Outside the technology complex, there is still no visible evidence of a broad AI-driven improvement in profit margins.The market has spent the better part of the past two years pricing AI as though it will eventually function like a universal profit-margin machine. Chips, data centres and hyperscalers may be collecting the revenue today, but the far larger valuation leap rests on a more ambitious assumption: that the rest of corporate America will soon use AI to do more with less, lifting margins across the S&P 493.TakeawaysThe AI trade still rests on a broad-margin miracle that has yet to appear outside technology. The market is pricing the destination while most of corporate America is still wrestling with the map.Tech can fold AI into code, products and workflows quickly. Banks, hospitals, utilities, manufacturers and governments have legacy systems, regulation, data-cleaning and human processes standing between the promise and the payoff.Falling token costs may expand usage, but they may also compress the revenue pool. More demand does not automatically mean every hyperscaler, model provider and data-centre operator earns the returns currently implied by valuations.The real risk is timing. If AI productivity gains arrive over five years rather than five months, corporate spending slows, ROI questions get louder and a market priced for instant margin expansion has to relearn patience.What AI Profit Margin Miracle?The market has spent the better part of the past two years pricing AI as though it will eventually function like a universal profit-margin machine. Chips, data centres and hyperscalers may be collecting the revenue today, but the far larger valuation leap rests on a more ambitious assumption: that the rest of corporate America will soon use AI to do more with less, lifting margins across the S&P 493.Apollo Chief Economist Torsten Slok is not seeing it yet.(as per his June 30 note)The first chart is the uncomfortable reality check. Outside the technology complex, there is still no visible evidence of a broad AI-driven improvement in profit margins. That matters because the AI trade is no longer valued merely on the earnings of a handful of semiconductor companies. It is valued on the expectation that the productivity gains will eventually spread through banks, insurers, manufacturers, logistics firms, health care providers and every other large corporate machine that has been promised an AI overhaul.That is the missing link in the story. The market is already paying for the destination, while the economy is still standing at the departure gate.Slok’s point is not that AI will fail to improve productivity. It is that investors may be assuming the benefits arrive in a straight line, at an implausibly fast pace, and with a much cleaner payoff than the real world is likely to allow. Software and technology firms can embed AI directly into products, workflows and code bases. In some cases, the benefit can be immediate. They are already living inside the digital factory.But that is not how most of the economy works.For a bank, insurer, hospital group, utility, defence contractor, airline, manufacturer or government agency, AI is not simply another software update. It often means rebuilding workflows, cleaning data, navigating regulation, managing compliance, retraining staff, redesigning accountability and integrating new systems with legacy infrastructure that may have been welded into the organisation for decades.The productivity promise may still be there. The timing is the problem.That distinction is becoming increasingly important as the market starts to focus on token costs, model routing and AI marketplaces. These are not just technical footnotes for engineers. They go to the heart of the business model. If the cost of intelligence keeps falling toward zero for a large share of use cases, then the industry may ultimately produce plenty of demand but not necessarily enough durable revenue to justify every hyperscaler, every data-centre buildout and every terminal-value assumption now embedded in the sector.The usual response is to invoke Jevons paradox: make compute cheaper, and demand explodes. That may well happen. But exploding demand does not automatically mean exploding margins for every participant in the value chain. A market can grow rapidly while returns become increasingly difficult to capture.That is where the second chart becomes more than a valuation argument. It is a question of corporate behaviour. Companies will keep spending aggressively on AI only if they can see a reasonably clear return on investment. If the promised margin uplift turns out to be a five-year project rather than a five-month one, finance directors will start asking harder questions.The danger is that equity markets have priced the productivity hockey stick before the factories, hospitals, insurers and logistics networks have even finished laying the ice.There is also an irony in the current setup. The software sector, which should in theory be one of the earliest beneficiaries of AI-driven efficiency, has already been punished this year by concerns that the same technology may compress the terminal values of established software businesses. AI may make companies more efficient, but it may also make their existing products less defensible.That is the tension investors are now being forced to confront.The AI story may still be right in the long run. But the margin miracle outside technology is not yet visible in the data, and the market’s patience may prove far shorter than the implementation cycle. For now, the sector is being priced for broad, fast and highly profitable adoption. Apollo’s warning is that the real economy may deliver something slower, messier and far more uneven.And in a market priced for perfection, slower is often enough to hurt.