For almost three decades, the global internet was an unusual technological artefact. Most large pieces of human infrastructure had borders, jurisdictions and tariffs. The web mostly did not. A packet from a São Paulo café to a Tokyo server might cross eight networks and three continents without anyone really thinking about it, and the price of that journey was somewhere around the cost of nothing.
That era is closing. Not loudly, not in a single moment, but with the patient certainty of a tectonic plate. The most significant story of 2026 is not who launched what model or which platform changed its algorithm. It is the quiet rebuilding of the underlying internet along regional lines.
What is actually changing under the hood
Three things are happening at once, and together they are reshaping the physical and legal map of the network.
The first is regional cloud sovereignty. The European Union finalised the implementation guidelines of the EU Data Act and the AI Act in late 2025, and the consequence is that critical data for healthcare, finance, public administration and large AI workloads now needs to live inside European-owned and operated infrastructure. The American hyperscalers have responded by creating fully separate European arms, in many cases with European board oversight and European employee-only access. The split is not always visible to the user, but it is real. A French citizen's medical record in 2026 is, by law, on different metal than a Canadian one.
The second is subsea cable diversification. For decades, almost half of intercontinental traffic ran over a handful of routes through the Red Sea, the South China Sea and the North Atlantic. After multiple cable cuts in 2024 and 2025, both accidental and otherwise, governments and operators have accelerated the construction of alternative routes. Saudi Arabia, India and Australia are co-financing a new south-Indian-Ocean cable corridor. Brazil and the EU completed the Ellalink upgrade. Japan and the Philippines launched a new Pacific route that bypasses the most contested chokepoints. The result is more capacity, but also more fragmentation.
The third is national or regional AI infrastructure. France, the UAE, Saudi Arabia, India, Brazil and a handful of others have committed at least eight billion dollars each, in some cases far more, to building sovereign compute. The cluster matters because if a country relies on someone else's GPUs to train its models, it relies on someone else's foreign policy too. This logic, once dismissed as paranoid, is now mainstream in finance ministries.
Why this is happening now
The trigger was not a single event. It was the cumulative recognition, across very different governments, that the open internet had become a strategic dependency. When something becomes strategic, governments stop being passive about it.
For Europe, the trigger was a mix of GDPR enforcement difficulties and the AI Act. For the United States, the trigger was the AI export control regime, which made compute and frontier models effectively national assets. For China, the trigger was the long-running Great Firewall logic combined with a new push for domestically-fabricated chips. For BRICS+, the trigger was the freezing of Russian central bank reserves in 2022, which woke up many non-aligned countries to the realisation that financial infrastructure could be turned off. Network infrastructure, by extension, could be too.
What is interesting is that none of these governments needed to coordinate. They reached similar conclusions independently because they were responding to the same underlying shift. Digital infrastructure stopped being neutral plumbing and started being a layer where power is contested.
What this looks like for users right now
For most people, day to day, the change is barely visible. You open the same apps. You watch the same videos. The latency you experience is, if anything, slightly better than five years ago, because the regional routing is more optimised.
The first place users feel the shift is in what is available where. Streaming catalogues, which have always been regional, are now even more sharply differentiated, because content licensing is increasingly bundled with data residency obligations. AI products differ more across regions too. The version of an assistant available in the EU may decline certain prompts that the US version accepts, and vice versa. A model deployed in the UAE may have different cultural defaults than the same model deployed in the UK.
The second place users feel it is in payments. Cross-border money is being routed through more regional rails. The Indian UPI, the Brazilian Pix, the EU's Wero and the African PAPSS systems all interoperate increasingly with each other and decreasingly with the legacy US-dominated card networks. The international transfer that took four days and cost forty dollars in 2018 now often takes thirty seconds and costs cents, but it travels through systems that the previous decade did not have.
The third, more subtle, is privacy. The amount of personal data that physically leaves your region is dropping. This is, on balance, probably good, but it also means the historical model of one global database for each major platform is being pulled apart into regional shards. Inconsistencies follow, and identity verification across borders is sometimes more painful than it used to be.
What this means for business
For any company operating across multiple regions, compliance has stopped being a back-office concern and become a core architectural decision. You either build region-aware systems from day one or you accept that scaling internationally will be more expensive and slower than it used to be. The companies that prepared early are quietly winning. The ones that did not are finding that the cost of multi-region rearchitecture is now the largest single line on their roadmap.
For startups, the implication is a little different. The default of launching a single global product from San Francisco has weakened. There is a new generation of founders building region-first companies that are deliberately optimised for their home market and only later expanding. Many of them never get global, and many of them do not need to. The unit economics of a region-only model are increasingly good, because regional infrastructure costs less to use than truly global infrastructure.
The next five years
The trajectory is clear in direction even if the speed is uncertain. The internet of 2030 will look more like four or five overlapping regional internets, with thinner but still real connections between them, than like one global commons. Most of the original promises of the open web survive inside each region. The cross-region experience will be more friction, more compliance, more cost.
The optimistic reading is that regional sovereignty makes the network more resilient. A single point of failure is harder to find when there are five physically and politically distinct stacks. The pessimistic reading is that this is the digital version of nineteenth-century imperial blocs, and that the cost of fragmentation will be paid by everyone who cannot afford to operate in multiple of them.
Both readings are partly correct. The system that emerges will not be the global internet of the 2010s, and it will not be a return to isolated national networks either. It will be something new. The work of the next five years is figuring out what to call it.
