The Map Has Already Changed: A Diagnosis of the Splinternet Era
On February 24, 2022, Russian tanks crossed into Ukraine, and the world watched three different wars begin.
On the Open Internet, users in Kyiv and London and Washington opened their phones to a war of imperial reconquest: Russian tanks converging on Kyiv and civilians filming rocket attacks from balconies. Elon Musk activated Starlink over Ukraine that weekend, enabling the country to stay online; Meta made an exception to its hate-speech rules for the phrase ‘death to Russian invaders’; TikTok feeds filled with vertical video of combat that Kremlin-linked accounts were racing to counteract.
On the Russian internet, users on VK saw a ‘special military operation’ against a US-installed regime, a defense of Russian-speaking populations, a delayed but righteous response to NATO's eastward creep. Foreign platforms went dark inside the country within a month. Journalists calling it a ‘war’ faced fifteen-year prison sentences.
On the Chinese internet, users on WeChat and Weibo saw a story about US and NATO provocation, a nuanced Chinese neutrality that leaned Russian, and a familiar frame about American hegemony finally meeting a peer. The word ‘invasion’ was rare in state coverage; the phrase ‘Ukraine crisis’ was preferred.
Three wars. Three information environments. Three sets of facts about the same tanks.
Bespoke Realities
One event, presented differently in various information environments and interpreted as reality in each of them. We’ve come to accept this as a part of what the internet has done to our world. We’ve come to accept that other people have “alternate facts,” not in the sense that multiple versions of reality may be true but we have come to accept that they believe their version of reality is the right one. Vote based on it. Raise their children based on it. Participate in local government based on it. Why?
Renee DiResta uses the phrase ‘bespoke realities’ to describe this 21st century phenomenon, when someone can have so much access to information and so little ability to discern what is actually true that they can inhabit their own version of reality. Their bespoke reality can be reinforced by the sources and voices they deem reliable, and they can engage with a detailed narration and interpretation of what is going on in the world from that bespoke perspective based on the various media and platform feeds they have cultivated, knowingly or unknowingly, over the years. This isn’t all the fault of the individual, as DiResta shows. The platforms we give our attention to optimize what they put in front of us in order to keep our attention, not to inform or educate us. But regardless of where the blame for the outcome lies, the result is a society in which individuals can construct their own fact-set, their own information environment, their own community of believers, and never meaningfully have to encounter anyone outside it.
The dynamic itself is not necessarily new. What is new is the scale, and the lock-in the platforms have created. And the challenge it presents is real: any pushback by the platforms, the government, the media, even friends and family, reads as censorship, judgement, or conspiracy. And the phenomenon is bipartisan, operating across the political spectrum in democracies around the world. It is also, importantly, not itself the root of the problem. It is an outcome of how we have built our Western-driven portion of the digital world. Bespoke realities are an outcome of how it is designed and governed, and the vulnerabilities that come along with that.
I say “our portion of the internet” because, as the stories above show, other parts of the world don’t experience the internet the way we do in America, or Western Europe, or Australia, Korea, Japan, or any other country that let’s its people access what we can think of as the Open Internet. Polarization, mis- and disinformation, and conspiracy theories are not unique to the bespoke realities of the Open Internet, but we build and experience our bespoke realities differently than someone may in China, or Russia, or Iran, or North Korea. This, the breakup of what was supposed to be a global commons built on information sharing, individual expression, and unfettered access, into different spheres composed of different platforms, different regulatory regimes, and different infrastructure stacks and supply chains, is the Splinternet.
The conflict, confusion, and conspiracy we observe others participating in here in the United States (but are always so sure we ourselves are avoiding) is a symptom of the Splinternet as seen from inside a single open society. And many of the prescriptions that have been proposed to deal with it (platform reform, content moderation, civic-tech intervention, community-driven correction along the lines of Community Notes on X) are designed to work on the information layer. But the bespoke realities Americans observe inside their own country are just the most visible surface of a structural fragmentation operating, simultaneously, at three other layers of the internet. And those other layers are where the real battle is being, and needs to be, fought.
Because the advent of AI is turning this into an arms race. While bespoke realities were impacting real world events at scale and in a disruptive way by at least the 2016 United States presidential election, they have so far been tolerated as “the discourse,” with calls for civility, fact-checking, and the prescriptions mentioned above. But as AI looks to be able to supercharge economic and military output and efficiency, the levels beneath the information layer are now the site of geopolitical competition.
Countries that want to win the AI race, or at least ensure their sovereignty as it unfolds, are investing in compute and physical infrastructure at a scale previously unseen, both domestically and through international partnerships. China, the EU, and even the Vatican are staking a claim at the governance level, attempting to set or influence new rules across their versions of the internet. The platforms, from the social platforms to the cloud providers to the AI labs, are trying to position themselves to lock-in as many users as possible. And up at the information layer, where many of us spend our time, maybe the proliferation of AI-slop will drive the average user away from the feeds we frequent today, but it’s unlikely to make the internet a more fun, more productive, more enjoyable place to spend time. More likely, the AI companies will further gift wrap our own increasingly bespoke realities in a new generation of apps: banal screens, this time with no comments or other users, and fewer linked and labeled sources, pulling us further into the silo of the digital world.
The Stack
The internet is not a single interface that people open up on a computer or a mobile device, as much as our collective screen time may portray it that way. The internet is a layered stack, a structure that enables the information we see in our feeds and inboxes and screens to be made and shared and reported on. Benedict Anderson argued 40 years ago in Imagined Communities that information structures build political community, so the gathering of believers into their chosen bespoke realities, and the political re-sorting underway because of it, should not take us by surprise. But the dynamic is now clear enough and, in the wake of smart phones and social media, vast enough, that organic effects, bad actors, and competitive nation states are contesting all layers of the stack.
Layer 1: Compute and Physical Infrastructure
What it is: The physical substrate that stores, moves, and processes information, including chips and fabs, data centers, subsea and terrestrial cables, satellite constellations, cellular towers, power grids, and the compute clusters those grids feed.
Diagnostic: If it has a location, a capex line, and a physical footprint, it lives at Layer 1.
Why the Splinternet is contested here:The physical layer defines what is possible at every layer above it. Which country could roll out 3G in the 2000s determined which populations could join the mobile-internet era; which country can build frontier compute now determines who can host the next generation of AI platforms. Rerouting a subsea cable changes latency and cost across whole economies. A data center built under one jurisdiction is subject to that jurisdiction's compulsion powers, including the compulsion to seize its contents or shut it off. Physical infrastructure is where sovereignty is most literal, and where the choice to build inside or outside a sphere is hardest to reverse once made.
What's happening: The physical layer has been contested for two decades. The 3G and 4G rollouts (2001–2015) determined which economies could support mobile-video-scale platforms and where they could run at all. Huawei's dominance of 4G and 5G equipment triggered the first infrastructure-layer geopolitical crisis, with the US, UK, Australia, and Japan coordinating restrictions from 2018 onward. China’s Belt and Road Initiative, announced in 2013, exported Chinese-built infrastructure across Central Asia, Africa, and Southeast Asia over the course of a decade. Subsea cable buildouts put Google, Meta, and Amazon in the position of owning private trans-oceanic connectivity by the late 2010s. The AI supercycle has now accelerated all of this. In the United States, the 2022 CHIPS Act redirected about $52B in capital to domestic and allied fabs (TSMC Arizona, Intel Ohio, Samsung Texas) under an export-control regime that has tightened quarterly since 2019. As of 2026, the hyperscalers continue to fund the largest economic build out in American history. China's June 2026 $295B state-directed AI data-center plan, paired with SMIC's 5nm pilot production and Huawei Ascend chips replacing Nvidia domestically, shows how China is building the new infrastructure needed to control its own sphere in the future. The EU launch Global Gateway in 2021, allocating €300B to counter the Belt and Road Initiative and its AI Factories program funds sovereign compute across seven member states. In the Gulf, the UAE’s G42 and Saudi Arabia's HUMAIN have committed roughly $100B combined to sovereign AI compute. And Starlink, activated over Ukraine on day one of the invasion, is now treated as strategic infrastructure rather than a commercial product. The physical substrate is being rebuilt on lines that no longer assume a single global commons.
Layer 2: Platforms
What it is: The software and service layer running on infrastructure that shapes what users can do.
Late 1990s: Search engines and web portals
Post-2005: social platforms after 2005 (Facebook, YouTube, Twitter, later WhatsApp, WeChat, TikTok); commerce (Amazon, Alibaba, Shopee, Flipkart); payments (Stripe, Alipay, PIX, UPI, M-Pesa); messaging (Telegram, LINE, KakaoTalk); cloud (AWS, Azure, Alibaba Cloud)Since 2020: AI platforms (OpenAI, Anthropic, Baidu ERNIE, Mistral)
Diagnostic: If users have accounts on it, it has terms of service, and its design shapes how they communicate, are entertained, or make money, it's a platform.
Why the Splinternet is contested here: Platforms are where the internet is actually experienced. The design choices platforms make (engagement dynamics, advertising models, moderation regimes, virality mechanics) determine what users see, what they can do, and how they can organize. Those design choices have reshaped political systems for two decades. The 2010–2012 Arab Spring wave was inseparable from Twitter and Facebook mobilization mechanics, and every major democratic disruption since has had a platform-layer story. Access to a platform is now a sovereignty decision, and the design of a platform is a policy decision by fiat.
What's happening: The platform layer has been splintered across spheres since the smartphone era began. Google left China in 2010, Facebook and Twitter were blocked shortly after, and the Great Firewall's platform blocklist expanded through the 2010s as WeChat, Weibo, and Alibaba built the sovereign Chinese super app economy. YouTube's recommendation engine instilled political opinions in Brazil and elsewhere in ways linear media never did. India blocked WeChat and 200+ Chinese apps in 2020, one of the first large-scale platform-layer sovereignty moves outside the US-China relationship. Since then the pace has accelerated:
The US forced TikTok's divestment in 2024–25
France arrested the founder of Telegram in 2024
Japan's data-localization rules forced LINE-Yahoo to wall Japanese user data off from its Korean parent
Some spheres now run sovereign super apps like WeChat or VK, while others accept segmented incumbents like TikTok. The platforms that twenty years ago were the visible face of one internet are now sphere-blocked, sphere-segmented, or sphere-built.
Layer 3: Governance
What it is: The rules constraining infrastructure and platform behavior: national laws, regulatory bodies, standards bodies, multilateral agreements, procurement criteria, sanctions, and export controls.
Diagnostic: If it is a rule, standard, or binding institutional decision that a state or supra-state body enforces, it's governance.
Why the Splinternet is contested here: Governance most directly encodes sovereignty, and so far it has been the layer that catches up last. Regulators write rules for platforms after platforms have already reshaped societies, like CAN-SPAM after email spam had already colonized inboxes, or GDPR after two decades of unrestrained data collection, or the DSA after platform dominance had already restructured news distribution. Governance is the layer where sphere boundaries are drawn most explicitly: a law says "this applies in this jurisdiction," and the jurisdiction becomes an edge of one sphere or another.
What's happening: The governance layer has been drawing sphere boundaries for over a decade. Post-Snowden, the invalidation of Safe Harbor (2015) demonstrated that even allied spheres could not maintain unified rules on transatlantic data flows. Russia's data-localization law (2015) required foreign platforms to store Russian citizens' data on Russian soil. India's IT Rules (2011, expanded 2021) built the framework for intermediary-liability control there. China's Cybersecurity Law (2017) formalized state control over data and content, codifying what the Great Firewall regime had been doing since the early 2000s. The EU's GDPR (2018) established the Brussels effect as the operating model for open-sphere governance, where the strictest jurisdiction sets the global compliance floor.
The current AI capex cycle has now accelerated the governance layer. The EU's DSA, DMA, and AI Act constitute the most comprehensive internet-governance regime in existence, extended in June 2026 by the proposed Cloud and AI Development Act, which turns cloud and AI sovereignty into a procurement test. On the Open Internet, the US governs the sphere's edges (export controls, EO 14105 outbound investment screening (2023), EO 14117 data-broker restrictions (2024), the 2026 GAAIA proposal), while Europe focuses more on its interior. In China, the July 2026 outbound investment controls redefine what capital and talent can leave. In Japan, FSA convened a 36-entity AI-security working group in June 2026. The Vatican issued Magnifica Humanitas on AI governance around the same time.
In the American and European models, regulation is reactive since it follows developments at the platform layer. In the Chinese model, the order is inverted: algorithm registration, sovereign-internet law, and outbound controls shape what platforms can be in the first place. The stack order is itself sphere-specific, which contributes to structurally different internets.
Layer 4: Information and Content
What it is: The content produced, curated, and consumed across the stack, from news to memes, search results to algorithmic recommendations. It also includes the ecosystem producing all this content: journalists, state media, publishers, influencers, creators, and now generative AI bots.
Diagnostic: If it is something a user sees, reads, or watches, and it is produced by authorship (human or machine) rather than being a platform feature itself, it's content.
Why the Splinternet is contested here: This is where the sphere's outputs meet the user, and where bespoke realities live. Contesting the information layer is as old as propaganda, but the free creation, distribution and consumption of the material for all users across the entire sphere, is a 21st century phenomenon. But this layer is dependent on the three beneath it: you cannot produce or consume content that infrastructure, platforms, and governance do not permit or enable. That makes the content layer the most visible fragmentation, the piece of the Splinternet you notice first because you live it everyday. But what happens at the layers beneath is often more structurally consequential.
What's happening: The information layer has been contested for as long as there have been platforms. The Iranian Green Movement in 2009 was the first major test of platform-mediated political mobilization. Russian interference in the 2016 US election marked the moment American open-sphere leadership recognized that its information layer was operationally exposed. The Cambridge Analytica scandal (2018) revealed the platform-content interaction as its own strategic terrain. WhatsApp networks organized a Brazilian presidential victory in 2018; Facebook's role in the Rohingya crisis in Myanmar (2017) established that platform design could enable mass violence. Iranian internet shutdowns during the 2009 Green Movement, the 2019 fuel protests, and the 2022 Mahsa Amini protests demonstrated the kill switch as content-layer policy.
In the AI supercycle, the fragmentation has become denser and more visible. Inside the Open sphere, DiResta's bespoke realities have made shared facts optional. Chinese state media (CGTN, Xinhua, Global Times) has expanded strategically across Africa, Latin America, and Southeast Asia. Russian information operations, banned in the EU since 2022, have evolved into more sophisticated networks that continue operating across the Global South. The 2024 Romanian presidential election was annulled after evidence of foreign information operations. Generative-AI content now saturates every open platform. The content layer is downstream of the other three since its shape is set by what infrastructure, platforms, and governance permit.
The internet has been fragmenting for years. The fragmentation is no longer episodic, regional, or speculative. It is structural, multi-layered, and accelerating.
The Asymmetry
Every sphere pays a cost for its information posture.
Open societies pay a manipulation cost. The same openness that produces innovation, freedom of speech, cultural assimilation, and coalition formation also produces a structural vulnerability: the potential for bespoke realities. A society in which any opinion can be shared, any voice can be amplified, any community can be assembled, and any narrative can find an audience cannot also close itself to the foreign actors willing to seed and accelerate fragmentation amongst it’s population. This is the cost open societies have spent thirty years pretending they do not pay. The 2024 Romanian presidential election annulment, the manipulation of the American information environment since 2015, and a steady flow of smaller cases across European democracies should be enough for us to stop pretending.
Closed societies pay a different cost: internal blindness. A regime that suppresses information about the outside world also degrades its ability to see what is happening inside its own borders. The kill switch buys insulation from external manipulation at the cost of seeing reality. Russian military intelligence misread the willingness of Ukrainians to resist invasion in February 2022 in part because the information sphere inside the Russian state did not surface contradicting evidence. Chinese authorities failed to process what was happening in Wuhan in the critical first weeks of January 2020 in part because the information sphere did not deliver the signal upward. The kill switch and the Great Firewall are real capabilities, but they come with a significant cost.
And then there is the legitimacy cost: the erosion of trust when the story a system tells about itself diverges from the ground truth. Open spheres pay it continuously, at the ballot box. Closed spheres pay it episodically, when regimes fall. What took decades in pre-digital information ages, like the fall of Selassie in Ethiopia or the collapse of the USSR, can happen faster now in closed information environments because a sudden galvanizing movement like the Arab Spring or the Maidan in Ukraine can spring forth in ways they couldn’t previously. And in open societies, like in the United States, in the wake of the 2008 financial crisis, the War on Terror, the opioid crisis, and the COVID-19 pandemic, we see the continuous erosion of trust in institutions and elites in a way that is more consistent and connected across decades than it would have been previously. Why? Because the information environment can amplify any voice, so a topic you scroll past on a social media feed everyday may still be salient to you even if it hasn’t shown up on the nightly news in 5 years.
Neither side has accurately priced the costs it is paying. The interesting strategic question is not which posture wins. It is what each sphere does with the cost it is paying. Open spheres can either pretend the manipulation cost does not exist or build the substrate to absorb it. Closed spheres can either pretend the blindness cost does not exist or build the parallel mechanisms that compensate for it. Both spheres can ignore the legitimacy costs they face in a world where elite action and decision are more visible than ever before, or they could embrace the requirements of a new era and reap the benefits of a more connected populace. The answer each side reaches for, often unevenly and over decades, will shape the next era of strategic competition more than any single policy choice.
What’s at Stake
The gap between the map most institutions still use and the map the world actually inhabits is not just an intellectual failure. It is a strategic vulnerability.
For multinationals, the operating assumption of most corporate strategy is still that "the internet" is a single global marketplace with rules that converge over time. This is the "global commons" idea that shaped a generation of market expansion decisions, but that model is dead. And planning from it results in two long term costs.
Stranded Investments: market expansion into a new sphere requires understanding the world from that sphere’s perspective. While something like a Russian ground invasion into Ukraine or China’s ability to build a domestic AI supply chain may have looked unlikely from the point of view of an executive sitting in North America, the foreign firms that shut down Russian operations after 2022 and the hyperscalers watching their Chinese AI market share dwindle today are examples of how one sphere’s worldview doesn’t map onto the reality of another.
Sphere Lock In: Sunk costs in one sphere create path-dependencies that make sphere exit expensive or impossible even after the strategic case for staying has collapsed. Where stranded investments are the cost of entering a sphere you can't operate in, sphere lock-in is the cost of not being able to leave one you already entered. Look at Apple’s iPhone supply chain, where home-sphere pressure required diversifying production, and moving even 20% of production into India has meant standing up an entire new industrial base. But lock-in cuts both ways. ByteDance built up an audience of about 170 million users in the United States and became so locked into the Open Sphere that when Washington demanded divestment in 2024, Beijing’s 2020 export controls on personalized-recommendation technology made a clean sale impossible.
Together these compound into strategic naiveté and organizational instability, as institutions that have built their planning around an old wrong map chart their path forward in a new world.
For governments, the miscalculation is bigger and older. This is not a set of technology-policy problems. It is a new Great Game. Compute, platforms, regulation, information — each layer is now a domain of geostrategic competition, and the choice to invest, to regulate, to defend, or to sit out, at each layer, is a choice about influence and the ability to project power:
Which sphere sets the technical standards the next billion internet users build on.
Which sphere gets to inspect a data center on foreign soil.
Which sphere writes the AI norms that developing countries adopt.
Which sphere wins hearts and minds in the seams of the map, and which loses them.
The United States has spent a decade governing the sphere's edges with export controls, sanctions, and CFIUSm while under-investing in the sphere's interior. That posture worked when the interior was assumed to be aligned, and under the control of Washington and Silicon Valley. It is not aligned anymore, and the cost of continuing to assume it is falls on American power projection and American commercial position simultaneously.
What is actually required, for both the public and private sectors, is a coherent sphere-aware posture. That doesn’t mean conflict, the goal is not a new Cold War over the internet. But it means awareness: naming which sphere you are operating in, what its rules are, what the competing spheres are doing, and where the seams offer influence or vulnerability. The prize for the multinational with a sphere-aware strategy is markets its competitors do not yet understand. The prize for the government with a sphere-aware doctrine is influence in places that are still choosing. The strategy that wins in this environment is the one that operates from the map that exists, not the one that was promised.
Recommendations
The first recommendation is diagnostic. Every strategy that touches the internet, from national security and regulatory strategy to industrial policy and corporate strategy, must name the sphere it applies to. ‘The global internet’ should be a phrase that requires justification, not the default assumption. A strategy is sphere-aware if it states which sphere is in scope, what the rules in that sphere are, what the relevant adversaries' positions inside it are, and what the costs of operating across spheres look like for the actor concerned. A strategy is not sphere-aware if it talks about ‘the internet’ without granularity. A strategy that is not specific about the sphere it applies to is a strategy whose costs are going to be an unnecessary surprise.
The second part is strategic. The information layer of open societies is structurally exposed, and it will be exposed for the foreseeable future. Recurring information-layer reform agendas are necessary, but they will not be sufficient. The contest is moving, and ought to move, down the stack to the layers where state and corporate action can move at scale. The infrastructure layer, where sovereign capacity is being constructed by patient state capital and by allied industrial policy. The platform layer, where design choices made now will shape the information layer's behavior for the next decade. The regulation-and-governance layer, where the rules of cross-sphere operation are being written, today.
The map has already changed. Those that accept that will be able to keep pace with it. Others won’t.