Open Source AI Mirrors Bitcoins 2014 Fight Against Gatekeepers

On July 1, 2025, Brownstone Research analyst Ben Lilly published a new installment of his Chain of Thought newsletter with a thesis that should sound familiar to anyone who watched Bitcoin grow from a curiosity into a trillion-dollar asset class. The open-source AI movement, Lilly argues, is following the exact same playbook that Bitcoin ran a decade ago. Permissionless access. Jurisdictional arbitrage. Corporate incumbents scrambling to maintain control over a technology that, by its nature, resists control. The parallel is not just rhetorical. It is structural, and it carries investment implications that the mainstream has barely begun to price in.

The Permissionless Pattern

The core of Lilly's argument rests on a single mechanism: permissionless access. In 2014, Bitcoin offered anyone with an internet connection the ability to send value without a bank's approval. No credit check. No intermediary. No government-issued permission slip. That was the radical promise, and it took years before the financial establishment stopped laughing at it.

Open-source AI is now making the same offer in a different domain. Meta's Llama family of models can run on consumer hardware costing a few thousand dollars. Hugging Face, the model-hosting platform, lists over 500,000 models available for free download, fine-tuning, and deployment. A developer in Lagos or Hanoi can now build an AI application with the same foundational tools available to an engineer at Google. No licensing agreement. No corporate gatekeeper.

This matters because the alternative is a world where three or four companies, principally OpenAI, Google DeepMind, and Anthropic, control the most powerful general-purpose technology since the internet. Their models are closed. Their APIs are metered. Their terms of service determine what billions of people can and cannot build. That is not a technology ecosystem. It is a feudal arrangement with a subscription fee.

The open-source countermovement rejects that arrangement on principle. And principle, in technology, tends to win over time. Linux beat proprietary Unix. The open web beat CompuServe and AOL. Bitcoin beat the premise that only licensed institutions should move money. The pattern repeats because permissionless systems attract more builders, generate more innovation, and prove harder to kill than centralized alternatives.

The Regulatory Replay

The regulatory parallels between AI in 2025 and Bitcoin in 2014 are almost uncanny. Lilly points to New York's BitLicense, the regulatory framework proposed in 2014 that imposed heavy compliance costs on cryptocurrency businesses. The BitLicense did not destroy Bitcoin. It destroyed New York's position as a crypto hub. Companies left. Developers moved to Wyoming, Miami, Singapore, and Zug. The technology grew everywhere except the jurisdiction that tried hardest to control it.

Today, the same script is running in AI. The European Union's AI Act, which took effect in stages beginning in 2024, imposes disclosure requirements, risk classifications, and compliance burdens on AI developers. California's SB 1047, which would have required safety evaluations for large AI models, passed the legislature but was vetoed by Governor Gavin Newsom in September 2024. Newsom cited concerns that the bill would drive AI companies out of California, the state where most of them are headquartered.

The veto was revealing. Even a progressive governor in the most regulation-friendly state in America recognized that overly restrictive AI rules would simply push development elsewhere. This is jurisdictional arbitrage, the same force that shaped Bitcoin's geography. Builders go where they can build. Capital follows. The jurisdictions that welcome open development attract talent and tax revenue. The ones that try to license and control it watch their industry migrate.

The EU, by contrast, appears committed to the control path. Brussels has chosen compliance-heavy rules that will almost certainly benefit large incumbents, the Microsofts and Googles that can afford armies of regulatory lawyers, at the expense of smaller open-source projects that cannot. This is precisely how New York's BitLicense worked in practice. It did not protect consumers. It protected incumbents.

The Infrastructure Investment Thesis

Lilly frames the opportunity in familiar terms: picks and shovels. During the California Gold Rush of 1849, the merchants who sold mining equipment made more reliable fortunes than most prospectors. During Bitcoin's early years, the equivalent was mining hardware and exchange infrastructure. Companies like Bitmain and Coinbase built the physical and financial rails on which the entire ecosystem ran.

In AI, the picks-and-shovels layer includes semiconductor manufacturers (Nvidia's data center GPU revenue hit $47.5 billion in fiscal year 2025), cloud providers willing to host open-source models, and protocol-level projects that pair on-chain payments with distributed compute. These are the companies building infrastructure that will matter regardless of which specific AI models win.

Lilly is careful not to name specific tickers, keeping his analysis at the thematic level. But the logic is clear. The companies closest to critical infrastructure, compute, data pipelines, and distribution platforms, are the ones most likely to survive the coming shakeout. Most AI startups will fail. Lilly draws the analogy to the altcoin boom of 2014, when hundreds of Bitcoin clones launched with grand promises and nearly all went to zero. The few that survived, Ethereum being the most notable example, did so because they offered genuinely differentiated infrastructure rather than marginal improvements on the original.

The same dynamic will play out in AI. The startups building slightly better chatbots or marginally improved image generators will not make it. The ones building the compute layer, the distribution platforms, and the payment rails for machine-to-machine transactions have a real shot.

Bitcoin's Deeper Lesson for AI

The Bitcoin analogy works at the surface level of market dynamics and regulatory arbitrage. But it runs deeper than Lilly's newsletter explores.

Bitcoin did not just decentralize payments. It proved that a system with no central authority, no CEO, and no marketing department could grow into a $2 trillion asset class held by sovereign wealth funds and central banks. It proved that permissionless systems are not just ideologically appealing but practically superior to their permissioned counterparts over long time horizons.

This is the Austrian economics insight that underpins the entire story. Friedrich Hayek argued in "The Use of Knowledge in Society" that no central planner can possess the dispersed knowledge held by millions of individual actors. Central planning fails because the planner cannot know what the market knows. Bitcoin proved Hayek right in the domain of money. Open-source AI is testing the same hypothesis in the domain of intelligence.

When OpenAI decides what its models can and cannot do, when Google DeepMind fine-tunes its systems to avoid certain topics, when Anthropic builds guardrails that reflect the preferences of a small team in San Francisco, they are acting as central planners. They are making decisions about what billions of people should think, create, and compute. Their intentions may be good. But Hayek's critique applies regardless of intention. No small group, however talented, can optimize a system as complex as human knowledge production.

Open-source AI distributes that decision-making power to the edges of the network, exactly as Bitcoin distributes monetary policy to the edges of its network. The result will be messier. It will produce models that some people find objectionable. It will enable uses that regulators would prefer to prohibit. But it will also produce more innovation, more adaptation, and more resilience than any centrally managed alternative.

The Bitcoiner who grasps this does not need Lilly's newsletter to see the pattern. The fight for open-source AI is the fight for individual sovereignty applied to a new domain. It is the same fight, with the same adversaries, playing out on a different battlefield.

The Skeptical View

Not everyone buys the analogy. Critics point out that AI development requires vastly more capital than Bitcoin ever did. Training a frontier model costs hundreds of millions of dollars. Nvidia's H100 GPUs, the workhorses of AI training, sell for $25,000 to $40,000 each, and a single training run can require thousands of them. This capital intensity, skeptics argue, makes AI inherently more centralized than Bitcoin, which anyone could mine with a laptop in 2009.

There is truth in this objection. The compute requirements for training frontier models do favor well-capitalized incumbents. But the objection misses a crucial distinction: the cost of running an already-trained model is falling fast. Meta's Llama 3 models can run inference on a single consumer GPU. The cost of fine-tuning an open model for a specific task has dropped from millions to thousands of dollars in the span of two years. Training is centralized. Deployment is not.

There is also a governance concern. Open-source AI models can be used to generate misinformation, produce dangerous content, or automate cyberattacks. The closed-source labs argue that their gatekeeping role serves a genuine safety function. This is a legitimate concern, but it is also the exact argument that banks and regulators used against Bitcoin for years. "If you let people transact freely, bad things will happen." Bad things did happen. Bitcoin was used on Silk Road. It was used for ransomware payments. But the technology also enabled financial sovereignty for millions of unbanked people, provided a hedge against hyperinflation in Venezuela and Turkey, and created an entirely new asset class that now anchors the portfolios of major institutional investors.

The benefits of permissionless systems are diffuse and long-term. The costs are concentrated and visible. This asymmetry always favors the critics in the short run and the builders in the long run.

What to Watch

Three developments will determine whether Lilly's Bitcoin-AI analogy holds up over the next 12 to 24 months.

First, watch the regulatory map. If the EU AI Act drives meaningful AI talent and capital out of Europe, the jurisdictional arbitrage thesis is confirmed. Early indicators include the relocation patterns of AI startups and the funding flows tracked by PitchBook and Crunchbase. If European AI venture funding drops relative to the US and Asia in 2026, the pattern matches Bitcoin's post-BitLicense migration.

Second, watch open-source model performance. The gap between closed-source frontier models and their open-source equivalents has been narrowing. If Meta's next Llama release, or a competitor like Mistral or Falcon, matches GPT-class performance on standard benchmarks, the closed-source moat evaporates. Hugging Face's Open LLM Leaderboard is the scoreboard to track.

Third, watch the convergence of AI and crypto infrastructure. Projects building decentralized compute networks, on-chain model registries, and machine-to-machine payment rails are the most direct expression of Lilly's thesis. If any of these achieve meaningful adoption, measured in actual compute hours sold or transactions processed rather than token market caps, the picks-and-shovels thesis becomes investable rather than theoretical. The next 18 months will tell us whether this is a genuine structural shift or just a clever analogy recycled for a new hype cycle.

---

Source: Bitcoin Magazine