AI Margin Mirage: BlackRock's Wei Li Bets on a Narrative the Data Hasn't Confirmed

Between the blocks, silence screams the truth. Last week, BlackRock's investment strategist Wei Li made a declarative statement that rippled through trading desks: AI-driven earnings growth would reshape portfolio construction, tilting the scales decisively toward US equities over government bonds. The conclusion was clean. The logic was seductive. The data, however, is a mess of unverified assumptions and concentration risks that the market has decided to ignore. Li's thesis rests on a simple causal chain: AI is moving from proof-of-concept to scaled commercialization, and that transition is generating enough profit to justify a structural preference for equities. On the surface, this aligns with the headline numbers. Microsoft's intelligent cloud segment posted 20% year-over-year growth. NVIDIA's data center revenue has been a hammer for six consecutive quarters. Enterprise AI budgets are climbing from roughly 2% of IT spend to a projected 10% by 2025. The signal seems loud. But when I strip away the earnings call euphoria and map the actual on-chain and cross-sector flows, the picture looks less like a broad-based earnings revolution and more like a liquidity event concentrated in a handful of names. Floors are illusions until you map the liquidity. The first problem with Li's thesis is the unspoken assumption that AI earnings growth is a wide-market phenomenon. It is not. The S&P 500's forward P/E sits near 21-22 times earnings, a multiple justified almost entirely by the Magnificent Seven, which trade at 30-35 times forward. The equity risk premium has collapsed to roughly 30 to 50 basis points. In plain terms, you are being paid almost nothing to own equities over risk-free bonds. The market has already priced in a future where AI delivers sustained, compounding profit growth. That is not a contrarian position. That is a crowded trade. My own audit experience in the DeFi Summer taught me that when yield looks too good, wash trading is usually the culprit. The same logic applies here. What portion of the AI trade is real cash flow, and what portion is narrative-driven multiple expansion? The numbers suggest the latter is doing most of the heavy lifting. AI-related revenue still accounts for less than five percent of total S&P 500 earnings. The gap between narrative and contribution is a structural vulnerability, not a detail. Li's preference for equities implies more than a sector rotation. It implies that AI-driven profit growth will outpace the headwind of a 4.0-4.5% ten-year Treasury yield. The historical record says that is a high bar. The 2022 winter, which I audited protocols through, showed me that leverage built on fragile collateral does not survive a repricing. Today's equity market is running on a similar collateral: an assumption that AI margins will remain sticky even as competition compresses API prices and infrastructure costs climb. The infrastructure layer is the one place where the data is unambiguous. AI capital expenditure is projected to exceed $200 billion in 2024 alone, driven by hyperscalers and model labs. NVIDIA holds over 80% market share in AI accelerators. That is the core of the trade. But here is the uncomfortable part: the profitability is accruing to the pick-and-shovel suppliers, not to the application layer where most of the hype lives. When I examine the earnings breakdown, the "AI profit growth" Li references is disproportionately a chip and cloud story. It is not a general market story. The application layer, with the exception of a few enterprise software names, is still burning cash to chase adoption. This concentration creates a fragile equilibrium. If the market is paying 60-70 times forward earnings for NVIDIA, it is betting on flawless execution for the next decade. Any slip in the Blackwell ramp, any quarter of decelerating data center growth, and the repricing will not stay contained to one ticker. The correlation between the Magnificent Seven and the broader index has been rising. A de-rating in AI names will not spare the rest of the equity complex. The competitive dynamics are equally underappreciated in Li's framing. The model layer is a three-horse race, but the race is cutting margins. API prices have collapsed over ninety percent since GPT-4's launch. OpenAI and Anthropic are growing revenue, but they are doing so against a backdrop of brutal cost curves. The open-source pressure from Llama and, more recently, DeepSeek, is not a theoretical threat. It is a live cap on the pricing power of closed models. The assumption that AI profit growth persists in perpetuity ignores the historical pattern of every infrastructure cycle. Capacity floods in, margins compress, and the late-cycle buyers get hurt. Regulatory overhang is the silent variable. The EU AI Act went into effect in August 2024. Over forty US states have proposed AI-related legislation. Copyright litigation against OpenAI and Anthropic is not a tail risk; it is a pending balance-sheet liability. The analysis I did on the FTX collapse and reserve discrepancies taught me that when a market refuses to price in legal and governance risks, the adjustment is violent, not gradual. Li's report does not mention these risks. That omission is not an oversight. It is a selection bias that favors the bullish case. The contrarian angle here is not that AI is a bubble. That is a lazy take. The contrarian angle is that AI earnings growth is a real but narrow phenomenon that has been mislabeled as a broad market tailwind. The data supports a concentrated bet on infrastructure and model quality, not a blanket preference for equities over bonds. The equity risk premium is offering no compensation for the variance. If you are a strategist recommending a structural shift based on AI, you need to answer one question: what happens when the rate of AI revenue growth decelerates from fifty percent to twenty percent? The multiple will not hold. Structure creates freedom; chaos demands order. The order here is to demand evidence of breadth. Track the percentage of S&P 500 companies mentioning AI in earnings calls that actually show AI revenue in the financials. The gap is staggering. Follow the enterprise software renewal rates, not the pilot announcements. A pilot is not a commitment. A renewal is. The market is treating every AI pilot as a revenue annuity. History, and my own audit of NFT floor prices in 2021, shows that volume without unique adoption is just a data artifact designed to deceive. What would change the trade? If the ten-year yield breaks above 4.5%, the relative value math breaks. If Microsoft or Amazon guide down on AI cloud capacity, the narrative cracks. If NVIDIA's Blackwell launch slips again, the leverage unwinds. These are the signals to watch. The next six months will separate the AI earnings story from the AI narrative trade. Li's recommendation is not a data-driven conclusion. It is a bet on the persistence of a margin miracle. The data I have mapped across infrastructure, application, and competitive dynamics says the miracle is real, but it is priced. Between the blocks, silence screams the truth: the market is not paying for AI earnings growth. It is paying for AI earnings growth without any downside variance. That is a price no rational book should pay. The takeaway is not to abandon equities. The takeaway is to demand a better price for the same risk. The next time a strategist tells you AI has reshaped the investment landscape, ask them to show you the renewal rates, the revenue concentration, and the margin compression curves. If they cannot, they are selling you a map where the territory is still unmapped. The data will tell you when the trade is real. It has not told you that yet.