The contract didn't lie. The market repriced the entire AI sector last quarter, and the market isn't waiting for the next model to be released. It's waiting for the next earnings report. That's the shift CITIC Securities is pointing at, and it's worth parsing because the sector's pricing logic just changed from a narrative to an execution game.
Every bull market, we're told a story. In 2023, it was the AI story. GPT-4 dropped, and the market looked at it like a magic trick. It was the largest technical breakthrough in a decade, so the market simply paid for the story. Anyone with a GPT wrapper and a demo day got a valuation. The market priced in "imagining the future," not "checking the code."
By 2025, the market is no longer reading whitepapers. It's reading income statements. CITIC's report, which is a critical look at the sector's correction, isn't a macro call. It's an industrial diagnosis. It says the market is now anchored on "commercialization progress," not on "technical trajectory." I'm a forensic reader of chains, but this report is a forensic read of a business cycle. The core takeaway is that the market has shifted its valuation anchor from the promise of the model to the execution of the P&L.
Let's break down the actual mechanics. CITIC identifies three pricing variables that have taken over: the speed of commercialization, the efficiency of compute conversion, and the evolution of the model gap. It doesn't just name them. It constructs a causal chain. The chain is: compute advantage -> market share -> model gap. That's the industrial circuit. And it's the only circuit that matters right now.
The Compute Conversion Gap
Compute is the input. It's the cost of goods sold for every AI company. The 70% of capital expenditure for the top firms is compute. That's not a new stat, but the market is now treating it as a bottleneck. But here's the thing that the report hints at, and it's the thing that most retail buyers miss: compute doesn't create value until it's converted into a product. It's a necessary condition, not a sufficient one. You can have the largest GPU cluster on the planet, but if you don't have a distribution channel, you have a data center, not a business.
This is why Google has a compute advantage but hasn't seen the commercial success of its AI. It has the compute, but its market share is not matching. The report calls this a "conversion efficiency" problem. It's a variable that isn't just about the chip. It's about the integration. The company that can convert its compute into a usable, deployable product with a low cost per query will win the market share. The company that just burns compute to train a slightly better model without a revenue-generating deployment is going to be sold off.
The Commercialization Time Lag
The report also highlights the "commercialization rhythm." The market's patience window is closing. OpenAI's annualized revenue is over $4 billion, but the inference cost is still high. Anthropic's revenue is growing, but the gross margin is under pressure. That's a unit economic problem. The market is now in the phase of "paying for market share," and it's not a sustainable model.
This is the classic "PS-to-PE" shift. When the market is paying for growth (PS), you can get away with spending money to make money. When the market shifts to PE, it wants to see the earnings. If the top AI companies don't deliver above-consensus commercial data in the next two to three quarters, the valuation system will shift from "Price to Sales" to "Price to Earnings." That's a systemic repricing, and it's not going to be a gradual one.
I've seen this pattern before. In crypto, when a project's funding runs out and the "testnet" phase ends, the token goes from being priced on "promise" to being priced on "Treasury." When that switch happens, the projects with real revenue survive, and the ones with only a narrative get liquidated. AI is a slightly more complex version of the same asset. The market is looking for a "real yield." It's looking for a protocol with fees.
The "Anti-Distillation" Wild Card
The report identifies "anti-distillation" as the biggest potential variable. This is where the technical gets interesting. Distillation is the process of using a large model's output to train a smaller, cheaper model. It's how the entire open-source ecosystem "stands on the shoulders of giants." If the top labs enforce a technical standard that prevents this (e.g., output watermarks, API usage restrictions), they cut off the "catch-up path" for smaller players.
If anti-distillation becomes the standard practice, the industry accelerates toward an oligopoly. The model gap becomes fixed, not because of the compute, but because of the data access. This is the "data moat" play. The top labs will not just have better compute; they will have exclusive access to high-quality interaction data. That's a positive feedback loop: compute -> model -> data -> compute.
The implication is clear: the market will reward the company that can maintain the model gap. If the gap is locked, the pricing power is locked. If the anti-distillation fails, the competitive landscape gets flushed out.
The Contrarian Take: What the Bulls Got Right
The report is bearish on the macro's ability to save the sector. But the bulls got one thing right: the fundamentals are strong. The market is not in a "bubble" in the sense of a zero-value asset. It's a market that's in a "repricing" phase. The value is real, but the price is not yet aligned with the execution.
The report's framework is correct to say that the interest rate isn't the "root cause." It's the industrial variables. But it's wrong to dismiss the macro entirely. The risk-free rate is still the denominator in every discounted cash flow model. A high interest rate environment means that a company's future earnings are worth less today. It compresses the multiple on even a good story. The report's "K-shaped divergence" signal is the real insight. If the dollar weakens and rate cuts are priced in, you could see a rebalancing of capital from US AI leaders to other markets. But that's a trade, not an investment. It's a short-term flow.
The Accountability Call
The real takeaway is a warning to the "grand narrative" chasers. The market is now in a phase where "grand narrative" is a liability. The "AGI is coming" story is no longer enough to justify a valuation. The market is asking for "proof of work."
The next 6-12 months will be a "distillation" of a different kind: a distillation of the true business models from the vapor. The market will separate the projects with real revenue and real customer retention from the ones with only a white paper and a demo. The "AI stock" era is over. The "AI business" era has begun. You don't have to be a technologist to see that the market has started to look for a "proof of revenue."
That's the new evaluation. I didn't have to parse a transaction log to see this; I just had to read the balance sheet. The question isn't "which model is smarter?" It's "which model is a business?" The market is no longer paying for the future. It's paying for the now. The bottleneck wasn't the compute. It was the conversion. And that's the code that needs to be audited.