I used to think AI was the one technology that could escape crypto’s boom-bust cycle. After all, AI had real products—chatbots, code generators, image creators—that people paid for. Crypto, in contrast, often felt like a solution in search of a problem. Then I read a detailed research report from Citic Securities on the recent tech stock adjustment, and I saw the same pattern of expectation versus reality that I had witnessed in the 2021 NFT bubble and the 2022 Terra collapse. The report’s core argument—that AI stock pricing is shifting from a macro-driven narrative to an industry fundamentals-driven one—is a mirror of what crypto has been going through since the 2022 bear market. But the report also revealed something deeper: the AI industry’s current obsession with “anti-distillation” is a direct parallel to crypto’s “code is law” debate. And both are facing the same governance problem: a few entities hold the keys, and the rest of us are left to hope they act in good faith.
Context: The Citic Securities Report Through a Crypto Lens
The Citic Securities report, which I dissected over several nights in my Beijing apartment, identified three key variables for AI stock pricing: the pace and scope of commercialization, the efficiency of converting compute power into market share, and the evolution of the model capability gap. It also flagged “anti-distillation” as the biggest potential variable—the idea that leading AI companies would use technical means to prevent competitors from using their model outputs to train new models. The report’s conclusion was that AI has entered an “expectation verification phase” where the market will pay for execution, not imagination.
For a crypto veteran, this language is painfully familiar. Crypto went through the same transition in 2022. The “tech stack” narrative of 2021—where every Layer 1, every DeFi protocol, every NFT project was valued on future potential—collapsed when the market realized that most of these projects had no sustainable revenue model. The survivors—Aave, Uniswap, MakerDAO—were those that had actual users, actual fees, and actual governance that didn’t rely on a single founder’s whims. The AI industry is now where crypto was in early 2022: full of narratives, short on unit economics.
Core: The Paradox of Centralized AI and the Myth of “Code is Law”
Let me start with the variable that the report calls the “biggest potential variable”: anti-distillation. The report describes it as a technical means to prevent competitors from using a model’s outputs to train new models. The hope is that this will protect the model’s competitive advantage. But based on my experience auditing smart contracts in 2017—when I found 12 critical logic flaws in Gnosis Safe’s multi-signature implementation—I know that any technical protection mechanism is only as strong as the governance around it. Who controls the anti-distillation mechanism? A small team of engineers at OpenAI or Anthropic. Who controls the model weights? The same team. Who controls the ability to change the rules? The same team.
This is exactly the “code is law” problem that crypto has been grappling with for years. In DAO governance, the ideal is that smart contracts enforce rules transparently and immutably. But the reality is that upgrade rights are almost always held by a few multi-sig signers. I’ve seen this in every Layer 2 I’ve audited: the optimistic rollup’s fraud proof system is elegant, but the upgrade key is a single Ethereum address controlled by a foundation. The user trusts the code, but the code trusts the keyholders. Anti-distillation is the same: it’s a technical solution to a governance problem. It will fail if the governance fails.
The compute advantage trap
The report’s second variable—compute power conversion to market share—is equally relevant to crypto. The report argues that having more compute allows faster iteration, lower costs, and better models. But it also warns that compute advantage alone does not guarantee commercial success. Google has the best TPUs, but its AI chatbot Gemini was rushed and had embarrassing errors. Anthropic has more compute than most startups, but its Claude model still struggles with long context tasks.
In crypto, the same pattern holds. Ethereum has the most compute (in terms of validator nodes and developer mindshare), but its Layer 2 solutions like Arbitrum and Optimism have captured more market share in transaction volume. Solana has superior compute architecture but has suffered from repeated outages. The lesson is that compute is a necessary condition, not a sufficient one. The market rewards products, not promises. I remember the 2020 DeFi summer when I watched Compound’s governance token crash wipe out the savings of friends in my Beijing study group. The protocol had strong compute (solidity code, security audits, liquidity), but when the market turned, the lack of real utility—beyond yield farming—was exposed. The same is happening in AI: models with strong compute but weak product-market fit are being punished.
The “anti-distillation” irony
Here is the contrarian angle that the Citic Securities report missed: anti-distillation is a sign of weakness, not strength. If a model is truly superior, it doesn’t need to prevent others from learning from its outputs—it will naturally win through adoption. The fact that leading AI companies are considering anti-distillation suggests that they fear commoditization. This is exactly what happened in crypto with oracles. Chainlink’s early dominance was due to its decentralization and reliability. But when competitors like Pyth and API3 emerged, Chainlink didn’t try to “anti-distill” its data feeds—it improved them. The result was a healthier ecosystem. Anti-distillation is a defensive move, and defensive moves rarely create long-term value.
The K-shaped divergence and crypto’s opportunity
The report discusses a “K-shaped divergence” where AI stocks in the US outperform while others fall behind, and then a potential convergence if the macro environment changes. For crypto, this divergence is already happening. Bitcoin and Ethereum are the dominant assets, but most altcoins are still far below their 2021 highs. The report’s concern about AI’s centralization—that a few companies will control the most powerful models—is actually a bullish signal for crypto. If AI becomes centralized, the need for decentralized alternatives becomes more urgent. Decentralized AI compute networks like Render Network and Akash Network are already filling this gap. Decentralized data storage like Filecoin and Arweave are essential for AI training data provenance. The “anti-distillation” debate is a reminder that the crypto ethos of transparency and permissionless innovation is the antidote to AI’s centralization risks.
Takeaway: Follow the commercialization, not the narrative
If you are a crypto investor, the Citic Securities report is a gift. It tells you that the market is moving from “code is law” to “commercialization is law.” The projects that will survive are those that have real users, real revenue, and real governance that is not controlled by a few multi-sig signers. I saw this firsthand in 2021 when I launched my own NFT project, “On-Chain Diaries,” minting only 50 artifacts that represented our daily interactions with Beijing. I manually coded the smart contract to ensure royalties went to local artists, bypassing large platforms. It was a small, quiet act of resistance against the commodification of creativity. But it taught me that the most important variable in any decentralized system is trust—not trust in the code, but trust in the community. The AI industry is learning the same lesson.
So, what should you do?
Follow the fear, not the chart. The fear in AI is that anti-distillation will lock in the power of a few companies. The fear in crypto is that our own governance models are still too centralized. Both are valid fears. But the solution is the same: build systems that are not just technically sound, but also ethically resilient. If you can’t build a decentralized AI, at least build a decentralized governance for it. The future belongs to those who can combine the soul of code with the heart of community.
If you can, look at the projects that are actually solving the anti-distillation problem—not by closing off their models, but by opening up their governance. That’s where the real value will be. The market is moving from the narrative to the numbers. And the numbers don’t lie.
