Society’s digital infrastructure

Digital sovereignty of nations, Estonia’s next digital leap, and local implementation

Software has become society’s invisible infrastructure. Money, communications, business, education, healthcare and public services depend ever more on systems whose presence we often notice only when something fails. Artificial intelligence takes this change into the next stage: software is no longer only a tool a person uses, but increasingly creates, processes data, communicates with other systems and acts on behalf of a person or an organisation. That also changes what we expect from digital infrastructure. Alongside speed, convenience and ordinary security, control, sovereignty and the ability to prove what a system actually did become steadily more important.

The EFTI source “Crypto Cloud: ICP as a Cloud Protocol and the Internet’s Next Software Foundation” looks at the technical foundation of this change and explains why the Internet Computer Protocol, or ICP, can be treated as a complete cloud protocol.[1]EFTI, “Crypto Cloud: ICP as a Cloud Protocol and the Internet’s Next Software Foundation”.efti.org This essay deliberately moves one layer further. The central question is no longer how ICP works technically, but what such a foundation allows a society to build. The view moves from the global scale to nations, from there to the next possible layer of Estonia’s digital state, and finally to the local community and the individual. These are not three separate worlds, but different scales of one digital system.

A new digital sovereignty

Cloud computing has made creating software extraordinarily powerful and accessible. A company, a school or a public institution no longer has to build its own data centre. It can use ready-made compute, databases, identity services and dozens of other infrastructure layers. The other side of that convenience is an ever deeper dependence on a few large cloud providers, their technology supply chains, their administrators and their jurisdiction. An organisation may own its data and its software, while an important part of what actually keeps them running remains outside its direct control.

For states, the question becomes especially sharp, because digital infrastructure has become part of the state’s own capacity to function. Digital sovereignty therefore does not mean only that a server is located on the state’s territory or that data is kept in the European Union. It is equally important who operates the infrastructure, who can change the system, what the keys and access depend on, and how far the correct operation of critical software rests on trusting a single operator or administrator. AI adds a new dimension: as software acts more independently, the need grows, together with its capability, to determine what it is allowed to do and later to check what it actually did.

In this context ICP becomes interesting for society. Cryptographic computation is not limited here to assets and transactions. A whole application can run on the same protocol, together with its logic, data, identity and the web layer a user sees. In 2026 this was joined by Cloud Engine: ICP’s private subnet model, in which an organisation can choose the underlying machines, their operators, their geographic location and the level of replication, while applications run on the same ICP software model.[2]Internet Computer, Cloud engines, Internet Computer Protocol and ICP Quick Start. The public network is a shared ICP environment; a cloud engine is a private subnet configurable by one owner, for which the nodes, operators, geography and replication can be chosen.internetcomputer.org Physical infrastructure can therefore remain under the control of a particular state or organisation, without simply building yet another conventional server farm on top of it.

A cloud engine is not a new protocol beside ICP, but another way of using the same protocol. The public Internet Computer is a shared global compute environment where anyone can install software and where compute, storage and network resources are paid for in cycles.[2]Internet Computer, Cloud engines, Internet Computer Protocol and ICP Quick Start. The public network is a shared ICP environment; a cloud engine is a private subnet configurable by one owner, for which the nodes, operators, geography and replication can be chosen.internetcomputer.org A cloud engine, by contrast, is a portion of the same protocol reserved for one owner. For the software, the canister model, replicated execution and consensus remain. The physical composition and location of the infrastructure can be shaped to a particular need.

The public Internet cloud and the private cloud engine

This distinction noticeably widens the kind of software that can be built on the same architecture. The public ICP network suits websites, public services, communities, wallets, SaaS applications and many other systems that have no reason to own separate physical infrastructure. ICP itself describes this as low-cost shared hosting by default, and public application subnets run on a usage-based cycle system.[2]Internet Computer, Cloud engines, Internet Computer Protocol and ICP Quick Start. The public network is a shared ICP environment; a cloud engine is a private subnet configurable by one owner, for which the nodes, operators, geography and replication can be chosen.internetcomputer.org An application running on the public network does not mean that all of its data is automatically readable by everyone: a canister’s internal state is isolated, and the application itself can control which functions are opened to whom.[3]Internet Computer Developer Docs, Canisters and Security model. Canisters hold isolated state and communicate by messages; using a public application subnet does not automatically mean that application state is publicly readable, but on standard application subnets the protocol does not currently guarantee full memory confidentiality from node operators.docs.internetcomputer.org

Not all software fits the same model. Some systems need data to reside in a particular place, nodes run by specific operators, separate resources, or stricter control over the environment in which their state is processed. In other cases the question is simply one of volume: with a large amount of data and many services talking to one another, usage-based accounting can be less natural than compute reserved for that organisation. On a standard public application subnet, sensitive data must also take into account that the protocol does not currently guarantee full memory confidentiality from node operators by default.[3]Internet Computer Developer Docs, Canisters and Security model. Canisters hold isolated state and communicate by messages; using a public application subnet does not automatically mean that application state is publicly readable, but on standard application subnets the protocol does not currently guarantee full memory confidentiality from node operators.docs.internetcomputer.org For needs of that kind, the cloud engine opens the other side of the same protocol.

Under the current technical model of the cloud engine, canisters running on it do not themselves have to hold cycles. Compute, storage, inter-canister messaging and direct HTTPS outcalls inside the cloud engine are priced at zero on the CloudEngine subnet’s cost schedule. That does not mean unlimited or free infrastructure: compute and storage capacity are still limited by the physical capacity of the chosen nodes, and the cloud engine’s own infrastructure and operators must still be paid for. Certain public ICP network services that themselves require cycles also remain separate, and using them from a cloud engine requires an intermediary running on the paid public network.[4]Internet Computer Skills, Cloud Engine Canisters and Deploy to Cloud Engine, updated 21–23 September 2026; Open Cloud, Platform. Under the current technical documentation, canisters on a CloudEngine subnet run on a zero-cycle cost schedule: compute, storage, messages and HTTPS outcalls do not consume the cloud-engine canister’s cycles; paid public-network services are used through an intermediary when needed. The cloud engine’s physical infrastructure and operators are still paid for, and capacity is limited by the power of the chosen nodes.skills.internetcomputer.org

This changes the economic logic of software. In the public Internet cloud, a shared resource is used and paid for according to use. In a private cloud engine, the emphasis is on compute the owner has chosen, and inside the cloud engine every byte of data, internal message or conversation between application components does not have to be metered separately in cycles. This can be especially meaningful for large multi-canister systems, substantial persistent data, and applications in which many software components communicate continuously. A cloud engine can at the same time choose which services to open to the wider network, so “private” does not have to mean a system cut off from the rest of the Internet.[4]Internet Computer Skills, Cloud Engine Canisters and Deploy to Cloud Engine, updated 21–23 September 2026; Open Cloud, Platform. Under the current technical documentation, canisters on a CloudEngine subnet run on a zero-cycle cost schedule: compute, storage, messages and HTTPS outcalls do not consume the cloud-engine canister’s cycles; paid public-network services are used through an intermediary when needed. The cloud engine’s physical infrastructure and operators are still paid for, and capacity is limited by the power of the chosen nodes.skills.internetcomputer.org

One protocol thus contains a fairly wide space of use. The public Internet cloud makes it simple, and with a low barrier, to build global services. A private cloud engine adds a way to move toward larger, more sensitive, more regulated systems, or systems that need a particular kind of infrastructure control. A developer does not necessarily have to change the whole software model when moving between these two worlds. That is an important step if the aim of a cloud protocol is to serve not only public web applications, but a broader part of society’s software.

From one protocol to many sovereign environments

The phrase Internet Computer for Nations has grown out of this. It should not be read as if ICP were a technology created chiefly for states. Companies, universities, developers, communities and individuals can build on the same foundation. The state is rather one of the most demanding tests, because jurisdiction, data residency, cybersecurity, long-term independence and high requirements for public trust all meet there at once. If the same technological foundation can be meaningful at that level, it shows the reach it may also have elsewhere in society.

The most visible national example at present is Pakistan. In February 2026 the Pakistan Digital Authority and DFINITY signed a cooperation agreement centred on ICP infrastructure dedicated to Pakistan, keeping sensitive data in the country, AI-ready public software and a pilot for a national messaging application. The cooperation also includes growing local development capacity in government, education and business.[5]Pakistan Digital Authority, Pakistan Digital Authority and DFINITY Partner for Sovereign Cloud Infrastructure and AI Software Systems, 10 February 2026. The cooperation covers ICP infrastructure dedicated to Pakistan, keeping data in the country, AI-ready software, a pilot for a national messaging application and growing local competence.pda.gov.pk At the same time Pakistan is developing a conventional government cloud, high-performance AI compute, and identity and data-exchange solutions in parallel, which shows clearly that digital sovereignty is not treated as choosing one technology for everything.[6]Pakistan Digital Authority, Working towards a sovereign digital backbone, 22 July 2026. In parallel, Pakistan is developing Government Cloud, national AI compute, directory services and other layers of digital infrastructure.pda.gov.pk

A different example is Swiss Subnet, based in Switzerland, which puts local jurisdiction, operators, data residency and verifiable execution at the centre. Its current direction is aimed above all at regulated, sensitive and mission-critical workloads in the enterprise and public sectors.[7]Swiss Subnet, Sovereign cloud infrastructure for the AI era, Infrastructure and About. Swiss Subnet focuses on verifiable environments running under Swiss jurisdiction for regulated and mission-critical workloads.subnet.ch Pakistan and Switzerland start from very different places, but both point to the same broader question: can environments whose physical and legal control remains local be built on an open global compute protocol?

UNDP’s cooperation with DFINITY gives this development an international dimension. As early as 2024, ICP was used in Cambodia in a prototype of the Universal Trusted Credentials initiative, whose aim was to create trustworthy infrastructure for digital credentials for small and medium-sized enterprises.[8]United Nations Development Programme, UNDP Partners with the DFINITY Foundation to Enhance Financial Inclusion of MSMEs, 3 July 2024. The first ICP-based prototype of the Universal Trusted Credentials initiative focused on digital credentials for small and medium-sized enterprises in Cambodia.undp.org In 2026 the Internet Computer describes its cooperation with UNDP through cloud engines, AI-based software and public-sector pilots across country offices, governments and civil-society organisations.[2]Internet Computer, Cloud engines, Internet Computer Protocol and ICP Quick Start. The public network is a shared ICP environment; a cloud engine is a private subnet configurable by one owner, for which the nodes, operators, geography and replication can be chosen.internetcomputer.org These examples do not yet form one finished global model, but they signal that sovereign computation is increasingly being explored as real institutional infrastructure.

From here a question arises that reaches beyond ICP itself. Digital sovereignty does not have to mean that every state builds itself a closed technological island, just as global infrastructure does not have to mean one central system. The Internet became universal through shared protocols by which very different independent networks can communicate. A similar logic at the level of computation opens the possibility that the same open protocol runs in different jurisdictions, with different operators and in environments with different purposes. Global does not have to mean centralised, and sovereign does not have to mean closed.

Estonia’s next digital leap

Estonia’s starting point differs from that of many other countries, because we do not first have to build a digital state. Digital identity, signing, registries, X-Road and a long history of public e-services already form a strong digital space of trust. X-Road is an especially important example here: its purpose is not to gather all of the state’s data into one centre, but to let independent information systems communicate securely and in a standardised way. Data remains under the control of the service provider, and the choice of an information system’s own architecture does not depend on X-Road.[9]Information System Authority, X-Road data exchange layer. X-Road lets information systems running on different platforms and architectures exchange data securely without centralising the data.ria.ee The strength of Estonia’s digital state has therefore been, from the beginning, the cooperation of many independent parts, rather than the construction of one large central system.

The age of AI places the next question in front of this mature digital state. In 2026 Eesti.ai brings together fifteen high-impact initiatives in education, healthcare, business, the public sector and AI infrastructure. One of them looks directly at Estonia’s need for AI compute and points out that the volume of domestic compute required to develop and operate the next generation of digital services needs to be assessed and developed separately.[10]Eesti.ai, Eesti.ai projects and “Estonia aims to become a world leader in the use of artificial intelligence”, 9 April 2026. The initiative brings together fifteen high-impact projects, among them mapping Estonia’s need for AI compute.eesti.ai This makes digital sovereignty in Estonia a very practical question: which compute it makes sense to own locally, what to use from an international cloud, and in which cases a different architecture is needed.

Still more significant is the direction of creating a digital identity for AI agents. In June 2026 Estonia decided to move ahead with a solution that would, in the future, allow artificial intelligence to act on behalf of a person, a company or an institution within clearly defined rights, in a way that can be checked and audited.[11]Government of the Republic of Estonia, “Michal: Estonia will become the first country in the world to create a digital identity for AI agents”, 17 June 2026.valitsus.ee This is a natural continuation of Estonia’s digital story so far, but it also brings a new architectural question. Digital identity helps prove who a person is, X-Road helps systems communicate securely, and an agent’s identity can determine on whose behalf software acts. The next question is how to make the activity of the software itself trustworthy and checkable.

This also connects with Estonia’s direction in cybersecurity. The national cybersecurity strategy for 2024–2030 aims at a cyber space that is secure, trustworthy and resilient to cyber threats, while the RIA strategy for 2026–2030 emphasises a unified digital architecture, interoperable solutions, the resilience of services and the strengthening of digital independence.[12]Information System Authority, Cybersecurity strategy 2024–2030 and RIA strategy 2026–2030.ria.ee Traditional cybersecurity remains indispensable here: networks and machines must be protected, access managed, software audited and incidents responded to. Verifiable computation does not replace those layers. It can add a different kind of guarantee in places where it is especially important to reduce dependence on the correct behaviour of a single machine, administrator or cloud operator.

Estonia’s opportunity is therefore not to rebuild the existing digital state on top of some new technology. It is much more interesting to ask where a new layer can complement the architecture that already exists. X-Road can continue to connect information systems and exchange data, a conventional cloud can continue to host existing systems, and GPU infrastructure can serve large AI models. A verifiable computation layer can be valuable where what is at stake is digital authority and keys, the critical actions of AI agents, authoritative digital state, software releases, or other processes for which we want, in addition to good security practice, a stronger technical guarantee of what happened in the system.

In that sense Estonia’s next digital leap could move from a secure digital state toward an increasingly verifiable one. This is not a finished national architecture, and it is not a claim that every system should become one. It is rather a research question that Estonia is in an unusually good position to ask, because we already have a strong foundation and a digital state mature enough to focus not only on digitisation, but on the next layer of the trust model.

subnet.ee as an Estonian pilot

EFTI has created the subnet.ee concept in order to investigate this possibility in practice. At present it is an initial research and pilot direction, not finished national infrastructure. The idea is to explore what an ICP cloud engine located in Estonia and operated by local parties could look like, and what it would actually make sense to build on such an environment. The concept joins local control over physical infrastructure, the ICP cloud protocol and verifiable full-stack software into one whole.[13]EFTI, subnet.ee — a pilot for Estonia’s sovereign smart cloud. An initial concept for practically investigating local infrastructure, an ICP cloud engine and verifiable software in Estonia.subnet.ee

Its difference from the usual idea of “our own cloud” is important. Placing servers in Estonia gives control over their physical location and operator, but it does not yet change the trust model of the application itself. On a cloud engine, the ICP protocol runs on the chosen infrastructure and binds separate machines into one replicated compute environment. Applications can use the same canister model as on the public Internet Computer, while the location of the infrastructure, the operators and the replication can be shaped to the needs of that particular environment.

The economic model of the cloud engine makes the subnet.ee possibility especially interesting from the application side as well. If, under the current technical model, storage and inter-canister communication inside the cloud engine are not metered in cycles, systems can be built on the same infrastructure whose volume of data and internal traffic would have a different cost structure on the public usage-based network.[4]Internet Computer Skills, Cloud Engine Canisters and Deploy to Cloud Engine, updated 21–23 September 2026; Open Cloud, Platform. Under the current technical documentation, canisters on a CloudEngine subnet run on a zero-cycle cost schedule: compute, storage, messages and HTTPS outcalls do not consume the cloud-engine canister’s cycles; paid public-network services are used through an intermediary when needed. The cloud engine’s physical infrastructure and operators are still paid for, and capacity is limited by the power of the chosen nodes.skills.internetcomputer.org This is not only a question of cost. One’s own cloud engine also makes it possible to choose the physical location and operators of the nodes, and to decide which services are connected to the public network. The same cloud protocol thus extends to software for which a shared public network may not be the natural choice because of volume, jurisdiction, the sensitivity of the data, or the organisation’s own need for control.

The aim of subnet.ee is not to build Estonia a separate Internet or to compete with the existing government cloud. Its value would rather be as a laboratory in which to study how a global open protocol and local digital sovereignty can complement each other. Real applications could be used there to try integration with X-Road and existing information systems, to explore possible operator models, and to assess in which use cases verifiable execution gives a real advantage. The point of the pilot is not to prove in advance that the answer must be ICP, but to build enough that the technological possibility can be judged from a working system, not only from an architecture diagram.

This makes subnet.ee one possible bridge between the global scale and the Estonian one. Pakistan shows how sovereign ICP infrastructure can be part of building a large country’s digital infrastructure. Switzerland ties a similar direction to jurisdiction and institutional use. Estonia can explore how verifiable computation fits the existing layers of an already very mature digital state. It is precisely the different local solutions that eventually help show which part of the new model is universal and which must remain for each society to shape.

Local implementation: Tark Türi

If the view stayed at the national level, an important part of the idea of society’s digital infrastructure would still be missing. The value of technology finally reaches a person through a school, a company, an organisation, a municipality and a community. AI makes this level especially interesting, because the technical barrier to creating software is falling quickly. Building a good system still requires understanding the problem, architecture and responsibility, but there is steadily less reason why every local need should wait for a large platform, a public procurement or the solution of an international technology company.

One of EFTI’s directions at the local level is Tark Türi, a digital space of cooperation for a local community, joining people, knowledge, discussions and initiatives. Its starting point is the town and municipality of Türi, but the principle itself is not limited to Türi. Every community could have the possibility of shaping a digital space in which local knowledge does not disappear into a social-media feed, but accumulates, becomes connected and gradually turns into a basis for acting together. It is an attempt to ask, at the local level, the same question that digital sovereignty asks at the level of the state: how much of our own digital space could be shaped by us and based on our own needs?

Tark Türi is not, in this sense, simply a website or a new forum. Discussion, documents, photographs, earlier decisions, proposals and the necessary background can gather around one topic, and a public way of reading also leaves room for those who do not themselves take an active part in the conversation. Around one park, a school route or a local development question, what remains is then not only a momentary conversation, but over time a community memory. One person may know the local history, another may notice an everyday problem, a third may be able to propose a solution and a fourth may know how to carry it out. The task of the software is to help that knowledge meet.

A possible role for artificial intelligence is to support what EFTI calls ühisaru, or collective digital intelligence. AI can help summarise long discussions, find related questions, bring back earlier decisions and make a large amount of local information usable for a person. It should not decide for the community or turn disagreements into an artificial consensus. The idea runs the other way: technology helps people notice, remember and make sense of their own collective knowledge, leaving choices and responsibility with the people themselves.

This is also where we arrive at the idea of conscious software. It does not mean consciousness in the software itself, but a conscious attitude toward the relationships and behaviour that technology supports. Where many of today’s platforms are optimised to increase attention and time spent, the aim of local public software can be to grow understanding, to make important information findable and to connect digital communication with real cooperation. Türi, as a spring capital, a garden town and a living environment close to nature, gives this a concrete local face. A nature-centred approach does not have to remain a matter of design. It can mean the principle that technology fits the person and the living environment, rather than the person fitting the technology. From that point of view, Tark Türi can be seen as the digital layer of a nature-centred centre for innovation.

At the same time Tark Türi offers a more general model that can be adapted elsewhere. The digital space of another town, municipality, school network or community would have a different identity and different functions, but the principle could remain the same: local knowledge, public memory and cooperation do not have to depend entirely on an external commercial platform. Such a system is also a good example of software for which the cloud-engine model may in the future prove natural: a large body of persistent community knowledge, software components that communicate with one another, and a possibly more sensitive data layer can run in one locally controlled environment, while the public part remains open to people. Which infrastructure is specifically best must still emerge from the application’s real needs and from practical experiments.

The capacity to build for ourselves

With society’s digital infrastructure it is easy to focus on servers, networks and protocols and to leave aside the most important resource: the people who know how to create something on top of them. Digital sovereignty remains incomplete if we have the infrastructure, while the understanding of it and the ability to build software are entirely somewhere else. AI changes this situation quickly, because part of the work that until recently required a large development team can now be done with far fewer resources. At the same time, the need to think in systems does not disappear. As creating code becomes easier, the ability to understand the problem, shape the whole, choose a suitable architecture and decide what is worth building at all becomes still more important.

Practical education therefore belongs naturally to EFTI’s software direction. The aim is not to teach as many programming languages as possible, or to tie learners to a complicated stack of cloud services, but to give the ability to move from a real problem to a working full-stack application, using AI and as complete a foundation as possible. That can mean professional software education, cooperation with schools and young people, or building concrete local projects together with their future users. A society’s digital autonomy grows not only when the state has better infrastructure, but when more people and organisations gain the ability to understand, adapt and create technology themselves.

The local level gives this its most direct meaning. In a school, software education can move from using finished applications toward creating them. A municipality can be a place to try new solutions. An organisation can solve a small but real problem with software whose construction would previously have been disproportionately expensive. AI does not reduce the person’s role here. It can give a smaller actor access to a capability that previously belonged only to large organisations. When that capability is joined to a conscious architecture and suitable infrastructure, creating software itself becomes one layer of societal capacity.

EFTI’s software direction

This is why society’s digital infrastructure is not, for EFTI, only one research topic. This source also defines the institute’s software direction: to investigate, discuss, teach and selectively build. At the global level that means following the development of ICP and of cloud architecture more broadly, and creating technical and architectural clarity. At the national level it means setting those possibilities alongside Estonia’s digital state, X-Road, digital identity, cybersecurity and the development of AI, and investigating concrete pilots such as subnet.ee. At the local level it means creating software together with schools, municipalities, companies, organisations and communities, from whose real needs these systems grow.

EFTI’s role does not have to be to operate every layer, or to announce a finished architecture of the future. The institute’s value can lie in connecting the different scales: bringing global development into the Estonian context, joining technical knowledge to a societal need, growing local skills, and building enough in chosen places that good ideas can be tested in the real world. subnet.ee is one possible research and pilot direction at the national level, and Tark Türi one direction of the local application layer. Both are at present open to development, not finally finished answers.

The same applies to cooperation. A field like this cannot mature inside one institute, because infrastructure, cybersecurity, public administration, education, software development, entrepreneurship and knowledge of local life all meet in it. EFTI therefore wants to be open to technical, institutional, educational and local partners who are interested not only in watching the next generation of digital infrastructure from the side, but in investigating and trying it together. One partner may bring infrastructure, another a problem, a third knowledge, and a fourth the people for whom the solution is actually being built.

One system, three scales

To understand the larger picture, these levels have to be seen together. Globally, a possibility is emerging to run full-stack software on an open compute protocol, and to use the same protocol both as a public Internet cloud and, through cloud engines, on infrastructure chosen by a particular state, company or organisation. One software model thus extends from a public web service to an enterprise system with a large volume of data, and from there to sovereign national infrastructure. At the national level this raises the question of how to preserve control, openness and interoperability in a world where software and AI are becoming steadily more critical. In Estonia that question can be asked from an especially high starting point, because we already have a mature digital space of trust, whose next layer does not have to replace what exists, but can complement it.

At the local level, large architecture finally becomes real life. A digital space like Tark Türi can help a community remember, make sense and act, while AI-based software development gives more and more people the ability to create such systems themselves. A local experiment can give knowledge to the national level, and one country’s experience can in turn help shape global development. The movement does not run only from the top downward. Working digital infrastructure can develop at the same time from the protocol toward society and from a local need toward a larger system.

Estonia’s first digital leap connected people, the state and data. The next one can extend the same space of trust to the activity of software itself, and at the same time give more people the possibility of creating within that space. ICP matters in this story not simply as one new technology, but as currently one of the more complete attempts to join a public Internet compute infrastructure, private sovereign cloud engines and verifiable full-stack software inside one protocol. Estonia’s possibility is to explore what this means for a mature digital state. Türi’s possibility is to show what a large architecture can mean at the level of a real community. EFTI’s role is to help build the bridge between them.

Society’s digital infrastructure is not only what our systems run on. It is the foundation on which people, communities and states can increasingly build their own digital future.