The United States Department of Justice walked into a courtroom and told the judge that the law itself might be the problem. Not the defendant. Not the plaintiff. The law. In a filing supporting OpenAI in the ongoing copyright litigation that has pitted some of the world's most prominent authors against the world's most prominent AI lab, the DOJ made a startling argument: restrict AI training data, and you don't just slow down a company โ you damage American prosperity itself.
Let me sit with that for a moment, because it's doing a lot of heavy lifting.
The DOJ isn't saying OpenAI didn't use the authors' books. It's saying that even if they did, the use should be protected. That the greater good โ American economic dominance in artificial intelligence โ outweighs the property claims of individual creators. That's not a legal argument. That's a national industrial policy dressed in a legal brief.
I've been watching this industry long enough to know when a narrative shifts. I audited 40+ ICO whitepapers in 2017 with Python simulations when everyone else was chasing hype. I watched DeFi Summer's liquidity fairy tale unfold in real time from Berlin. I covered the Beeple auction and asked who actually owns the soul of crypto art. And I've learned that the most dangerous moments in this industry aren't the crashes โ they're the moments when the rules get rewritten quietly, in legal filings that most people never read.
This is one of those moments.
The Context: A Three-Year Legal War
Let me back up and give you the battlefield map, because this didn't happen in a vacuum.
Since 2023, a wave of copyright lawsuits has crashed against the AI industry's shores. The New York Times sued OpenAI and Microsoft, alleging that millions of its articles were used to train GPT models without permission. A coalition of authors โ including George R.R. Martin, John Grisham, and Jodi Picoult โ filed suit, claiming their books were ingested into training corpora. Getty Images sued Stability AI over its training data. The list goes on.
The core legal question is deceptively simple: does the "fair use" doctrine โ which allows limited use of copyrighted material without permission for purposes like commentary, criticism, and research โ extend to training large language models on massive scrapes of the internet?
The AI companies say yes. The training is "transformative," they argue. The models don't reproduce the works; they learn patterns from them. The authors say no. The models are competing directly with the works they were trained on, and the scale of the copying โ billions of words, entire libraries โ is anything but "fair."
For three years, this question has hung over the AI industry like a sword. Every funding round, every enterprise contract, every partnership negotiation has had to account for the possibility that a court might rule that training on copyrighted data without permission is infringement. That's not a hypothetical risk. That's a line item in every AI company's risk assessment.
And now the DOJ has stepped in and said: the sword shouldn't fall.
The Core: What the DOJ Actually Argued
The DOJ's filing โ made in the context of the authors' lawsuit against OpenAI โ argues that restricting AI training data would "undermine American prosperity." The government's position is that AI models need access to vast amounts of text to remain competitive, and that imposing copyright liability on training data would cripple American AI leadership.
Let me break down what this actually means across the dimensions that matter.
The Technical Reality: Data Is the New Oil, and the Well Is About to Run Dry
Here's the thing that most people don't understand about how large language models actually work. The current paradigm โ the one that OpenAI, Anthropic, Google, and Meta all follow โ is built on a simple equation: more data plus more compute equals better models. This is the scaling law, and it has held remarkably steady since it was first articulated in 2020.
The models are trained on enormous corpora of text scraped from the internet โ books, articles, forums, code repositories, academic papers. The size of these corpora is staggering. GPT-4 was trained on an estimated 13 trillion tokens. That's roughly 10 trillion words. To put that in perspective, the entire Library of Congress contains about 170 million items. We're talking about training models on the equivalent of tens of thousands of Libraries of Congress.
The scaling law says that as you increase both data and parameters, model performance improves predictably. But here's the dirty secret: we're running out of high-quality text. Researchers have estimated that the stock of publicly available, high-quality text on the internet will be exhausted by 2026 or 2027. We're approaching the data wall.
This is why the copyright question is so existential. If AI companies can't use copyrighted material without paying for it, the cost of training data becomes prohibitive. If they can't use it at all, they hit the data wall even faster. The DOJ's position is essentially: let the models eat, or America loses the AI race.
Based on my experience auditing tokenomics and data claims in the crypto world, I can tell you that the data quality question is always more complex than it appears. In 2017, I ran simulations on ICO whitepapers and found that most of them had fundamentally broken token models. The same kind of scrutiny applies here. The question isn't just whether AI companies can legally use the data โ it's whether the data they're using is actually making the models better, or just bigger.
There's a deeper technical issue that gets lost in the legal debate. The scaling law isn't just about quantity โ it's about diversity. Models trained on homogeneous data plateau quickly. The reason web-scale training works so well is that the internet contains an incredible diversity of writing styles, domains, languages, and perspectives. Restricting access to copyrighted material would disproportionately remove the highest-quality text from training corpora โ published books, professionally edited journalism, academic research. What would remain is the long tail of the internet: blog posts, forum comments, social media. And that's a fundamentally different โ and worse โ training signal.
I've seen this dynamic play out in crypto markets. When liquidity gets fragmented across dozens of Layer2 chains, each individual pool becomes thinner and less efficient. The same thing happens with data. If you restrict access to the highest-quality sources, you don't just lose volume โ you lose the diversity that makes the whole system work. The DOJ's position, whatever its legal merits, is technically coherent: it preserves the data diversity that the current paradigm depends on.
The Commercial Calculus: Legal Risk as a Cost Center
Let me talk about money, because that's where this gets real.
OpenAI's business model is built on a simple premise: train the best models, sell access to them. The company's API, its ChatGPT subscriptions, its enterprise agreements โ all of these revenue streams depend on the ability to iterate quickly. And iteration depends on data.
The copyright lawsuits represent a massive financial uncertainty. If OpenAI loses, it could face billions in damages. More importantly, it would have to fundamentally change how it trains models โ either by licensing data (which would be enormously expensive) or by relying on synthetic data (which is still inferior for many tasks).
The DOJ's support doesn't resolve the lawsuits. Courts are not bound by the DOJ's opinions. But it signals something important: the executive branch of the United States government is on OpenAI's side. That matters for investor confidence. It matters for enterprise customers who are worried about legal exposure. It matters for the company's valuation.
I've seen this dynamic before. In 2020, during DeFi Summer, I watched projects with questionable fundamentals raise millions based on narrative momentum alone. The market was pricing in optimism, not reality. The same thing is happening here. The DOJ's filing is being priced into OpenAI's valuation as a reduction in legal risk โ even though the actual legal outcome is far from certain.
There's also a cost structure angle that doesn't get enough attention. If the DOJ's position prevails, OpenAI avoids the need to pay licensing fees for training data. That's not a trivial saving. The licensing market for high-quality text is enormous โ publishers charge millions for access to their archives. If AI companies can use this material under fair use, they capture that value for free. That's a direct improvement to their unit economics.
But here's the counterpoint that keeps me honest: the legal uncertainty doesn't disappear just because the DOJ has an opinion. Courts are independent. Juries are unpredictable. And the content industry has deep pockets and strong incentives to fight. The DOJ's filing is a signal, not a settlement. The actual legal risk remains substantial.
The Industry Ripple: A New Power Dynamic
This goes far beyond OpenAI. The DOJ's position, if adopted by courts, would create a "safe harbor" for the entire AI industry. Every AI company training on web-scraped data would benefit. The compliance costs that have been building up โ the data provenance tracking, the licensing negotiations, the legal teams โ could be dramatically reduced.
But here's the flip side: the content industry is not going to take this lying down.
Publishers, authors, artists, and news organizations have been watching their work get absorbed into training corpora with growing alarm. The DOJ's position tells them that the government doesn't care about their property claims โ that their work is, in effect, a public resource to be mined for the greater good of American AI dominance.
This is going to create a backlash. We're already seeing it. The Authors Guild has been lobbying Congress for legislation that would require AI companies to license training data. News organizations are exploring collective licensing arrangements. The creative community is organizing.
And here's the uncomfortable truth: the DOJ's position might actually be counterproductive for the AI industry in the long run. By picking a side so explicitly, the government has politicized the copyright question. It has turned a legal dispute into a cultural war. And in a democracy, cultural wars tend to produce legislation โ and legislation tends to be more restrictive than court rulings.
There's also a structural shift happening in the content industry itself. The traditional gatekeepers โ publishers, record labels, film studios โ are being disintermediated by AI. The DOJ's position accelerates this process. But it also creates an opening for new business models. I've been watching the emergence of "data licensing" as a category โ companies like Reddit and Shutterstock have signed deals with AI companies to license their data. These deals are the beginning of a new market, and the DOJ's position doesn't eliminate that market โ it just changes the bargaining power of the parties.
The Competitive Dimension: America vs. Everyone Else
Let me zoom out to the global picture, because this is where the DOJ's filing gets really interesting.
The United States is in an AI arms race with China. The Biden administration and now the Trump administration have both made clear that maintaining American leadership in AI is a national priority. Export controls on advanced chips, restrictions on Chinese access to American AI technology, massive investments in domestic compute infrastructure โ the policy toolkit has been deployed aggressively.
The DOJ's position on training data is part of this same playbook. If American AI companies can freely use the world's copyrighted text, they maintain a data advantage over Chinese competitors, who face their own restrictions on data access. The DOJ is essentially arguing: we can't afford to handicap our own companies with copyright liability when our geopolitical rivals don't have the same constraints.
This is a coherent strategic argument. It's also a dangerous one. Because it frames copyright protection as a luxury that America can't afford โ which is exactly the kind of framing that content creators will resist.
There's also a domestic competitive dimension. The DOJ's support for OpenAI specifically โ not for the AI industry generally โ creates an uneven playing field. OpenAI is the most visible beneficiary. Its competitors, including Anthropic and Google DeepMind, will also benefit from any favorable ruling, but they don't get the same explicit government endorsement. And open-source models, which have been the counterweight to OpenAI's closed approach, may find themselves at a disadvantage if the regulatory environment becomes more permissive for closed commercial models.
I've seen this dynamic play out in crypto. When the SEC started signaling favoritism toward certain projects โ or hostility toward others โ it distorted the market in ways that had nothing to do with technical merit. Government signals become market signals, and market signals become capital allocation decisions. The same thing is happening here.
The international dimension adds another layer of complexity. The European Union has taken a very different approach to AI training data, with the AI Act requiring transparency about training data sources and imposing stricter obligations on foundation models. If the US establishes a permissive regime while the EU maintains a restrictive one, AI companies will face divergent compliance requirements across jurisdictions. That's not just a legal headache โ it's a competitive disadvantage for companies operating globally.
The Ethical Tension: Who Pays for the Data?
Let me talk about the ethical dimension, because this is where I struggle.
The copyright question is not just about money. It's about authorship. It's about the right of creators to control how their work is used. It's about the dignity of labor โ the idea that if your words, your art, your research, your reporting are valuable enough to train a model, you should have some say in that transaction.
The DOJ's position essentially says: your work is a public resource. The greater good of American AI dominance outweighs your individual property claims. That's a hard pill to swallow for the thousands of writers, artists, and journalists whose work has been absorbed into training corpora without consent or compensation.
I think about this a lot. I've spent my career writing about blockchain and crypto โ creating content that is, in turn, scraped into training datasets. My words are probably in some model's training corpus right now. And I have complicated feelings about that.
On one hand, I believe in the transformative potential of AI. I've seen what these models can do โ the creative possibilities, the analytical power, the ability to democratize access to knowledge. On the other hand, I believe in the value of human creativity. I believe that writers deserve to be paid for their work. I believe that the labor of creating โ whether it's a novel, a news article, or a research paper โ has intrinsic worth that shouldn't be erased by a legal doctrine.
The DOJ's position doesn't resolve this tension. It just picks a side. And the side it picks is the side of the machine.
There's also a privacy dimension that gets overlooked in the copyright debate. Training data isn't just books and articles โ it's also personal information, private communications, and sensitive content that was never intended to be part of a training corpus. The DOJ's position, by making it easier to use data without permission, potentially makes it easier to use personal data without consent. That's a separate legal issue โ privacy law is distinct from copyright law โ but the policy direction is the same: data should flow freely, and individual rights should yield to the greater good.
The Investment Angle: What This Means for Valuations
Let me talk about what this means for investors, because that's where the rubber meets the road.
The legal risk has been a persistent discount factor in AI company valuations. Every time a new copyright lawsuit is filed, you can see the anxiety ripple through the market. The DOJ's filing reduces that anxiety โ at least temporarily.
For OpenAI specifically, this is a meaningful boost. The company has been in fundraising talks at valuations that have been reported at $300 billion or more. A favorable legal environment supports that valuation. It reduces the risk that the company will have to pay billions in damages or fundamentally change its training approach.
But here's the thing I've learned from watching crypto markets for a decade: government signals are not the same as legal outcomes. The DOJ's opinion is just that โ an opinion. Courts are not bound by it. And the backlash from the content industry could easily produce legislation that overrides whatever the courts decide.
I've seen this pattern before. In 2017, when the SEC issued its famous DAO report, the market initially treated it as a death knell for ICOs. Then the market recovered. Then it crashed again. The point is: regulatory signals create volatility, not certainty. And volatility is not the same as direction.
There's also a broader market implication. If the DOJ's position prevails, it could trigger a wave of consolidation in the AI industry. Companies with the most data access and the most compute would be best positioned to benefit. Smaller players, who might have relied on licensing deals or alternative data sources, could be squeezed out. That's a concentration risk that investors should be watching.
The Infrastructure Angle: The Data-Compute Nexus
There's a less obvious dimension to this that I want to flag: the relationship between data policy and compute infrastructure.
If the DOJ's position prevails and AI companies can continue to train on vast amounts of copyrighted data, the demand for compute will continue to grow. More data means bigger models, which means more GPUs, more data centers, more energy. The entire AI infrastructure buildout โ the $500 billion Stargate project, the massive data center construction boom, the global GPU shortage โ is predicated on the assumption that the data pipeline will keep flowing.
If the copyright restrictions were to tighten, the economics of the compute buildout would change. AI companies might shift toward more data-efficient approaches โ synthetic data, retrieval-augmented generation, smaller specialized models. That would reduce the demand for massive training clusters and change the economics of the entire infrastructure layer.
The DOJ's position, by keeping the data pipeline open, indirectly supports the compute buildout. It's a signal to infrastructure investors that the current paradigm โ data-hungry, compute-intensive training โ will continue. That's a meaningful signal, even if it's not the one that gets the headlines.
But there's a wrinkle here. Even if copyright restrictions are relaxed, the data wall is still approaching. The internet's supply of high-quality text is finite. At some point, AI companies will need to find new sources of training data โ synthetic data, licensed data, user-generated data. The DOJ's position buys time, but it doesn't solve the underlying problem. The infrastructure buildout is betting on a paradigm that has a built-in expiration date.
The Contrarian Angle: The Government's Embrace Is a Trap
Now let me give you the contrarian take, because I think there's a real one here.
The DOJ's support for OpenAI might be the worst thing that could happen to the company. Here's why.
First, it politicizes OpenAI. The company has spent years trying to position itself as a responsible AI leader โ working with regulators, publishing safety research, advocating for oversight. The DOJ's endorsement makes OpenAI look like a government favorite, a corporate champion backed by the state. That's a liability. It invites scrutiny. It invites resentment. It makes the company a target for every critic who believes that AI is a tool of corporate power.
Second, it invites legislative backlash. The DOJ's position is so explicit, so one-sided, that it practically begs Congress to respond. Content creators are already lobbying for legislation that would require AI companies to license training data. The DOJ's filing gives them ammunition: "Look, the executive branch is siding with Big Tech against creators. We need Congress to step in and protect us." The result could be legislation that is far more restrictive than anything a court would have imposed.
Third, it creates a moral hazard. If AI companies believe they have the government's blessing to use copyrighted data without compensation, they have less incentive to build legitimate licensing arrangements. That's a long-term problem. The content industry is not going to disappear. Publishers, authors, and artists will continue to create โ and they will continue to demand compensation. The AI industry needs to find a sustainable relationship with the creative community, and the DOJ's position makes that harder, not easier.
I've seen this dynamic play out in crypto. When the government signaled support for certain projects or approaches, it often created a bubble of false confidence. The projects that survived were the ones that built real value, not the ones that relied on government favor. The same will be true here.
There's also a deeper structural concern. The DOJ's position treats copyright as an obstacle to innovation. But copyright is also a mechanism for funding innovation. Writers need to eat. Journalists need salaries. Researchers need grants. If the economic foundation of content creation is undermined, the supply of high-quality training data will eventually dry up โ not because of legal restrictions, but because the creators will have moved on to other work. The DOJ's position might be winning the battle while losing the war.
The Takeaway: The Rules Are Being Rewritten, and Nobody's Watching
Here's where I land.
The DOJ's filing in support of OpenAI is not a legal footnote. It's a signal that the United States government has made a strategic choice: American AI dominance matters more than copyright protection. That choice will shape the AI industry for years to come โ not just in the United States, but globally.
But here's the thing that keeps me up at night: this choice is being made in legal filings and policy memos, not in public debate. Most people have no idea that the DOJ has taken this position. Most people have no idea that the copyright question is the single most important unresolved issue in AI. And by the time they understand what's happening, the rules will already be written.
I've spent my career watching the blockchain industry rewrite the rules of finance and ownership. I've seen what happens when technology outpaces law. I've seen the chaos, the opportunity, and the collateral damage. The AI copyright question is the same story, playing out at a much larger scale.
The question isn't whether the DOJ is right or wrong. The question is whether we โ as a society โ are going to have a real conversation about who owns the data that powers the machines, and who gets to benefit from that data. The DOJ has made its choice. The courts will make theirs. Congress will make theirs. And the rest of us will live with the consequences.
Where the code meets the chaotic human heart, the copyright question is the sharpest edge. It's the place where the cold logic of scaling laws collides with the warm reality of human creativity. And right now, the cold logic is winning.
But here's the thing about ledgers: they can always be rewritten. The question is who gets to do the rewriting, and whose stories get told in the new version.
Rewriting the ledger, one story at a time. That's what I do. And this story โ the story of who owns the data that powers the machines โ is the most important one I've covered yet.
The next chapter is being written right now, in courtrooms and legislative chambers and corporate boardrooms. The question is whether the rest of us will be paying attention when it's published.
The math doesn't lie, but it also doesn't tell the whole story. And the story here is still being written.
