Only a couple of years ago, artificial intelligence was mostly talked about as something experimental: impressive demos, amusing conversations, pictures “on demand”. Today it is increasingly just an ordinary tool, like search, spreadsheets or email. And that raises an important question: why, in some countries and cities, does AI quickly become a working lever for business and government, while in others it remains a “toy for the chosen few”?
The answer is not that someone “lacks talent” or “got unlucky with technology”. The reason is almost always mundane: for AI to be useful, it needs a foundation. The same kind that electricity was for factories, or roads for logistics. Without that foundation, AI either never takes off or widens the gap between those who have everything and those who don’t.
In this article we look at the bricks that make up a “foundation for AI”, why the world is already facing a new kind of digital inequality, and what helps to close the gap in practice.
The world is entering the age of AI — but not all at once
Technologies spread unevenly. That is normal: first they appear in rich centres, then they gradually get cheaper and reach everyone else. That is how it went with the internet, smartphones and the cloud. But AI has one particular feature: it depends heavily on resources — computing power, data and people who know how to work with it properly.
So “we have the internet too” does not guarantee that AI will be adopted as quickly as in the leading countries. You can download an app, open a chatbot and even bring a couple of tools into the office. But as soon as the conversation turns to large-scale impact on the economy — teaching in schools, higher productivity in companies, the growth of local services — it turns out that the foundation is not the same everywhere.
And this leads to a new type of divide: not simply “internet / no internet”, but “good-quality digital infrastructure and skills / none”.
Four foundations AI cannot work without
To understand what exactly a given country (or even a single region within a country) is missing, it helps to break “AI readiness” down into four parts. Think of them as the four legs of a chair. If one is broken, sitting will be uncomfortable, and sometimes dangerous.
Connectivity and access: internet, devices, energy
Let’s start with the obvious: AI tools live on the internet. But what matters is not just having a connection but its quality.
Many countries have taken a huge step forward: more people have got online, and mobile coverage has expanded. The problem lies elsewhere: speed and reliability are improving unevenly. Where connections are slow or expensive, people and companies simply cannot use modern digital services on a regular basis. And AI, especially in the form of “assistants” and cloud platforms, likes fast, stable internet.
Energy is part of the same picture. Data centres, networks, servers — all of it needs electricity. If the power keeps cutting out, the digital economy will always be spinning its wheels.
Compute: where the “brains” for models come from
AI is not just a piece of software. It is computation. To train a model, or even just to run it at a useful scale, you need powerful servers, graphics accelerators, data centres and access to cloud platforms.
Here the global picture resembles the oil fields of the past: resources concentrate where there is money, industry and supply chains. As a result, some countries have a large “computing safety margin”, while others have only a limited one.
For most people this sounds abstract until they run into reality:
— local companies cannot afford expensive capacity;
— public services do not scale;
— universities struggle to do research;
— start-ups depend on other people’s clouds and other people’s rules.
Context: data, language, local reality
AI becomes useful when it understands your reality: the language, the documents, local processes, domain knowledge.
If models are trained mostly on data from large countries and in widely spoken languages, they will work worse for many regions. Even if “everything is clear” in general, quality can drop precisely where it matters: in legal wording, medical terms, educational materials, local names and specifics.
Context is not only about language. It is also data on how the economy and society work: which services are in demand, which documents are in circulation, what mistakes people typically make, which risks matter here in particular.
Without context, AI turns into a “one-size-fits-all adviser” that talks well but often misses the details.
Skills: people’s abilities and the capacity to implement
The most underrated part is people.
You can buy the tools, connect to the cloud and even get access to good models. But if the population has low digital literacy and companies have no specialists who know how to implement solutions, the effect will stay patchy.
And this is not only about programmers. In the age of AI, what matters is skills for the many:
— how to frame tasks and check the result;
— how to work with data and think critically;
— how to stay safe and not “leak” personal information;
— how to rebuild processes around new tools.
Skills are the bridge between technology and real benefit. Without that bridge, AI stays at the level of experiments.
The new digital divide: quality matters more than simply being connected
In the past, much of the discussion revolved around the question “how many people are connected to the internet”. Now something else matters more and more: what that internet lets you do.
In rich countries, data use per person and connection quality are growing very fast. That means people there can comfortably study online, watch long lectures, work with demanding services, use video calls without stress and plug in cloud tools as a “second operating system”.
In poor countries access is growing too, but it often remains expensive, unreliable or too slow for modern use cases. On paper, the internet is there. But it cannot handle proper learning, remote work and large-scale adoption of digital services.
And this is where AI becomes not a “great equaliser” but an accelerator: where the base is strong, growth is faster.
“Small AI” as an opportunity: not everyone needs a supercomputer
There is good news too.
To get value from AI, you do not have to build giant models “from scratch”. At the start, what most countries and companies need more is applied solutions:
— assistants for public services and help desks;
— automation of standard documents;
— translating and simplifying texts;
— analysing customer requests;
— tools for training and professional development.
All of this can be done on top of cloud services, ready-made models and open technologies. This approach is sometimes called “small AI”: less fanfare, more use.
But there is a condition: even “small AI” needs the four foundations. Without decent connectivity, access will not be convenient. Without skills, it cannot be implemented. Without context, the quality will fall short. And without compute (even if it is rented in the cloud), there will be no scale.
In other words, “small AI” is not a magic button but a more realistic route.
Open source lowers the barrier to entry but does not remove the cost of implementation
Open models and tools really are changing the market. They make AI more accessible: you can adapt solutions to your own tasks, build local products, teach students and run pilots without huge licence fees.
But it is important to grasp one simple thing: a “free model” does not mean “free implementation”.
The real costs are often hidden elsewhere:
— collecting and cleaning data;
— infrastructure and security;
— training people;
— support and improvements;
— responsibility for mistakes.
So a sensible strategy usually looks like this: use open solutions where they give flexibility and independence, but do not forget about quality, security and organisational maturity.
AI and the job market: demand is rising, but skills are unevenly spread
When people say that “AI will take our jobs”, they usually paint the same picture: robots pushing people out.
Reality is usually more complicated. In practice, the job market changes through skills.
Right now, demand is growing for the ability to work with AI tools. And not only in IT. More and more tasks in marketing, design, content, analytics and even office administration are starting to require a basic understanding of how to use AI and how to check the result.
But here inequality shows itself again: where basic digital literacy is low, people cannot retrain fast enough, and companies cannot quickly rebuild their processes.
There is another layer to the problem: the brain drain. Data and AI specialists often leave for places with higher salaries, better infrastructure and more room to grow. This widens the gap: strong ecosystems become even stronger.
Why regulation matters, but must not stifle development
AI brings not only benefits but also risks: data leaks, fraud, errors in important decisions, amplified bias.
So you cannot talk about AI without talking about rules.
But it is easy to get the focus wrong here. If a country tries to “write the perfect AI law straight away” without any experience of implementation or basic infrastructure, the result can be odd: there is nothing to regulate yet, but the restrictions are already getting in the way.
A workable approach is usually more pragmatic:
— strengthen data protection and cybersecurity;
— introduce clear requirements for accountability and transparency in sensitive areas;
— develop standards and risk-management practices;
— teach organisations to use AI properly, rather than only banning things.
In other words, the goal is not to slow the technology down but to make it work safely.
What a sensible route looks like for a country or region
There is no universal formula, because the starting conditions differ. But the logic is usually the same: first close the basic gaps, then add complexity.
If a region has weak infrastructure, it should start by improving connectivity, the availability of devices and basic digital literacy.
If the infrastructure is acceptable, the next step is context and implementation: local data, language support, applied products, training implementation specialists.
If a region wants to become a development centre, it will have to think about a computing base, research centres and a venture capital environment as well.
What matters is this: almost everywhere, the biggest effect comes not from one-off “showcase projects” but from large-scale programmes that reach schools, universities, companies and public services at the same time.
How to narrow the gap in practice
Access to a model is only the first layer. You also need stable internet, digital literacy, clear interfaces, training and a way to check the result. Without these conditions, AI strengthens users who are already strong faster than it helps those who never had basic support.
The main takeaway
AI is not magic and not “an app you need to download”. It is a layer of technology that rests on infrastructure and people.
When there is good connectivity, affordable compute, local context and strong skills, AI quickly becomes part of the economy and helps raise productivity.
When those things are missing, AI remains a superficial tool and can even deepen inequality.
So the question “are we ready for AI?” is really a question of whether our digital foundation is ready.
References:
World Bank. Digital Progress and Trends Report 2025: Strengthening AI Foundations. 2025.
