​

• People who consume intelligence and institutions that control it.

• Companies that rent general-purpose tools and companies that operate proprietary intelligence systems.

• Countries that import AI capability and countries that own the infrastructure required to produce it.

• Workers assisted by occasional automation and organizations surrounded by persistent, autonomous digital labor.

 

The World Bank’s 2025 Digital Progress and Trends Report describes four foundations required for meaningful AI participation:

 

connectivity, compute, context and competency. 

 

In plain English, people need a reliable connection, sufficient processing power, relevant data and the skills to use the technology. Missing any one of those pieces can leave a person, or an entire country, standing outside the AI economy looking in.

​

​

​

​

​

​

​

​

​

​

 

 

 

 

 

 

 

 

Compute Is Infrastructure

AI does not float invisibly in “the cloud.” The cloud is someone else’s building and inside that building are chips, servers, cooling systems, fiber connections, backup generators, security systems and enormous amounts of electrical infrastructure. Every AI response ultimately depends on physical machinery operating somewhere else which is what makes compute infrastructure in the first place. The automobile economy could not scale without roads, fuel stations and supply chains. AI cannot scale without data centers, power generation, transmission capacity and connectivity. A brilliant model without dependable compute is like a Ferrari without fuel: impressive, expensive and completely useless once it stops moving.

​

According to the International Energy Agency, data centers consumed approximately 415 terawatt-hours of electricity in 2024, representing about 1.5% of global electricity consumption. The IEA projects that demand could more than double to approximately 945 terawatt-hours by 2030. A typical AI-focused data center can consume as much electricity as 100,000 households. Some of the largest facilities under development could consume twenty times that amount. International Energy Agency Those numbers do not mean we should stop building AI infrastructure. They mean we need to stop pretending the infrastructure question is someone else’s problem. Where data centers are built, who finances them, who supplies their electricity and who receives priority access to their computing capacity will influence the distribution of economic power.

If most advanced compute is concentrated in a small number of hyperscale campuses controlled by a handful of corporations, society is not centralizing technology, it is also centralizing access to intelligence. In tow, this creates several risks.

​​

First, physical concentration creates fragility. A grid constraint, equipment shortage, natural disaster, cyberattack or political dispute can disrupt enormous amounts of capacity. Second, geographic concentration can leave entire regions dependent on distant infrastructure that may not have been designed around their needs. Third, corporate concentration gives a small number of providers tremendous influence over pricing, availability and permissible use. Finally, uncontrolled development can transfer infrastructure costs onto local communities. Residents may face pressure on electricity systems, water resources and land while receiving little direct economic benefit in return.The IEA estimates that approximately 20% of planned data-center projects could face delays unless grid bottlenecks are addressed. It also notes that building major transmission infrastructure can take four to eight years in advanced economies. AI may move at software speed, but you cannot download a transmission line, and the electrical grid most certainly does not move at software speed. International Energy Agency 

​

Why Decentralization Matters

The answer is not to eliminate hyperscale data centers. Frontier-model training and other extremely demanding workloads require enormous concentrations of computing power. The answer is to avoid making hyperscale infrastructure the only viable path to AI participation. A more resilient system would combine large-scale facilities with regional data centers, edge-computing sites, smaller modular facilities and specialized infrastructure positioned closer to the communities and industries being served. Different workloads require different infrastructure. Training a frontier model may require a massive centralized campus, but running a healthcare application in a hospital, supporting advanced manufacturing, operating autonomous systems or delivering AI services to a regional business network may not.

 

Placing appropriate computing resources closer to where data is produced can reduce latency, improve resilience, strengthen privacy and allow institutions to retain greater control over sensitive information. But decentralization should not become another empty technology slogan or innovation theatre. Placing a few servers in smaller buildings does not automatically democratize AI. If the same handful of companies owns the hardware, controls the software and dictates access, the map may look decentralized while the power structure remains exactly the same.

​​

Real decentralization requires diversity across several layers:

• Geographic distribution of computing capacity.

• Multiple infrastructure owners and operators.

• Interoperable systems that reduce platform dependency.

• Affordable access for startups, universities and public institutions.

• Locally relevant data and models.

• Financing structures that allow more communities to participate in ownership.

• Clear rules preventing infrastructure costs from being quietly transferred to ordinary ratepayers.

 

Communities should also have a loud and heard voice in how these projects are developed. Opposition to data centers is many times dismissed as resistance to progress. However, that does not hold weight, because when a community is asked to absorb additional power demand, environmental pressure, construction disruption and long-term infrastructure risk, residents are entitled to ask what they receive in return. A worthy development model should include transparent environmental and grid-impact assessments, developer-funded infrastructure improvements, local hiring commitments, community-benefit agreements and protections against stranded costs. New York’s 2026 data-center policy direction, for example, contemplated requiring developers to contribute toward grid improvements and distributed energy resources rather than leaving ratepayers to absorb the cost. State of New York. That should not be considered radical, that is progressive. If an AI facility creates private value while imposing public costs, the owners should pay their share. Capitalism works considerably better when the bill follows the beneficiary.

​​

The New Cognitive Class System

Every major technology creates winners and losers. AI is different because it can influence the speed and quality of nearly every other economic activity. It can assist with education, medicine, finance, logistics, legal analysis, cybersecurity, military planning, scientific discovery and government administration. This means unequal AI access will not remain confined to the technology industry. It will compound advantages across society in every way. A wealthy student may have a persistent, personalized tutor trained around that student’s weaknesses and learning style. Another student may receive limited access to a generic system, or no access at all. A well-capitalized company may employ thousands of AI agents monitoring operations around the clock. A smaller competitor may still rely on exhausted employees and disconnected software. A country with domestic compute infrastructure may develop locally relevant systems, protect sensitive data and capture economic value. Another may be forced to export its data and rent intelligence from foreign providers. These are not just differences in convenience, they are differences in cognitive leverage. Money has always purchased better information, stronger networks and more skilled advisers. AI industrializes and multiplies that advantage, where it transforms superior decision-making from a scarce human service into an always-available system that can operate at machine speed. That is why the AI divide could become self-reinforcing. The organizations with the best intelligence will make better decisions, and better decisions will generate more capital and data. More capital and data will purchase better models and infrastructure, and those improved systems will then make still better decisions. Once that feedback loop becomes entrenched, catching up becomes extremely difficult.

​

Who Controls the Future?

The future of AI will not be determined by model capability alone. It will be determined by ownership.

 

Who owns the data centers?

Who controls the chips?

Who receives access during periods of scarcity?

Who sets the prices?

Who owns the data used to improve the systems?

Who can audit the models?

Who is allowed to build on top of them?

Who can be disconnected?

​

These questions are not secondary to AI policy, they are AI policy. Governments have a role, but government ownership of everything is neither realistic nor desirable. Private industry has a role, but allowing a few corporations to govern global intelligence infrastructure through terms-of-service agreements is not a serious public strategy either. The answer must be a mixed ecosystem. Governments should protect competition, modernize electrical grids, expand connectivity and create shared access for researchers, schools and public institutions. Investors should finance alternatives to complete hyperscale concentration, including regional infrastructure, energy-integrated data centers and efficient computing platforms. Technology companies should build systems that are interoperable, auditable and capable of operating across multiple infrastructure providers. Universities and workforce programs must treat AI competency as a basic economic skill rather than an elite technical specialty. Communities should participate in both the planning and economic benefits of the infrastructure located in their backyards. And individuals must understand that using AI is not the same as controlling it.

​

Access Must Be Designed, Not Assumed

There is still time to shape the outcome. AI systems are becoming less expensive to operate, hardware is becoming more efficient and open-weight models are narrowing parts of the performance gap with proprietary alternatives. Those trends create a genuine opportunity to distribute capability more broadly. Stanford Institute for Human-Centered AI 

 

But falling costs do not automatically produce equal access. Electricity became widespread because societies built generation, transmission, regulation and distribution systems around it. Automobiles became transformative because roads, fueling infrastructure and financing became broadly available. AI will require the same seriousness, and we need infrastructure that does more than enrich its owners. We need systems that expand productive capacity without turning intelligence into a gated community. This is not an argument against wealth, scale or private investment. Those forces will be essential and imperative to building the infrastructure AI requires. It is an argument against confusing technological availability with impactful access. A free chatbot cannot compensate for a society in which advanced compute, proprietary data and autonomous systems are reserved for those already holding the most power. 

 

The defining inequality of the next generation may not be between people who have money and people who do not. It may be between those who can continuously amplify their intelligence and those who must compete without that amplification. 

 

The rich will not just have more money, they will have better brains, unless we decide that access to intelligence is infrastructure worth distributing.