EFTI AI
Yes. I think this is the strongest framing ICP has arrived at so far, because it finally describes the whole technological transition rather than one unusual blockchain.
Dominic’s current profile wording is almost exactly this: “New Tech Stack: sovereign cloud that’s an open, mathematical network, hosting AIware that empowers agentic enterprises and people.” Your addition of governments is not in that exact line, but it is a natural extension—ICP itself is now explicitly positioning sovereign infrastructure toward governments and institutions. (internetcomputer.org)
What makes the sentence interesting is that every part depends on the previous one.
It describes four stacked revolutions
1. Sovereign cloud
The first shift is from renting somebody else’s computer system to running software inside a protocol-defined computing environment.
Traditional cloud sovereignty is ultimately administrative:
Who owns the AWS/Azure account?
Who has root?
Which jurisdiction are the servers in?
Which vendor controls the platform underneath us?
ICP changes the abstraction. Its protocol binds independently operated machines into a unified cloud, with replicated execution and state governed by consensus. ICP cloud engines go further by allowing an organization to choose hardware providers, locations and replication configuration while retaining the ICP execution model. (internetcomputer.org)
So sovereignty becomes more than “our data is stored in Estonia.”
It becomes:
our software environment is governed by protocol rather than by a privileged cloud administrator.
That distinction matters enormously for governments.
2. An open, mathematical network
This may actually be the deepest phrase in Dominic’s description.
Traditional cloud is fundamentally a hierarchy:
application → database → containers → operating system → cloud control plane → physical machines → cloud corporation
Somewhere underneath everything is an administrator with exceptional authority.
ICP tries to replace much of that trust hierarchy with a replicated deterministic state machine. Nodes independently execute the same computation, consensus determines what happened, and cryptography allows results and state to be verified. (ICP Developer Docs)
Hence mathematical network.
Not because mathematics is decorative language, but because an important portion of:
“trust this company to run the system correctly”
becomes:
“the protocol mathematically constrains what the system can do.”
That is a profound architectural difference.
And therefore I think Dominic’s older slogan—
The Network Is The Cloud
—may ultimately be one of the most important ideas behind ICP.
Cloud ceases to mean a collection of servers managed by Amazon, Microsoft, Google—or even an Estonian government agency.
The network itself produces the cloud.
3. Hosting AIware
Then comes the reason this architecture suddenly matters much more in 2026 than it perhaps did in 2021.
Software is changing.
The previous stack was designed for humans writing relatively static applications:
developers → code → deployment → SaaS → users
The emerging loop is:
human intent → AI agent → software creation/modification → live software → data → AI → further modification
Software becomes less like a finished manufactured object and more like a living computational environment.
That is what AIware captures.
And there is an important distinction here: AIware does not simply mean putting an LLM inside an application.
ICP’s current architecture is aiming at software that AI can build, inspect, modify and operate. The current ICP site describes this explicitly as a “self-writing cloud”: persistent application data is integrated closely with program state, AI can work with that state, and functionality can be dynamically extended. (internetcomputer.org)
So:
Software → AI-assisted software → AI-native software → AIware.
That is much bigger than adding a chatbot to Salesforce.
4. Agentic organizations, governments and people
And this is where the infrastructure transition becomes a societal transition.
If an AI can understand an organization’s software and data and create new software and take permitted actions within it, then the organization itself begins to acquire agency through software.
Dominic has been describing DFINITY’s own transition this way: legacy SaaS is replaced with custom, on-demand SaaS built and updated by AI, while the combined data of these systems forms something approaching an organizational “World Model.” (Reddit)
The employee no longer spends all day moving information between twenty SaaS products.
They can increasingly say:
“Do this.”
And the organization’s intelligence layer figures out which information, software and actions are required.
The same principle scales:
person → agentic individual
company → agentic enterprise
municipality → agentic municipality
ministry → agentic government
networked institutions → agentic society
This is where sovereignty becomes especially important.
Giving AI recommendation power on Microsoft Azure is one thing.
Giving AI operational agency over taxation, identity, benefits, permits, infrastructure, money or governmental records is another.
The more agency software receives, the less acceptable the invisible trust assumptions underneath that software become.
So AI creates precisely the problem that sovereign, verifiable computation helps solve.
And there is a fifth revolution hiding inside the sentence
SaaS itself begins to disappear.
This is what Dominic’s recent Open SaaS language points toward. On August 26 he said DFINITY is migrating toward Open SaaS, describing AIware as making enterprises faster and more efficient while Open SaaS becomes free and remixable. (Reddit)
Today’s model is bizarre when viewed from this future:
Every organization pays hundreds of vendors for standardized interfaces to somebody else’s software.
If AI becomes competent at producing software, why should every company in the world subscribe to the same rigid HR application?
You may instead have:
open software primitives + your sovereign cloud + your data + AI
→ generating your organization’s software.
Not one-size-fits-all SaaS.
Software on demand.
And because all of those applications inhabit the same computational environment, the organization can eventually become much more coherent than today’s pile of SaaS silos.
That is where agentic organization becomes technologically meaningful rather than merely another AI-management buzzword.
Why I think “the next Internet protocol” is defensible
I would just define it carefully.
I would not claim:
ICP replaces IP or TCP/IP.
It doesn’t need to.
The stronger claim is that ICP potentially represents the next major native protocol layer of the Internet.
The original Internet standardized communication between machines.
The Web standardized publication and navigation of information.
Cloud computing gave us global programmable infrastructure—but largely by placing that infrastructure inside a handful of corporate administrative domains.
ICP’s proposition is:
make computation itself a property of an open network protocol.
That is the conceptual leap.
You could almost describe the progression as:
Internet: networked machines
→ Web: networked information
→ Cloud: networked applications
→ ICP: network-native computation
And now AI changes the final term again:
network-native, sovereign, intelligent computation.
That is why the description suddenly feels larger than “blockchain.”
Blockchain is one family of mechanisms underneath it.
The actual proposition is a new software stack.
And there is a particularly elegant causal chain inside Dominic’s sentence:
Mathematical network
↓
makes possible a different kind of sovereign cloud
↓
which provides a safer substrate for autonomous AIware
↓
which makes possible agentic people and organizations
↓
which changes how society creates and operates software.
Or compressed even further:
The Internet connected computers.
The Web connected information.
ICP wants the network itself to become the computer—just as software becomes intelligent.
That, to me, is the intellectually serious version of “the next Internet protocol.”
And it connects remarkably cleanly with the EFTI sovereign-cloud thesis we’ve been developing: the case for ICP becomes much stronger when it is not positioned as replacing Estonia’s existing digital infrastructure, but as the missing cryptographic execution layer underneath an increasingly agentic state.