The Anti-Distillation Gambit: Why AI Valuation Has Shifted from Narrative to Execution

The numbers didn't lie, but my trust did. For years, I watched the AI sector trade on the promise of what models could become. The GPT-4 launch felt like a seismic event, a moment when the technology narrative justified every multiple. But sitting here in mid-2025, after a decade of auditing code and watching markets digest hype, I see a different game unfolding. The recent CITIC Securities report on tech stock adjustments cuts through the noise with a chilling precision: the market has stopped paying for imagination and started paying for execution. The report identifies three verifiable pricing variables—commercialization pace, compute conversion efficiency, and model gap evolution—and then drops a bombshell that most retail investors have completely missed. They call it the largest potential variable in the entire AI trade: anti-distillation. This is not just a technical footnote. This is the mechanism that could freeze the competitive landscape in place, turning the AI market from a dynamic race into a permanent oligopoly. The numbers didn't lie, but my trust in the market's ability to price this risk rationally has been shaken. To understand why this matters, we have to strip away the macro narrative. For the past eighteen months, every tech stock correction was blamed on rising bond yields. It was a convenient excuse, a way to avoid looking at the underlying rot. The CITIC report does something radical: it shifts the attribution framework from external macro factors to internal industry variables. This is the correct move. The AI sector has entered what I call the expectation validation phase. The market is no longer asking if AI will change the world; it is asking when and at what unit economics. The valuation anchor has switched from technical breakthrough potential to commercial realization metrics. Revenue growth, customer retention, and gross margins are now the only signals that matter. I built a liquidity pool, but lost my liquidity, and I see the same pattern in AI stocks. The liquidity of narrative has drained out of the market, replaced by the cold, hard liquidity of quarterly earnings. The CITIC framework is built on three pillars that deserve deep scrutiny. First, the pace and scope of commercialization must keep pace with market expectations. Second, compute advantages must convert into market share and pricing power. Third, the model gap evolution will determine the competitive hierarchy. Each pillar is verifiable, each is measurable, and each is currently under pressure. Let me dig into the commercialization variable first, because this is where the disconnect between hope and reality is most acute. The report correctly identifies that AI companies are still in the acquire-customers-at-any-cost phase. OpenAI crossed the $4 billion annualized revenue mark, but their inference costs remain stubbornly high. Anthropic is growing revenue impressively, yet their gross margins are compressed. This is the classic sign of a market buying growth rather than profitability. The unit economics have not been validated. I have seen this movie before. In DeFi, we called it liquidity mining. Projects subsidized their TVL numbers with token emissions, and when the incentives stopped, the real users vanished. AI companies are doing the same thing with enterprise discounts and aggressive sales teams. Microsoft Copilot's penetration rate is under dispute. Salesforce's Einstein GPT adoption has been underwhelming. The enterprise AI budget is growing, but the deployment speed is lagging the optimistic projections from twelve months ago. The market's patience window is narrowing. If the top players cannot deliver surprisingly strong commercialization data in the next two to three quarters, the valuation system will shift from price-to-sales multiples to price-to-earnings logic. That shift will trigger a systemic de-rating that most portfolios are not positioned for. The report hints at this but does not quantify the risk. Based on my experience auditing token models and watching liquidity evaporate, I can tell you that the market is already pricing in this shift, and the pain has only just begun. The second pillar, compute conversion efficiency, is where the game theory gets interesting. The report draws a clear causal chain: compute advantage leads to faster model iteration, lower service costs, and more flexible customer response. All of this converts into market share. This is true, but it is not the whole truth. I have watched Google sit on the most powerful compute infrastructure on the planet and still lose the AI commercialization race to OpenAI. Compute is a necessary condition, but it is not sufficient. The conversion of compute into market share requires productization, distribution channels, and service infrastructure. The report understands this, but the market does not. Investors are still treating compute as a moat when it is really just a commodity input. The real moat is what you do with the compute. The CITIC report introduces a concept that should terrify anyone who believes in open innovation: anti-distillation. This is the technical and legal effort by leading model vendors to prevent competitors from using their outputs to train new models. Output watermarking, API usage term restrictions, and legal action against scrapers. If this becomes the industry standard, the catch-up path for smaller AI companies is severed. The era of standing on the shoulders of giants ends. The industry shifts from a diverse ecosystem to a concentrated oligopoly. This is the largest potential variable because it determines whether the model gap becomes permanent. Let me be contrarian here, because this is where most analysis goes wrong. The market is focused on the model capability gap, but the real gap is in reasoning cost and long-context capability. The report notes that the capability gap has narrowed from a generational difference to an intra-generational difference. GPT-4 to GPT-4o was a smaller leap than GPT-3 to GPT-4. But the cost gap is widening. Inference cost per token is diverging. Long-context handling is diverging. This means that even if models become functionally equivalent in benchmarks, the cost structure and capability boundaries will maintain the competitive advantage of the incumbents. This is the silent killer. Most retail investors are looking at model leaderboards and ignoring the cost curves. I see the pattern before the price does, and the pattern here is clear: the market is mispricing the sustainability of the incumbents' cost advantage. The anti-distillation angle adds another layer. If leading vendors can lock up their training data through technical means, they create a data moat that compounds over time. Compute buys you training capability. Data buys you model quality. Anti-distillation protects the data. This creates a positive feedback loop: compute enables model training, models generate user interactions, those interactions become proprietary data, and anti-distillation prevents that data from leaking to competitors. This loop is the ultimate moat, and it is being built right now, silently, through API terms and watermarking techniques. Silence is the loudest audit. The CITIC report's silence on the quantitative metrics is telling. They propose a framework without providing the specific numbers that would validate it. This is the weakness of the analysis. We need to see the unit economics. What is the LTV to CAC ratio for enterprise AI deployments? What is the conversion rate from pilot to full deployment? What is the gross margin trend for the top AI vendors? These are the signals that will determine the direction of the next twelve months. The report also does not adequately address the China question. The concern about anti-distillation is implicitly a concern about the Chinese AI industry, which has relied heavily on open-source models and distillation to catch up. If the distillation path is blocked, and with export controls limiting access to high-end GPUs, the gap between US and Chinese AI capabilities could widen significantly. This is a geopolitical risk that the market is not pricing. The report mentions the K-shaped divergence and the potential for capital rebalancing from US AI leaders to other markets, including A-shares. But this rebalancing is contingent on the fundamentals supporting valuation convergence. Without commercial verification, the capital will not flow. Art burns hot; patience burns colder. The market is in a phase where execution matters more than vision. The top three risks are clear. First, commercialization continues to disappoint, triggering the PS to PE valuation shift. Second, anti-distillation freezes the competitive landscape, accelerating industry concentration. Third, compute supply chain disruptions delay training plans and blow up cost estimates. The opportunities are equally clear. Companies with verifiable commercialization paths will earn a premium in the valuation divergence. Companies that improve compute efficiency through algorithmic innovation will gain a competitive edge in a compute-constrained environment. And the K-shaped divergence creates trading opportunities for those who can time the capital rebalancing. But these opportunities require a level of diligence that most market participants are not prepared for. Flows change, but the current remains. The underlying current is the transition from narrative to execution. This is not a temporary adjustment; it is a permanent regime change. The AI sector is no longer a story stock market. It is a fundamental stock market with technology tailwinds. The metrics that matter are revenue growth, gross margin expansion, and customer retention. The strategies that work are those that can verify progress quarter by quarter. The CITIC report provides the framework, but the market needs the data. I have spent my career auditing code and watching protocols fail when the incentives are misaligned. The AI market has the same structural issue. The incentives have shifted from building the best model to proving the most sustainable business. The market will reward those who can show the numbers, and it will punish those who are still selling dreams. The next two to three quarters will be the tell. Watch the earnings reports. Watch the gross margins. Watch the customer retention rates. And most importantly, watch the API terms and conditions. If anti-distillation becomes standard practice, the window for new entrants closes, and the incumbents solidify their grip. The market is always whispering. You just have to listen. The current is moving toward verification, and those who cannot prove their worth will be swept away.