The AI Race Is About Sovereignty, Not Just Intelligence
Thinking Machines released Inkling with downloadable weights and a vision for customizable AI. The real race is over who controls the intelligence.
TFTC - Truth for the Commoner Bitcoin Brief | |||||||||||||||||||||||||
Sup, freaks. Yesterday we covered the physical layer of the AI buildout: land, power, data centers, and the split between builders and blockers. Today we move up the stack. Who owns the model? Who can modify it? Who decides what it is allowed to think? Who controls access when the model becomes critical infrastructure for a business, a family, or a country? The AI race is no longer only about who builds the smartest model. It is about who controls the intelligence. | |||||||||||||||||||||||||
LEAD STORY | |||||||||||||||||||||||||
The AI Race Is About Sovereignty, Not Just IntelligenceThinking Machines Lab released Inkling yesterday, and the most important thing about it is not where it sits on a benchmark leaderboard. Inkling is a 975-billion-parameter Mixture-of-Experts model with 41 billion active parameters. It was pretrained from scratch on 45 trillion tokens spanning text, images, audio, and video. It accepts text, image, and audio inputs, supports a context window of up to one million tokens, and gives users control over how much reasoning effort it spends on a problem. Thinking Machines also released the full weights for download. That last point is the story. Most people interact with AI by renting access to a black box. They send data to a company, receive an answer, and remain exposed to whatever the provider decides to change. The model can be updated. The price can rise. The acceptable-use policy can tighten. A government can pressure the provider. The service can disappear. That may be acceptable for casual chat. It is a dangerous foundation for critical business processes, personal knowledge systems, medical research, financial workflows, and national infrastructure. Thinking Machines is making a different argument. Its mission essay says AI should extend human will and judgment instead of making people passive consumers of a frozen system aligned by a handful of laboratories. The argument is Hayekian. Much of the knowledge that makes an organization valuable is tacit, local, private, and created through work. It exists in the heads of employees, in the exceptions buried inside workflows, in relationships, and in hard-earned judgment that rarely makes it into a clean dataset. A centralized lab cannot collect all of that knowledge. It cannot define one alignment layer that properly reflects the values of every person, company, community, and country. Trying to do so creates a massive locus of power. Customizable models offer another path. An organization can encode its own specialized knowledge and judgment into the weights instead of relying entirely on prompt engineering or renting the same generic model as everyone else. Thinking Machines demonstrated the idea by asking Inkling to fine-tune itself into a lipogram model that would never use the letter “e.” Inkling wrote the training job, generated an evaluation, ran the job through the company’s Tinker platform, and staged the new weights. A supervisor then relaunched the updated checkpoint. That was a controlled company demo, not an AI autonomously escaping into recursive self-improvement. But it showed the workflow clearly: the model helped customize itself around a user-defined objective and produced a new set of weights. This matters for the United States too. Yaël Ossowski argues that America can lead the closed-model benchmark race while losing the more consequential contest over which technology stack the world adopts. Chinese open-weight families such as DeepSeek, Qwen, and GLM are spreading because developers can download, modify, and build on them. The country that supplies the foundation may shape the next layer of global products, standards, and technical dependencies. Winning the benchmark while losing the ecosystem would be a hollow victory. Inkling is an important American counterexample. Thinking Machines independently pretrained a massive multimodal foundation model, made its weights downloadable, and built customization tooling around it. The base model was trained from scratch, although the claim that it is completely free of distillation is wrong. Thinking Machines says its post-training began with a small supervised-fine-tuning bootstrap using synthetic data generated by open-weight models including Kimi K2.5. Most post-training compute then went into more than 30 million reinforcement-learning rollouts. We also need to be precise about “open.” Inkling is open-weight, not fully open source in the strongest meaning of that term. Its model card lists Apache 2.0, but Thinking Machines separately subjects the weights and derivative models to an acceptable-use policy that the company can update. The training dataset is not publicly reproducible. The hardware barrier is real too. Running the BF16 checkpoint requires at least two terabytes of aggregate VRAM. The quantized NVFP4 checkpoint still requires at least 600 gigabytes. This is not a model most people will run on a laptop. Open weights are not the end state of AI sovereignty. They are one necessary step toward it. The bitcoin parallel is obvious. Bitcoin gives people the ability to own money without renting permission from a central bank or financial intermediary. Open and customizable AI points toward the ability to own intelligence without depending completely on a corporate API. Neither form of sovereignty is convenient at first. Both demand infrastructure, competence, and responsibility. Both are worth building because systems this important should not be controlled by a handful of institutions. The AI race is not only about intelligence anymore. It is about sovereignty. | |||||||||||||||||||||||||
SIGNAL | |||||||||||||||||||||||||
BITCOIN DEVELOPMENT Open Source Is Not CostlessOpenSats’ transparency page shows the scale of the funding machinery behind bitcoin and freedom tech: 415 grants, $35.3 million allocated, and roughly 41.9 billion sats sent at the time of today’s live check. The more important story is where those sats go. OpenSats supports Bitcoin Core contributors including Marco Falke, Andrew Toth, Sjors Provoost, 0xB10C, Bruno Garcia, Jon Atack, and Calvin Kim. It funds privacy tools such as Dana Wallet and Dan Gould’s Payjoin Dev Kit; mining projects including skot’s Bitaxe and Stratum V2; ecash work around Calle’s Cashu, Cashu Dev Kit, Minibits, and Nutstash; and foundational libraries including BDK, rust-bitcoin, and bitcoinfuzz. The same funding network reaches nostr through Damus, NDK, nak, Amber, Citrine, and Zapstore. Open source does not mean costless. Sovereign software requires sustained funding for maintainers, reviewers, security researchers, educators, and the invisible work that keeps Bitcoin dependable. | |||||||||||||||||||||||||
BITCOIN POLICY America Is Testing Bitcoin While China-Aligned Groups Fight AI InfrastructureThe Bitcoin Policy Institute’s Q2 report ties together several fights over open technology, national security, and infrastructure. The cleanest signal came from the military. In an official transcript published by Rep. Lance Gooden, Adm. Samuel Paparo said INDOPACOM has a node on the Bitcoin network and is running operational tests using the protocol to secure and protect networks. The military is not mining bitcoin. It is studying Bitcoin’s cryptography, proof of work, and network architecture as tools for cybersecurity and power projection. BPI also investigated the political campaign against American AI infrastructure. Its Singham Ground Game report documents 21 campaigns across 14 states involving the Party for Socialism and Liberation. BPI says those campaigns contributed to ten data-center moratoria, one permanent ban, and four rejected or abandoned projects representing approximately $23.6 billion of proposed investment. The report links PSL leadership to nonprofits funded by Neville Roy Singham, a Shanghai-based Marxist who has been the subject of congressional inquiries over documented ties to the Chinese Communist Party. BPI does not claim that every American opposing a data center is a foreign agent. It explicitly acknowledges legitimate concerns about electricity prices, water use, and local development. It also does not prove that China directly finances PSL. The problem is that PSL’s finances are shielded from public disclosure, leaving the public unable to answer that question. That uncertainty is exactly why transparency matters. Authentic local concerns can coexist with a foreign-aligned network working to amplify division and block infrastructure that strengthens a geopolitical rival. The policy fight extends to open-source developers. The Senate Banking Committee advanced the CLARITY Act with Section 604, the Blockchain Regulatory Certainty Act, still in the reported bill. Section 604 says a non-controlling developer should not be treated as a money transmitter merely for publishing software, providing self-custody tools, or supporting distributed-ledger infrastructure. It preserves liability for someone who specifically intends to transfer known criminal proceeds or support unlawful activity. Keeping Section 604 was a win. It was not the end of the fight. Law-enforcement critics argue the language could create loopholes. Developers argue that vague definitions of control and conduct outside the safe harbor may leave prosecutors too much room to treat code publishers as intermediaries. The DeFi Education Fund also notes that a markup amendment removed separate Section 301 language explicitly shielding non-controlling developers from that section’s rulemakings. Open-source developers need protection that is durable and legible. A safe harbor that requires years of litigation to understand is not much of a safe harbor. | |||||||||||||||||||||||||
CULTURE / DEMOGRAPHICS The Young Are Rich in Televisions and Poor in LifeJohann Kurtz argues that the economic statistics telling Millennials and Zoomers they are doing fine are measuring the wrong things. Electronics, entertainment, clothing, and other consumer goods are cheaper. The foundations of a durable middle-class life are not. The median single-family home reached five times median household income in 2024, compared with 3.2 throughout the 1990s. The median first-time homebuyer reached a record age of 40 in 2025. Housing is only part of the problem. Education, childcare, health care, safe neighborhoods, good schools, and stable family formation have become expensive private purchases. Younger households must buy back at retail what previous generations often received through functioning families, communities, churches, schools, and high-trust institutions. Freaks who listened to my conversation with Johann will recognize the thread. In our episode on wealth and Western decline, we focused on why the West stopped preserving wealth, duty, and family continuity across generations. His new essay shows the same collapse from the other side: the young now have to repurchase the social and material inheritance their parents received more cheaply or for free. Currency debasement and asset inflation are the trunk. But money is not the whole story. A society can debase its currency and liquidate its social capital at the same time. | |||||||||||||||||||||||||
POLITICS / ECONOMICS Gen Z Is Turning Against CapitalismPeter St Onge points to new Cato polling that should terrify anyone who believes in free markets. Cato found that 53% of Gen Z respondents viewed socialism favorably, compared with 45% who viewed capitalism favorably. That does not mean a majority supports abolishing markets or understands socialism the same way. It does mean the word “socialism” now carries less baggage for young Americans than the economic system they associate with the status quo. St Onge blames two forces. Schools teach young people to distrust capitalism, while Washington and the Federal Reserve give them an economy that looks rigged. Cheap money lifts the assets owned by older households and connected institutions. Bailouts protect incumbents. Housing and groceries run away from young workers who own few assets and absorb the inflation. You cannot socialize losses, privatize gains, debase the currency, protect incumbents, and then act surprised when the generation left outside concludes that capitalism failed them. What they have experienced is not a free market. It is fiat cronyism. The tragedy is that more central planning would deepen the same concentration of power that created the backlash. The answer is sound money, lower barriers to building, an end to bailouts, and rules that apply equally to insiders and outsiders. Bitcoin provides the monetary foundation for that alternative. | |||||||||||||||||||||||||
MACRO / LIQUIDITY The AI Boom May Be Absorbing Financial-Market LiquidityMichael Howell’s latest Capital Wars analysis offers a useful warning for bitcoin investors. Howell argues that the AI capex boom is moving capital out of financial markets and into the real economy. Global Liquidity momentum is rolling over while economic growth accelerates. In his model, that is a late-cycle regime that favors industrial commodities, especially base metals, while pressuring long-duration bonds and reducing the relative dominance of gold and bitcoin. That is not a permanent bearish call on bitcoin. It is a cycle call. Bitcoin can remain a long-run hedge against monetary debasement while underperforming the copper, energy, and physical inputs required to build the next investment boom. | |||||||||||||||||||||||||
| |||||||||||||||||||||||||
| |||||||||||||||||||||||||
⚡ FREEDOM TECH CORNER | |||||||||||||||||||||||||
| |||||||||||||||||||||||||
DATA SNAPSHOT | |||||||||||||||||||||||||
| |||||||||||||||||||||||||
| |||||||||||||||||||||||||
| |||||||||||||||||||||||||
Forward this to a freak who understands that sovereignty requires builders. See you tomorrow, Marty | |||||||||||||||||||||||||
Marty Bent: https://x.com/MartyBent?ref=tftc.io TFTC: https://x.com/TFTC21?ref=tftc.io Nostr: https://primal.net/marty?ref=tftc.io YouTube: https://www.youtube.com/@TFTC Podcast: https://www.tftc.io/tag/podcasts/ | |||||||||||||||||||||||||
