The Anti-Distillation Trap: Why AI's Real War Is About Who Owns the Output

Over the past 72 hours, I've been digging through CITIC Securities' latest research on the AI sector correction, and something kept nagging at me. The report is being circulated as a sober analysis of why tech stocks are bleeding. But strip away the institutional language, and you'll find a confession: the market has finally realized that AI's pricing power isn't about who has the smartest model. It's about who controls the data pipeline. And buried in the footnotes of this analysis is a term that should terrify anyone who believes in open innovation: anti-distillation. Let me rewind. For the past year, the crypto community has been watching the AI industry with a mix of envy and horror. Envy because AI companies have no problem raising capital. Horror because the centralization happening in AI makes even the most captured Layer 2 sequencer look like a model of decentralization. We've spent years warning about the dangers of concentrated power in blockchain networks. Meanwhile, the AI industry has built a system where a handful of companies control the compute, the models, and now, potentially, the very data that allows anyone else to compete. CITIC's report identifies three pricing variables for AI stocks: commercialization pace, compute conversion efficiency, and model gap evolution. All three are valid. But the report's real contribution is naming anti-distillation as the "largest potential variable." This is the moment where the AI industry's centralization problem becomes explicit. Anti-distillation is the practice of preventing competitors from using your model's outputs to train their own models. It's the digital equivalent of a walled city. And it's being framed as a legitimate business strategy. Here's what the report gets right: the AI industry has shifted from a single-dimensional competition over model capability to a multi-dimensional war over compute, commercialization, and ecosystem lock-in. OpenAI's partnership with Microsoft, Anthropic's compute deal with Amazon, Google's full-stack advantage. These aren't just business arrangements. They're moats. And the report correctly notes that the window for catching up is closing faster than most market participants expect. But here's where my analysis diverges from the institutional consensus. The report treats anti-distillation as a technical problem. It's not. It's a philosophical one. And it's a problem that the blockchain community has been wrestling with for years. We call it the oracle problem. The verifiability problem. The trust problem. When you rely on a centralized entity to provide you with data, you're not building on open infrastructure. You're renting access to someone else's garden. Let me give you a concrete example from my own experience. In 2022, I audited a series of failed DeFi protocols. The pattern was always the same: beautiful smart contracts, decentralized governance, and then a single admin key that could drain everything. The code was open. The execution was closed. The AI industry is heading toward the same fate. The models are open in the sense that you can query them. But the training data, the compute infrastructure, and now the output rights are all controlled by a handful of entities. Anti-distillation is the admin key of the AI world. The report's analysis of compute as a strategic asset is solid. It notes that compute-related spending now accounts for over 70% of capital expenditures at leading AI companies. This is the equivalent of the oil economy. But the report misses a crucial point: compute without open access is just another form of feudalism. The reason Bitcoin succeeded where so many digital currencies failed isn't because it had the best technology. It's because anyone could participate. The reason Ethereum became the settlement layer for DeFi isn't because it was the fastest. It's because the rules were transparent and the barriers to entry were low. Now, let's talk about the contrarian angle. The report assumes that anti-distillation, if successful, will cement the advantages of incumbents. I'm not so sure. In fact, I think anti-distillation could be the catalyst that breaks the AI oligopoly. Here's why: every attempt to lock down information flows creates an incentive for alternative systems to emerge. The open-source community has already demonstrated this with models like Llama and Qwen. They're not as capable as the frontier models, but they're improving. And they're improving because the open-source community is doing what it always does: finding creative workarounds. There's a deeper issue here that the report touches on but doesn't fully explore. The report mentions that the model gap has narrowed from "generational differences" to "intra-generational differences." GPT-4 to GPT-4o was a smaller leap than GPT-3 to GPT-4. This is the commoditization curve. And it's the same curve we saw in blockchain infrastructure. The first movers build the infrastructure, but the value eventually migrates to the application layer. The question is whether anti-distillation can prevent this migration. My bet is that it can't. Not because the technology won't work, but because the incentives are wrong. Let me be specific about what I mean. The report identifies three risks: commercialization shortfalls, anti-distillation leading to industry consolidation, and compute supply chain constraints. All three are real. But the report misses the fourth risk: the risk that anti-distillation becomes so aggressive that it triggers regulatory intervention. The EU AI Act is already moving toward requiring transparency in training data. If anti-distillation becomes standard practice, it's only a matter of time before regulators step in. And when they do, the incumbents who built their moats on closed systems will find themselves on the wrong side of the law. I've been thinking about this in the context of my own work. I run a Web3 community in Buenos Aires, and I've seen firsthand how the promise of decentralization can be co-opted by centralized interests. The same thing is happening in AI. The narrative is about democratizing intelligence. The reality is about consolidating power. And the market is starting to price this in. The report's shift from macro factors to industry fundamentals is a recognition that AI stocks are no longer a bet on technology. They're a bet on governance. Here's what I think the market is missing. The report frames anti-distillation as a variable that will determine the competitive landscape. But it's actually a variable that will determine the legitimacy of the entire AI industry. If the industry becomes a closed system where a few companies control the means of intelligence production, it will face the same backlash that Big Tech faced in the 2010s. The difference is that this time, the stakes are higher. We're not just talking about social media algorithms. We're talking about the infrastructure of human knowledge. The report's recommendation to avoid "excessive grand narratives" is telling. It's an admission that the market has been pricing in AGI timelines and productivity revolutions that may not materialize. This is the same pattern we saw in crypto during the ICO boom. The narratives were beautiful. The execution was lacking. And when the market realized that most projects were just PowerPoint presentations, the correction was brutal. The AI industry is heading for a similar reckoning. Not because the technology isn't real, but because the business models haven't been validated. Let me offer a different framework. Instead of asking whether anti-distillation will succeed, we should ask what kind of AI ecosystem we want to build. The blockchain community has spent years developing solutions for verifiable computation, decentralized identity, and transparent governance. These tools are directly applicable to the AI industry's centralization problem. We can build AI systems where model outputs are verifiable, where training data is auditable, and where the value created by user interactions is distributed to the users themselves. This isn't a pipe dream. It's the natural evolution of both industries. The report's analysis of the K-shaped divergence is also worth examining. It suggests that a weaker dollar and reduced rate hike expectations could trigger capital rotation from US AI leaders to other markets, including China. This is a trading signal, but it's also a reflection of a deeper truth: the AI industry's center of gravity is shifting. The compute constraints in China are real, but they're also forcing innovation. Chinese AI companies are developing more efficient algorithms, better quantization techniques, and alternative chip architectures. Necessity is driving creativity. And that creativity will eventually find its way into the open-source ecosystem. I want to end with a question that the report doesn't ask. What happens when the AI industry's centralization problem becomes so acute that it threatens the industry's own legitimacy? The blockchain community has an answer. We've been building systems that don't require trust in centralized entities. We've been building systems where power is distributed by design, not by convenience. The AI industry can learn from this. Or it can repeat our mistakes. The choice is theirs. But the market is already voting. And the market is telling us that the era of paying for imagination is over. We're entering the era of paying for execution. And execution, in the AI world, means building systems that are open, verifiable, and accountable. Freedom isn't a feature. It's the foundation. And the only moat that matters is the one built by our shared vision.