The Cypherpunks Won the Wrong War

You built sovereignty at the monetary layer. The cognitive layer is being captured right now. And nobody is fighting it.

The cypherpunk diagnosis was the most prescient political insight of the late twentieth century.

Not because of Bitcoin. Not because of price. Because of the underlying claim that preceded all of it by two decades: that freedom in the modern world is not primarily a legal or political question. It is an infrastructure question. Whoever controls the settlement layer controls the outcomes. The visible institutions, central banks, payment processors, correspondent banking networks, were legitimacy wrappers around execution layers they governed but did not publicly acknowledge governing. And once you understood that infrastructure was political, the only coherent response was to build parallel infrastructure outside it.

That diagnosis was correct.

It produced the most successful monetary experiment in history. Satoshi published the whitepaper in 2008 drawing on decades of cypherpunk intellectual work: Tim May’s crypto-anarchist manifesto, Eric Hughes’s cypherpunk manifesto, Hal Finney’s proof-of-work research, Adam Back’s Hashcash, Nick Szabo’s bit gold. The technical implementation was Satoshi’s. The political theory was the movement’s. Together they built what they said they would build and the world changed around it.

Bitcoin is now held by national governments, including through the United States Strategic Bitcoin Reserve established by executive order in March 2025, as well as by corporate treasuries and institutional funds worldwide. The Lightning Network processes transactions outside legacy rails. Stablecoins have become the de facto dollar system for tens of millions of people outside OECD banking access.

And yet something feels wrong.

Not financially. Financially the project succeeded beyond any reasonable 2009 expectation. What feels wrong is the story. The original thesis that freedom requires sovereignty over your own infrastructure, that captured systems cannot reform themselves from within, that the answer is parallel construction outside the captured layer, that thesis has been partially absorbed and largely neutralized. Bitcoin became an asset class. The revolution became a portfolio allocation. The energy that once felt like it was building a parallel civilization now mostly feels like it is tracking a price chart.

The OGs who built this feel it. The people who came for the original thesis rather than the financial upside feel it. The winter is not primarily about price. It is about narrative. The original story has lost its forward motion.

Here is why.

The cypherpunks won the wrong war. Not because the monetary battle was unimportant. It was important. But while that battle was being fought and partially won, a more fundamental capture was proceeding at a deeper layer of the stack. A capture more consequential than anything a central bank has ever done. A capture with less democratic accountability than any institution in modern history. A capture that is closing right now, in this window, before anyone has built the parallel infrastructure that would contest it.

The cognitive layer is being captured.

And almost nobody is fighting it.

What the Cypherpunks Actually Understood

The cypherpunk insight, articulated by Tim May, Eric Hughes, and Hal Finney in the early 1990s, was not primarily about money. It was about the relationship between infrastructure and freedom.

Hughes wrote in 1993: “We cannot expect governments, corporations, or other large, faceless organizations to grant us privacy out of their beneficence.” [1] The point was not that these institutions were evil. The point was structural: institutions optimized for their own continuity will use whatever infrastructure they control to preserve that continuity. If they control the communication layer, they will surveil it. If they control the payment layer, they will censor it. The solution was not to reform the institutions. It was to build infrastructure they could not control.

The thesis had three components worth separating:

Infrastructure determines politics. Not the other way around. Laws and policies are downstream of the systems that execute them. Control the execution layer and you control the outcome regardless of what the visible institutions declare.

Captured systems cannot self-reform. The Federal Reserve will not voluntarily surrender its ability to inflate. SWIFT will not voluntarily remove its ability to exclude. Reform from within is structurally impossible when the institution’s power derives from the infrastructure being reformed.

Parallel construction is the only answer. Not lobbying. Not voting. Not petitioning. Building infrastructure outside the captured system that makes the captured system irrelevant by providing a superior alternative.

All three components were correct. All three remain correct. And all three apply with full force to a layer the original cypherpunks could not yet see.

The Layer That Is Being Captured Now

The monetary layer was never the only infrastructure layer that mattered for freedom. It was the most legible one in 1993. So that is where the movement built.

But infrastructure determines politics across multiple layers simultaneously. And while the monetary layer battle was consuming the attention of the most adversarially minded builders in the world, three other layers were being quietly captured in ways that make central bank control of money look modest by comparison.

The Compute Layer

Ninety percent of all AI compute is managed by American and Chinese companies. [2] TSMC, a single company on a single island, manufactures roughly 90 percent of the world’s most advanced chips, the ones that matter for frontier AI. [3] Three hyperscalers, Amazon Web Services, Microsoft Azure, and Google Cloud, mediate the majority of the world’s cloud infrastructure. [4]

This is not ordinary market concentration. This is territorial capture.

Compute is not like other industries. It has the properties of estate rather than factory. A factory produces output. Estate controls the conditions under which others can produce at all. The compute estate, chips, data centers, energy systems, cooling infrastructure, and fiber corridors, is the physical substrate on which synthetic intelligence runs. You cannot think at scale outside it. You cannot build at scale outside it. You cannot compete at scale outside it.

The International Energy Agency projects that global data center electricity consumption will more than double by 2030, reaching roughly 945 terawatt-hours, comparable to Japan’s entire current electricity consumption. [5] Five major technology companies spent more than $400 billion on AI infrastructure in 2025 alone. [5] McKinsey estimates that AI-related data center infrastructure will require roughly $5.2 trillion in cumulative capital expenditure by 2030. [6]

This is estate being built at a speed that forecloses alternatives before they can form. Once compute infrastructure exists at this scale, shutting it down carries systemic risk. Regulating it threatens competitiveness. Building competing infrastructure requires capital that only states and the largest corporations can deploy.

The window in which parallel compute infrastructure could be built by independent actors is not infinite. It is closing.

The Cognitive Layer

The compute layer is the physical substrate. The cognitive layer is what runs on it.

Frontier AI models, the systems that will increasingly mediate how institutions see, classify, decide, and act, are controlled by a handful of private firms with no democratic mandate and no public accountability.

Consider the comparison. The Federal Reserve has a congressional mandate. Its governors are Senate-confirmed. Its balance sheet is public. Its chairman testifies before Congress twice a year. You can disagree with its decisions. You can hold it accountable through democratic processes. You can read its minutes.

Anthropic, OpenAI, and Google DeepMind are writing the standards that govern what AI systems are allowed to do, which capabilities trigger additional safety requirements, and what vocabulary the entire global regulatory conversation uses to discuss these questions. Anthropic’s Responsible Scaling Policy became one of the first private governance frameworks whose vocabulary shaped how regulators, policymakers, and international bodies discussed frontier AI risk, before most of them had a working definition of what frontier AI was. The vocabulary of AI Safety Levels, capability thresholds, and deployment gates entered international regulatory discussions because one private company wrote it down first. [7]

This is not regulatory capture in the ordinary sense. Ordinary regulatory capture happens after democratic institutions create regulatory frameworks that get colonized by industry. This is pre-regulatory capture. The standard is being written before the law exists to mandate it. By the time democratic institutions arrive, the vocabulary, the risk taxonomy, the evaluation methodology, and the compliance architecture will already have been established by the institutions most financially interested in their outcome. [8]

The institution that writes the standard writes the law.

That institution is currently a small number of private firms in two countries.

The Perception Layer

This is the most consequential capture and the least discussed.

A modern state cannot act on what it cannot see. Before policy, there is a map. Before enforcement, there is classification. Before targeting, there is sensor fusion. Before welfare allocation, there is eligibility logic. Before military action, there is battlefield cognition.

AI companies are entering exactly there.

The U.S. Army awarded Palantir a $178.4 million contract to develop the Tactical Intelligence Targeting Access Node, a system that integrates sensor data across space, aerial, and terrestrial platforms to improve targeting and reduce the sensor-to-shooter timeline. [9] The U.S. General Services Administration struck a government-wide agreement giving all three branches of the federal government access to Anthropic’s Claude. [10] NHS England awarded a £330 million contract to a Palantir-led consortium to connect health data across NHS systems. [11] OpenAI launched OpenAI for Government in June 2025 specifically to bring AI tools into public sector workflows across the United States. [12]

The general still authorizes the strike. The minister still announces the policy. The caseworker still signs the benefit decision. The formal chain of command remains intact.

But the environment in which the general sees the battlefield, what threats appear, how they are ranked, what actions seem available, and what timeline feels urgent, has begun to pass through a private machine layer. The officer does not experience this as a transfer of authority. She experiences it as clarity.

That is the form of capture that is hardest to see and hardest to reverse.

A state that loses territory can imagine recovery. A state that loses perception begins to confuse borrowed cognition with its own sovereignty. The eye of power belongs to someone else. The flag still flies. The officials still speak. The institutions still function.

But the map is no longer theirs.

The Dependency Ratchet

The reason this capture is more urgent than it appears is a mechanism that makes it progressively harder to contest the longer it runs.

Every institutional adoption of AI infrastructure follows the same sequence.

A government agency faces backlogs, budget pressure, and understaffed technical teams. A vendor offers tools that solve real problems: faster processing, better fraud detection, improved patient flow visibility, and reduced sensor overload. The agency adopts the tools. Efficiency improves. The adoption looks like a success.

Then something changes.

The weekly meeting agenda reorganizes around the platform’s categories. Staff are trained on the vendor’s workflow. Budget lines adapt. Internal vocabulary shifts. Performance metrics are defined by what the system can measure. Ministerial briefings are structured around the dashboard.

By the time the contract comes up for renewal, the agency cannot leave. Not legally — the exit clause is right there in the contract. But operationally: leaving would mean retraining staff, rebuilding workflows, recreating historical context, explaining short-term performance drops, and exposing the institutional weakness that made the adoption necessary in the first place.

The system is no longer a tool the agency uses. It is where the agency remembers how it works.

This is the dependency ratchet. Each adoption makes the next exit more expensive. Each integration converts a vendor relationship into institutional architecture. Each workflow adaptation makes the underlying question, who actually governs here, harder to ask.

The cypherpunks understood this mechanism in the monetary domain. The correspondent banking system did not capture global finance through a single decision. It captured it through accumulated dependency, each bank, each transaction, each clearing relationship adding one more tooth to the ratchet until exit became structurally impossible for any individual actor.

The same ratchet is running now in the cognitive domain. But it is running faster, because software dependencies form faster than banking relationships. And it is running deeper, because the systems being adopted govern perception itself, not just transactions, but what institutions can see, classify, and act on.

The window in which parallel cognitive infrastructure can be built before the ratchet closes is not fifty years. It is not twenty years. By this analysis, the infrastructure investment numbers, the procurement timelines, and the institutional adoption curves together suggest a window of five to ten years at most.

After that, contesting the cognitive layer will be as structurally difficult as contesting the correspondent banking system was before Bitcoin. Possible in theory. Prohibitively expensive in practice. Requiring decades of parallel construction before any meaningful alternative exists.

What the Cypherpunk Project Actually Is

The cypherpunk movement built the thesis and the early infrastructure. PGP. DigiCash. Hashcash. Anonymous remailers. Satoshi took that foundation and built the specific implementation that proved the thesis at scale. Together they demonstrated three things almost nobody else in the world had demonstrated simultaneously: that infrastructure is political at a bone-deep operational level, that the adversarial assumption is correct, and that building outside captured systems without institutional permission is possible.

Those three things are not taught in universities. They are not rewarded by corporate employment. They are not produced by the normal processes of elite institution formation.

They are, however, heavily concentrated in one specific community that built one specific thing and proved the thesis.

The cypherpunk project, stated completely, is sovereignty across the full stack:

Monetary layer: partially built. Real progress. Not finished.

Communication layer: partially built. Signal, Tor, end-to-end encryption. Not finished.

Compute layer: essentially unbuilt. The problem has barely been named.

Cognitive layer: unbuilt. The capture is proceeding uncontested.

Identity layer: unbuilt. Self-sovereign identity remains marginal.

Compliance layer: unbuilt. Who decides who can transact is still determined by captured institutions.

Perception layer: unbuilt. Who the state uses to see is entirely in private hands.

The project is maybe fifteen percent complete.

The hardest layers are the ones being actively captured right now, in this window, by actors who understand exactly what they are building and have the capital to build it faster than any distributed community has yet organized to contest.

The Window

The Federal Reserve was established in 1913. By then, the correspondent banking system had been building for decades. The infrastructure of monetary capture was already so deeply embedded in global commerce that contesting it required either building entirely outside it, which is what the cypherpunk tradition eventually made possible, or accepting it as a fixed constraint.

That is where we are with the cognitive layer, except the timeline is compressed.

The correspondent banking system took roughly fifty years to reach the point where exit was structurally prohibitive. The cognitive capture is moving at software speed. The procurement contracts are being signed now. The institutional dependencies are forming now. The standards are being written now. The perception infrastructure is being installed now.

The ratchet is turning.

Every quarter that passes without parallel cognitive infrastructure being built is a quarter in which the capture hardens. Not dramatically. Not visibly. One procurement contract at a time. One standard adopted. One workflow reorganized. One institutional memory transferred from human judgment into a vendor platform.

By the time the capture is visible to most people it will already be irreversible in the way that monetary capture was irreversible before 2008. Technically contestable, practically prohibitive, requiring decades of parallel construction before any alternative is viable at scale.

The window is open.

It will not stay open.

The Completion of What Was Started

The people who need to build the cognitive sovereignty layer are not the AI safety researchers at frontier labs, who are institutionally captured by the very problem they are supposed to solve. They are not the governance academics, who arrive after the architecture has hardened. They are not the technologists building products on top of existing AI infrastructure, who are tenants of the very estate that needs to be contested.

They are the people who already understand, at a bone-deep level, that infrastructure is political. Who have already built once outside captured systems. Who already have the adversarial mindset, the tolerance for operating without institutional permission, and the demonstrated capacity to build things that serious people said were impossible.

The skills that built Bitcoin are the skills the cognitive layer needs.

The energy sitting in narrative winter, waiting for a story that matches the original conviction, is the energy the next layer requires.

This is not a pivot away from what was built. It is the completion of what was started.

The cypherpunk project was never about money. Money was the first layer where the thesis could be demonstrated. It was demonstrated. The thesis is proven. Infrastructure is political. Parallel construction is possible. Captured systems will not self-reform.

Now the thesis needs to be applied to the layers that determine not just who can transact but who can think, who can see, who can know, and who can act at scale in a world where cognition is becoming infrastructure.

The first layer required a whitepaper and a genesis block. The next layers will require their own inventors.

Satoshi did not ask the Federal Reserve if it was okay to build Bitcoin.

The people who build cognitive sovereignty will not ask OpenAI or Anthropic or AWS if it is okay either.

The question is not whether it is permitted.

The question is whether it gets built before the window closes.

Sources

[1] Eric Hughes, “A Cypherpunk’s Manifesto,” March 9, 1993. https://www.activism.net/cypherpunk/manifesto.html

[2] Deloitte Technology Predictions 2026, citing Oxford Internet Institute research: 90 percent of all AI compute is managed by U.S. and Chinese companies. https://www.deloitte.com/us/en/insights/industry/technology/technology-media-and-telecom-predictions/2026/tech-sovereignty.html

[3] Counterpoint Research, Q3 2025: TSMC captured approximately 72 percent of the global semiconductor foundry market. Industry estimates place TSMC’s share of the most advanced process nodes at approximately 90 percent. https://dataconomy.com/2025/12/23/tsmc-dominates-foundry-market-with-72-share-in-q3-2025/

[4] Cambridge Core, “Sovereign AI in 2025”: Amazon Web Services, Microsoft Azure, and Google Cloud together mediate the majority of the world’s cloud infrastructure. https://www.cambridge.org/core/journals/natural-language-processing/article/sovereign-ai-in-2025/C51560626AF518BDF280891C406E9553

[5] International Energy Agency, “Key Questions on Energy and AI,” 2026: global data center electricity consumption projected to more than double to roughly 945 TWh by 2030; five major technology companies spent more than $400 billion on AI infrastructure in 2025. https://www.iea.org/reports/key-questions-on-energy-and-ai

[6] McKinsey and Company: AI-related data center infrastructure could require roughly $5.2 trillion in cumulative capital expenditure by 2030. https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/the-cost-of-compute-a-7-trillion-dollar-race-to-scale-data-centers

[7] Anthropic, Responsible Scaling Policy, Version 3.0, effective February 24, 2026. https://www.anthropic.com/responsible-scaling-policy

[8] National Institute of Standards and Technology, AI Risk Management Framework 1.0, January 2023. https://doi.org/10.6028/NIST.AI.100-1

[9] U.S. Army Program Executive Office, “Army Tactical Intelligence Targeting Access Node (TITAN) Ground Station Prototype Award,” March 6, 2024. https://cpeisw.army.mil/2024/03/06/army-tactical-intelligence-targeting-access-node-titan-ground-station-prototype-award/

[10] U.S. General Services Administration, “GSA Strikes Another OneGov Deal with Anthropic to Offer Claude AI to all Branches of Gov,” August 12, 2025. https://www.gsa.gov/about-gsa/newsroom/news-releases/gsa-strikes-onegov-deal-with-anthropic-08122025

[11] UK Parliament Hansard, “NHS Federated Data Platform: Contract Award,” November 21, 2023. https://hansard.parliament.uk/commons/2023-11-21/debates/23112160000011/NHSFederatedDataPlatformContractAward

[12] OpenAI, “Introducing OpenAI for Government,” June 16, 2025. https://openai.com/global-affairs/introducing-openai-for-government/