On October 24, 2026, Apollo Global Management released a research report. Its conclusion: outside the technology industry, artificial intelligence has so far produced no measurable productivity gains.
The report spans 12 major industries, tracking productivity data from more than 2,400 companies in North America and Europe between January 2024 and September 2026, and interviews 340 chief information officers. The study controls for economic cycles, labor force changes, and capital expenditure patterns to isolate AI's independent impact. The data reveals a stark divide: tech companies achieved 18% to 23% AI-driven productivity gains, while indicators in financial services, healthcare, manufacturing, retail, and professional services show no significant difference from before AI deployment.
Apollo Chief Economist Torsten Slok said: "At this point, we're just not seeing that yet."
A Divide, Two Realities
The Apollo data is directly reflected in stock market valuations: revenue per employee and profit margins at the seven largest tech companies continue to climb, while the same metrics at the other 493 S&P 500 companies remain essentially unchanged over the same period. This mirrors differences in AI penetration depth.
Non-tech companies spend an average of $47 million on AI systems per 1,000 employees. For a mid-sized manufacturer with 10,000 employees, that means an annual investment of approximately $470 million, yet productivity indicators show no quantifiable return.
Non-tech enterprise spending on AI tools and infrastructure is projected to reach $300 billion in 2026.
The Mechanism Puzzle: Same Tools, Why Only One Type of Company Benefits
The Apollo report points out that the root of the gap lies in how AI is integrated with workflows. Tech companies build "AI-native" processes—from product design and code review to customer support, AI is a core component.
Traditional industries adopt a "transplant" strategy: layering AI tools on top of existing processes. An insurance company might add an AI-assisted system to its underwriting workflow, but the workflow itself, the data standards, and the decision points remain fundamentally unchanged. AI's efficiency ceiling is constrained by the floor of the processes it is embedded in.
Intuit data shows that 88% of financial services companies already use AI to support relevant functions, and 59% of finance leaders use AI for daily decision-making. But Apollo data shows their productivity gains are close to zero.
Widespread use does not equal deep integration, and deep integration does not equal productivity gains.
The Limits of Counter-Evidence: What the Data Claiming Returns Actually Says
Grand View Research cites a joint Microsoft-IDC study claiming that 79% of healthcare organizations use AI, with a payback period of 14 months and a return of $3.20 for every $1 invested. Atlanta Fed data shows financial industry labor productivity rising about 0.5% per quarter on average, with high-skill services implying growth of about 0.8% in 2025.
These data points must distinguish between two things: efficiency gains from individual tools in controlled settings, and measurable aggregate productivity gains across an entire industry. The Microsoft-IDC study is based on a self-selected sample of proactive adopters and cannot represent the industry average. A 0.5% quarterly increase annualizes to about 2%, which remains far from an "AI-driven productivity revolution" within statistical noise.
One reverse finding: on tasks beyond AI's capability boundaries, workers using AI had accuracy rates 19 percentage points lower. Misuse not only fails to improve performance but causes losses.
How Much Comfort Can Historical Precedents Offer
It took roughly thirty years from electricity's spread to its driving a manufacturing productivity revolution, and more than fifteen years for computers to contribute measurable productivity gains. The prerequisite was redesigning factory layouts and management organizations—not that the technology itself was "useless." The Apollo report points in the same direction: current AI deployment resembles the 1980s, when computers were placed on desks while paper-based processes continued.
The problem lies in companies' lack of willingness to restructure processes and their inadequate data infrastructure. MIT's Daron Acemoglu analyzed in 2024 that without organizational redesign, AI's contribution to total factor productivity over the next decade may be only 0.5% to 1%, below current market valuation expectations.
The Fundamental Misjudgment in Corporate AI Strategy
Apollo's sample selects companies that publicly announced major AI investments, with average spending of $47 million per 1,000 employees but no productivity improvement. Companies investing less are likely to perform even worse.
The current mainstream deployment path has structural flaws. Companies equate AI investment with buying software, neglecting high-quality internal data, redesigning decision-making processes, workforce capability transformation, and awareness of AI's boundaries.
Of the $300 billion in AI spending in 2026, a considerable portion purchases the signal value of "already using AI" to boards and markets, rather than operational efficiency.
Assessment
The Apollo report reveals that AI requires extremely high supporting conditions to deliver value, and the vast majority of non-tech companies do not yet possess them. Tech companies benefit first because their product forms, data foundations, and organizational cultures are designed for rapid integration of new tools.
For decision-makers in other industries, the data points to the question: what prerequisites are needed for a company to achieve AI productivity gains, and how many does it currently have. Without an answer, $300 billion has already demonstrated what that invoice can buy.
© 2026 Winzheng.com 赢政天下 | 转载请注明来源并附原文链接