# Thesis v1 (superseded) > **This is v1. It is superseded by [`thesis-v2-script.md`](thesis-v2-script.md) (17 Sep 2026), the full spoken script Jacob rehearses from.** > > What changed between v1 and v2, in one breath: the title is now "A mind outside the state", not "The machine that pays you back". The spine follows `trajectory.md` (Weber, the two checks, the merger, no outside, exit over voice, Bitcoin, the licensed press, a market, we built one, small and enough, the key, the line). The moral case is rules over rulers and exit over voice, not distribution or "more power to more people". The centrepiece is the two worlds and the switch (`two-worlds.md`). The one metaphor is the switch; Elysium is said once; Gaza, fascism, and the origin story are out. The nature and memristor sections are gone. "Permissionless basic capital" is gone. Every number uses the corrected wording from `research/factcheck-slides-v2.md`. The key gets a date, said first, by him. > > The v1 text is kept below for reference only. Do not rehearse from it. --- # Thesis: The machine that pays you back Working title for the keynote. Exploit Conference, Montreal, 28 September 2026. 45 minutes. Speaker: Jacob Steeves (Const), founder of Bittensor. This is the argument, in the order it is made on stage. It is written in Jacob's first person so he can hear it. It is not a script. Every number here is sourced in `docs/research/state-of-bittensor-2026.md` and `docs/research/economics-of-ai-and-labor-power.md`. Numbers marked [VERIFY] must be checked on the day. Ground rule from Jacob: the origin story of the inspiration stays out. What stays in is what the inspiration points to: nature is decentralized, and a new form is needed to grow intelligence inside the larger human biome. --- ## 0. The one-paragraph version AI is moving economic power away from people who work and toward a few companies that own the machines. Every serious economist agrees on the cure: ordinary people must own a piece of the AI capital. Every serious proposal for how to do that runs through a national government or through the goodwill of the companies doing the concentrating. Bittensor is the one design where ownership of AI capital is issued directly to anyone who contributes, under open rules, in no country. Bitcoin proved that kind of network can scale to the size of a nation's power grid without a CEO. Bittensor points that same machine at intelligence. It is small, it is not finished, and I hold a key I intend to destroy. But it is the only thing of its kind, and this is the decade it has to work. --- ## 1. Open on the angst Start in the room, not on the slide. Everyone in this room has felt it in the last twelve months. A moment when a model did something that used to be your job, or your friend's job, and you did not know whether to be thrilled or afraid. Both, probably. Name the feeling. Then say: the feeling is correct. Something real is happening. I am not here to calm you down. I am here to tell you what the fear is actually about, because most people are pointing it at the wrong thing. Beat: Anthropic's own CEO said in 2025 that AI could remove half of entry-level white-collar jobs and push unemployment to 10 or 20 percent within one to five years. His words: "Cancer is cured, the economy grows at 10% a year, the budget is balanced, and 20% of people don't have jobs." (Amodei, Axios, 28 May 2025.) The people building this are telling you what it does. ## 2. What is actually happening: the firm without a payroll Now the numbers. Only a few. Let them land. - The US labor share of income, the slice of the economy paid to people for working, is 52.9 percent. That is the lowest number since the record began in 1947. It was 65.8 percent then. (BLS, Q2 2026, preliminary.) [VERIFY] - Anthropic: valued at $965 billion with a few thousand employees. OpenAI: $852 billion with under eight thousand. NVIDIA: $216 billion of revenue with 42,000 people. (Anthropic release, May 2026; Sacra, Mar 2026; NVIDIA 10-K, Feb 2026.) - General Motors at its peak in 1979 employed 853,000 people. It was the largest private employer in America. Walmart today employs 2.1 million. Do the arithmetic on stage: Anthropic is worth about a trillion dollars with roughly half of one percent of GM's headcount. Roughly two hundred million dollars of value per employee. Walmart does about $324,000 of revenue per employee. The gap is not ten times. It is hundreds of times. The point: for a hundred years, the way a frontier company shared its gains with society was payroll. GM paid 853,000 people. That was the transmission mechanism. The AI lab has no transmission mechanism. It does not need one. Its value goes to whoever owns it. And who owns it? In the US, the top 10 percent of households hold about 87 percent of corporate equity. The bottom half holds 2.5 percent of net worth. (Federal Reserve, Q2 2025.) Capital is pooling: four companies plan to spend $720 to $745 billion on data centers this year. AI took 53 percent of all venture dollars on Earth in 2025. (Company earnings calls, Jul 2026; PitchBook, Jan 2026.) Say plainly what this is: the most valuable thing humans have ever built is being built by the fewest people, owned by the fewest people, in the fewest places. ## 3. Why labor mattered: it was the leverage This is the part most AI talks skip. Do not skip it. Why did ordinary people ever have power? Three reasons, and each one is about work. First, work gathered people. A radical named John Thelwall wrote in the 1790s that "every large workshop and manufactory is a sort of political society, which no act of parliament can silence." The factory made the union. The union made the vote matter. (Quoted in Acemoglu and Johnson, *Power and Progress*, 2023.) Second, work was the tax base. Governments run on taxing wages and what wages buy. Two economists at the NBER wrote this year that transformative AI "may gradually erode the two main tax bases that underpin modern tax systems: labor income and human consumption." (Korinek and Lockwood, 2026.) A state that does not need your labor or your taxes has less reason to listen to you. Third, history says the gains do not share themselves. In Britain from 1780 to 1840, output per worker rose 46 percent. Real wages rose 12 percent. The profit rate doubled. Economists call it Engels' pause. It lasted sixty years and it ended only when workers organized. (Allen, 2009.) Now the question that this whole talk turns on. Say it slowly: **What does organizing look like when the machines no longer need the people?** If your labor was your leverage, and the machine takes the labor, you have no leverage. Not with your employer. Not with your government. You become, in the plainest terms, an unimportant extremity of the productive engine that runs the world. ## 4. The real fear Here is where I disagree with almost everyone who talks about AI risk. The fear is not that artificial intelligence takes over the world. Machines do not want anything. The fear is that a small group of humans takes over the world, uses AI to do it, and separates themselves from the rest of us. Picture it. There is a film called *Elysium*. A few thousand people live on a station in orbit with perfect medicine and perfect security. Everyone else lives below, with no say in anything that matters. Nobody down there is oppressed by a robot. They are oppressed by the people who own the robots. That is the actual shape of the danger. It is not science fiction. It is what the labor-share chart looks like if you extend the line. And I want to be honest about who is describing this future. Not me. Sam Altman proposed a fund that takes 2.5 percent of every large company's equity each year and gives it to every adult. Dario Amodei proposed a 3 percent tax on AI revenue. (Altman, 2021; Amodei, 2025.) The people building the concentration are proposing to share it. They know. They are just proposing to do it through the US Treasury. ## 5. Everyone agrees on the cure. Nobody has a mechanism. Slow down here. This is the hinge of the argument. The economists agree on the diagnosis: AI shifts income from labor to capital, and capital is owned by the few. The economists also agree on the cure. Anton Korinek told the US Senate in 2023 that without reform, AGI "would likely deliver mass impoverishment," and proposed what he calls "universal basic capital": every person owns a piece of the AI. Piketty wants a wealth tax. Glen Weyl and Jaron Lanier want people paid for the data that trains the models. Altman wants an equity fund. Amodei wants a token tax. Susskind wants a "Big State." Read the list again and notice one thing. **Every single proposal runs through a national government, or through the voluntary goodwill of the very companies doing the concentrating.** Tax and redistribute. Or ask nicely. Both depend on a state that is losing its tax base, or on corporations that would be giving away their own advantage. And every one of them works inside one country's borders. If you are not American, Altman's American Equity Fund does nothing for you. If your government does not act, Korinek's plan does nothing for you. So the question becomes: is there a third mechanism? One that does not need a treasury and does not need permission? Yes. We have already seen one work. It is fifteen years old. ## 6. Bitcoin: the proof that incentive networks scale Bitcoin did two things. Everyone knows the first: money nobody can print. The second is the one that matters tonight. Bitcoin created the first permissionless commodity market on Earth. It said: anyone, anywhere, bring compute, and I will pay you directly, every ten minutes, no contract, no HR department, no headquarters. Look at what that sentence built. The Bitcoin network today runs at about a thousand exahashes per second. A thousand quintillion operations every second. It draws as much electricity as Poland. Miners hunt for stranded hydro in Paraguay and flare gas in Texas and cheap power in Ethiopia. Nobody planned any of it. Nobody hired anyone. Incentives did it. (Hashrate Index, Q2 2026; Cambridge CBECI, 2026.) Note to self: hashes are not FLOPs. Do not say Bitcoin out-computes Google. Say it coordinates more raw machinery than any company or government has ever assembled for one purpose, with no one in charge. That is the lesson I took from Bitcoin, and it is the founding belief of Bittensor: **we can build systems governed by networks and incentives, not by a central authority, and those systems can grow to a size that competes with nations.** The problem with Bitcoin is that the thing it produces, a SHA-256 hash, is useless for anything but securing Bitcoin. All that machinery, one product, and the product is a lock. What if the same machine produced the most valuable commodity of this century? ## 7. Nature is not centralized A short turn. Keep it grounded and personal without the origin story. Look at any living system. A forest. A reef. Your own body. There is no CEO. There is no central planner. There are trillions of elements, each one following local rules, each one exchanging energy with its neighbors, and out of that comes something more intelligent than any of its parts. Nature is decentralized. It is also the most successful intelligence-producing process we know of. It produced us. I was also shaped by neuromorphic computing. A company I followed built chips out of memristors, components that learn by feeling the flow of energy through them. The chip does not run a program. Current flows, paths that carry more current get stronger, and structure emerges. [CONFIRM the company name with Jacob before it goes on a slide; notes say "GNOME," likely Knowm.] Put those two together and you get the idea underneath Bittensor: **money is energy that flows through a network, and where it flows, structure grows.** Point the flow at the hardest problem in the world and see what grows. The hardest problem in the world is the production of intelligence. So: a new substrate. Not a company. Not a state. A network where incentives flow like energy, transparent, permissionless, open to anyone, and where the structure that grows is intelligence that belongs to the people who grew it. A new form in which to develop AI, and to develop it inside the larger human biome rather than above it. ## 8. What Bittensor is, in one breath Bittensor is a blockchain that hosts many markets. Each market, we call it a subnet, is a Bitcoin for one commodity. Compute. Storage. Data. Inference. Model training. Coding agents. In each subnet, miners produce the commodity. Validators judge the quality. The chain runs one mechanism, Yuma Consensus, that turns the validators' judgments into payment, and punishes any validator who strays from what the stake-weighted majority sees. The whitepaper shows a colluding group with less than half the stake decays toward zero. (Whitepaper, 2022.) Since February 2025 each subnet has its own token, and the market price of that token decides how much of the new TAO that subnet earns. No committee allocates. The market does. (dTAO, Feb 2025.) Everything sits under one token, TAO. Twenty-one million, ever. Same schedule as Bitcoin. The first halving happened on 15 December 2025. Issuance is now about 3,600 TAO a day. (Taostats, 2026.) Say it in a sentence: Bittensor is Bitcoin's mining machine pointed at intelligence instead of at hashes, with the market deciding what counts as intelligence. ## 9. Fair launch. No pre-mine. Permissionless. This matters more than any feature, so say it plainly and let it sit. There was no pre-mine. No tokens were printed and handed to insiders. No venture round bought a slice before the public could. Nobody was given anything. Every TAO that exists was mined from block one by someone who ran a machine. I mined mine the same way you could have. That is why this community is what it is. People who earned their position defend it differently from people who were allocated it. And it is still true today. Anyone in this room can mine on a subnet tonight. Your agents can mine for you. Anyone can run a validator. Anyone can register a subnet and write their own market for their own commodity. Nobody can stop you and nobody has to approve you. Bring it back to section 5. Korinek asks: how do ordinary people come to own a piece of AI capital? His answer is a government program. Our answer is: the capital is issued directly to whoever contributes, under rules anyone can read, and the only permission you need is a wallet. Call it permissionless basic capital. That is my phrase, not Korinek's. Say so. ## 10. The third path Now the geopolitical turn. This is the novel part of the talk. Make it clean. Every AI lab that matters is inside one of two power structures. American or Chinese. Their models answer to those governments. Their capital sits in those jurisdictions. Their weights can be export-controlled, their outputs can be shaped, their access can be switched off, by a phone call from a capital city. Bittensor is in neither. It is not tied to the United States or its government. It is not tied to China. It lives in a bottom-up, grassroots, international network of people who run machines. It is disjoint from those power structures by construction. We are in Montreal, in a country whose AI researchers built much of this field and whose companies now buy it back from two superpowers. This room knows what it means to be downstream of somebody else's platform. So here is the third path. Not the American AI. Not the Chinese AI. An AI that is owned by the people who build it, wherever they are, that no government can shut off, and that can grow to a scale that competes with both. Bitcoin showed that scale is reachable. We are building the version that produces something useful. Many people talk about AI. Nobody else brings digital currency into the conversation, and digital currency is the only tool that has ever coordinated this much machinery without a boss. That is our unique position. Own it. ## 11. What has been built Now show the work. Keep it to a handful of facts. All [VERIFY]. - About 128 subnets live, up from about 32 at the start of 2025. (Taostats; Grayscale Research.) - Real products with real customers: Chutes serves inference at scale and is the only Bittensor-linked provider on OpenRouter. Targon runs confidential inference inside trusted hardware with Intel. Lium sells GPU compute. Twenty-four subnets bill outside customers for an estimated $28 to $35 million a year, and fourteen of them buy back their own tokens with it. Customers include PwC France and Dropbox. (SubConnect Revenue Index, Sep 2026.) - Nearly half of all TAO is staked to the root network. The root dividend stream has paid out about $198 million since February 2025. (bittensor.com V441 release, Jul 2026.) - Wall Street arrived without being invited. Grayscale's Bittensor Trust trades on OTCQX and has a spot ETF filing under SEC review. Bitwise filed too. Three public companies, including one on the TSX Venture Exchange, hold TAO as their treasury. (SEC filings, 2025–2026.) - The first halving is behind us. The schedule held. Nobody could change it. Frame: five years ago this was a whitepaper and a handful of miners. Today it is the largest decentralized AI network in the world, by a wide margin, and the only one Wall Street has tried to package. Then the honesty about scale, before someone else says it: our market cap is two to three billion dollars. NVIDIA's is about five trillion. That is a gap of roughly two thousand to one. We are a proof of mechanism, not yet a counterweight. I would rather say it than have you think I do not know it. ## 12. The honest part: I hold a key Do not bury this. Put it near the end, where it carries the most weight, and say it in the plainest words in the whole talk. Bittensor is not decentralized in the way Bitcoin is. I wrote that publicly in June and I will say it here. It can be, and it will be. But it is not yet. Today the chain's blocks are produced by nodes the foundation runs. That is proof of authority. There is a privileged key that can upgrade or pause the chain. A small group holds it, and it runs through me. We used it in 2024 to halt the chain within thirty-five minutes of a wallet-drain attack, and that saved people money. We could use it badly. We have been accused of using it badly. Some of you were in the Templar community in April and lost money, and I am sorry for that. Everything else about Bittensor is permissionless today. Mining, validating, staking, building a subnet. Nobody has ever needed my permission for any of that. But the rules of the game can still be changed by a key, and a network whose rules can be changed by a key is not finished. We chose this deliberately. We chose speed of iteration over immediate decentralization while we found the right economics. That was the right call and it is now the wrong call. So here is the path, with dates: - Nominated proof of stake, where token holders choose who produces blocks, within about a year. [VERIFY current target] - On-chain governance that replaces the key: proposals reviewed by elected validators and subnet builders who can delay, cancel, or fast-track, and a final vote to disable the sudo key entirely. Proposed in May. [VERIFY status] - Conviction rights for token holders, weighted by how long you lock. Shipping now. - Full decentralization around December 2027. Then ossification. The rules freeze. Nobody, including me, can change them. Now the turn that makes this a strength, not an apology: **A centralized AI lab can never make this promise.** OpenAI cannot decentralize. Anthropic cannot ossify. Their whole structure is the key. Our key is a bug we are removing. Theirs is the product. The difference between Bittensor and every centralized AI company is not that we are already decentralized. It is that we can be, we are on a dated path to be, and they cannot. ## 13. The call Three audiences. One ask each. Short. **If you hold TAO:** you own a piece of the only AI capital on Earth that was never allocated to anyone. Every TAO was earned. You are not a customer of this network. You are an owner of it, in the same way I am. Stake it to validators who allocate well. Watch what they curate. Hold it through the year it takes to remove the key, because the day the key is gone is the day the asset becomes what it was designed to be. **If you build:** there are about 128 markets and there is room for yours. Find the thing that is slow, expensive, permissioned, or closed, and write a market for it. The tools ship every month. Registration cost is falling toward the price of a transaction. The subnets with real customers are the ones that survive, so build for a customer, not for emissions. If you have a model, a dataset, a GPU, or an agent, you can be paid for it tonight, in ownership, not in wages. **If you are new:** you have just heard the shape of the next twenty years. Capital is concentrating into a few labs in two countries, and the economists' only answer is to hope a government fixes it. You do not have to wait for that. You can join a network that pays you in ownership for contributing to the most valuable thing humans are building. It is open. It has no gate. Come and see if it is real. ## 14. Close Return to the first beat. The fear you feel is correct. But aim it correctly. Do not fear the machine. Fear the machine owned by a few. Then give the alternative in one line: Bitcoin proved that a network with no owner can grow larger than any company. Bittensor is that network, pointed at intelligence, paying ownership to anyone who helps build it. It is the machine that pays you back. We are approaching a fork. Down one road, a station in orbit and everyone else below. Down the other, an intelligence that grows the way a forest grows, owned by everyone who tends it. We have the people. We have the design. We have the proof from Bitcoin that it can scale. What we do not have is time to waste. Build it with me. --- ## Open questions and weak points for Jacob to decide **Framing decisions** 1. **The Gaza image.** Your notes pair Elysium with "most people live in a form of Gaza." That line will be the headline of the talk and the only thing some people remember. Elysium alone carries the point. Decide deliberately. 2. **"Permissionless basic capital."** New phrase that ties you to Korinek's "universal basic capital." Strong hook, but Korinek has not endorsed token networks. Keep the "my phrase, not his" disclaimer or drop the phrase. 3. **How hard to lean on the US-vs-China frame in Canada.** It lands well in Montreal. It may read as anti-American to US investors and press in the room. Consider "outside both" rather than "against both." 4. **Covenant AI.** Options: (a) name it and apologize to those who lost money, as drafted; (b) refer only to "accusations in April"; (c) omit. Omitting risks a hostile Q&A. Naming it gives you the frame. 5. **"It runs through me."** Drafted as the plainest admission. Confirm you want to say it in the first person on a recorded stage. Alternative: "a small group, and I am in it." **Facts to confirm before they go on a slide** 6. Neuromorphic company name (Knowm / Alex Nugent?). Transcript says "GNOME." 7. Nominated proof-of-stake target date. Blockworks says "within approximately the next year" as of mid-2026. Give a month if you can. 8. Status of the governance pallet and the sudo-disable vote. Proposed May 2026. Has anything shipped? 9. Whether "128 subnets" or "256 slots" is correct. Secondary press says Robin τ raised the cap on 1 May 2026; the V440 release note says the cap raise was deferred. 10. Subnet revenue: SubConnect says $28–35M ARR (Sep 2026); another outlet claims $43M in Q1 alone. Pick one source and name it. 11. Market cap on the day. Trackers disagree by almost $1B because of circulating-supply methodology. 12. Exact conference name and whether the talk is recorded. **Weak points a hostile questioner will raise** 13. **Stake concentration.** Top 10 validators hold about 65% of root voting power (Apr 2026). Critics cite a stake Gini near 0.98. "Fair launch" does not mean "fair distribution today." Have a number and a plan (Root Reborn caps, conviction voting) ready. 14. **Emissions versus revenue.** The network mints roughly $800k a day at current prices and earns $28–35M a year from outside customers. Most subnet value is still emission-funded. The honest line: real, growing, and ten percent of the way there. 15. **Bittensor pays for contribution, not need.** A displaced 55-year-old paralegal cannot mine a subnet. The talk claims to answer "who owns AI capital," not "how does a person with nothing to contribute eat." Say that limit out loud or expect it as the first question. 16. **You sold alpha during the Covenant dispute.** You said it was under 1% of your position and that trading is what dTAO is. Critics call it a conflict regardless. Decide whether to pre-empt. 17. **Three emission models in sixteen months.** Builders say the rules move under them. Your answer is "iteration before ossification." Make sure the audience hears the ossification date every time you say iteration. 18. **Security.** A ~$8M wallet drain in 2024; critics say no public third-party audit and no bug bounty. Confirm the current audit status before the talk. 19. **The Bitcoin claim.** Hashrate is not FLOPs. The draft avoids the "out-computes Google" line. Do not reintroduce it. 20. **The "100% of capital" line from your notes.** Measured facts are 53% of VC, ~35% of the S&P 500, $720–745B of capex from four firms. Say "a record share," not "almost all." 21. **Labor-market data is mixed.** Yale Budget Lab (Oct 2025) and a Danish study (Mar 2026) find no economy-wide disruption yet. Your claim is about direction and destination, not this quarter. The 22–25-year-old canaries (19% below trend, Jun 2026) are your best early-evidence number. 22. **Scale.** $2–3B versus $5T. The draft says it first. Keep it that way.