Apollo Report: AI Delivers 'Zero Financial Return' Outside the Tech Industry

A Report That Sparks Some Hard Thinking

Apollo (Apollo Global Management) has just released a study with a striking conclusion: AI has not yet delivered quantifiable financial returns outside the tech industry. On Hacker News, this news appeared with little fanfare—6 points and 1 comment—but its significance goes far beyond that. As generative AI capital expenditure pours into data centers on a scale of hundreds of billions of dollars, this judgment from an asset management firm is essentially asking one question: apart from the people selling shovels, who has actually struck gold?

Why Returns Are 'Zero': Three Overlooked Structural Causes

The Mismatch Between Capex and Returns

The dominant narrative around AI investment centers on compute, models, and infrastructure. The money flows to chipmakers, cloud providers, and model labs—these segments form a closed loop within the tech industry itself. AI procurement in traditional industries, by contrast, often remains stuck in pilot projects and proof-of-concept stages, and has yet to translate into auditable profit increments. Costs are booked, but returns are hard to attribute, so financially it naturally shows up as "zero."

Organizational Change Lags Behind Tool Deployment

Deploying an AI tool takes only weeks, but transforming the processes, job responsibilities, and decision-making chains to match it takes years. For most non-tech enterprises, the bottleneck is not technological availability but managerial inertia. Apollo's conclusion more likely describes "returns that have not yet been identified and measured" rather than "returns that do not exist."

The Measurement Framework Itself Has Blind Spots

Measuring financial returns depends on accounting line items. The hidden benefits of efficiency gains—such as employee time freed up, lower error rates, faster customer response—are hard to map directly onto the income statement. Evaluating a general-purpose technology with traditional financial metrics tends to systematically underestimate its value, similar to the historical experience of electricity and computers entering the enterprise.

General Purpose Technology (GPT) is a core concept in economics, referring to technologies that can permeate almost all industries and continuously spawn complementary innovations, such as the steam engine, electricity, and the internet. The value release of GPTs has a pronounced "delay" characteristic: enterprises need to restructure processes, accumulate new skills, and adjust organizational structures before technological potential can be converted into productivity gains. Economist Paul David's classic research on electricity showed that factory electrification began spreading in the late 19th century, but the significant jump in total factor productivity did not appear until the 1920s—a lag of nearly thirty years. The reason was that early factories simply replaced steam engines with electric motors without changing shop-floor layout or production processes. This historical pattern suggests that evaluating the returns of a GPT with short-cycle financial metrics easily leads to a "zero" conclusion, which more likely reflects premature measurement timing rather than a lack of value in the technology itself.

What the Tech Industry's 'Exception' Tells Us

The reason the tech industry can realize returns is that it is simultaneously the supplier and the demander: AI is directly embedded in its product pricing (API calls, subscription upgrades) and cost structure (code generation, customer service automation). This dual identity amplifies the visibility of returns. For other industries, AI is more like a general-purpose technology whose value release follows the logic of slow variables and requires accompanying process reengineering and organizational learning.

Another angle worth noting is the competitive landscape. If AI returns in non-tech industries remain zero for a long time, corporate adoption incentives will shift from "pursuing growth" to "defensive following," which will flatten the AI penetration curve and lengthen the payback period for infrastructure investment.

Picks and shovels is a business model in the investment world that describes "selling tools rather than digging for gold yourself during a gold rush," originating from the 19th-century California Gold Rush, when most prospectors did not get rich, while merchants selling shovels, jeans, and equipment profited handsomely. In the AI context, this metaphor refers to chipmakers (such as Nvidia), cloud providers, and model providers—they benefit directly from AI infrastructure procurement demand across all industries, with returns that are certain and auditable. In contrast are the "gold diggers," i.e., enterprises trying to transform their own businesses with AI, whose returns are constrained by process change, organizational learning, and the difficulty of return attribution. The conclusion of the Apollo report essentially says: at present, most of the profit in the AI value chain is concentrated in the "picks and shovels" segment, and value capture on the demand side is still in its early stages. This pattern also explains why there is a significant gap between the scale of AI capital expenditure and returns outside the industry.

Implications for Practitioners

For corporate decision-makers, the practical value of this report is the reminder that the business case for AI projects needs more refined attribution design. Tracking efficiency metrics, quality metrics, and financial metrics side by side is more feasible than waiting for a vague "AI return rate."

For AI practitioners, the signal is equally clear: the next stage of competition is not in model capabilities but in industry know-how and engineering for implementation. Whoever can translate general capabilities into measurable outputs for specific roles can break the "zero return" judgment.

Proof of Concept (PoC) is a common initial stage when enterprises introduce new technologies, with the goal of verifying technical feasibility through small-scale experiments rather than generating quantifiable business value. A large number of AI projects in non-tech enterprises remain stuck in the PoC stage for a long time, for reasons including: disconnect between experimental and production environments, lack of engineering capability to scale pilots into full deployment, and inability to translate technical metrics (such as model accuracy) into business metrics (such as unit cost reduction or revenue growth). There is usually a "valley of death" between PoC and scaled implementation, requiring accompanying data governance, process reengineering, and change management. This also means that if enterprises measure AI progress by the number of PoCs, they tend to overestimate the actual level of implementation; but if measured by financial returns, they will underestimate long-term potential because projects have not yet crossed the scaling threshold.

One-line Summary

Apollo research finds that AI has yet to deliver financial returns outside the tech industry, owing to a triple blind spot in attribution, organization, and measurement.

Paragraph Summary

A new study by Apollo Global Management concludes that AI has not yet delivered quantifiable financial returns outside the tech industry. The article explains this phenomenon through three structural causes: capital expenditure is concentrated in tech supply chain segments such as chips, cloud, and models, while procurement in traditional industries mostly remains at the pilot stage, making returns hard to attribute; organizational and process change lags behind tool deployment, and managerial inertia constrains value release; and traditional financial metrics struggle to capture the hidden benefits of efficiency gains. The tech industry is the exception because it is simultaneously the supplier and demander of AI, giving returns higher visibility. The article reminds enterprises to establish parallel tracking of efficiency, quality, and financial metrics, and points out that the key to the next stage of competition lies in industry know-how and engineering implementation capability.