Apollo Economist Finds Zero Financial Payoff from AI for Non-Tech Firms
Torsten Slok's analysis reveals flat profit margins across the S&P 493 despite massive enterprise investments.
Despite hundreds of billions of dollars poured into artificial intelligence infrastructure and software over the past few years, non-technology enterprise companies are seeing virtually no return on investment. A striking note published by Torsten Slok, chief economist at private equity giant Apollo Global Management, demonstrates that corporate profit margins and worker productivity outside Big Tech remain entirely flat.
Key Details
Torsten Slok’s research isolates the performance of what he terms the "S&P 493"—the entire S&P 500 index excluding the seven tech behemoths: Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla. While the Magnificent Seven continue to generate record profits driven by AI hardware sales and cloud hyperscaler infrastructure, the remaining 493 companies display no statistically significant gains in earnings or revenue per employee attributed to AI integration.
The report highlights several critical observations regarding the enterprise AI landscape:
- Flat Profit Margins: Across traditional industries such as manufacturing, retail, financial services, and healthcare, profit margins have remained static despite widespread enterprise AI adoption announcements.
- Unrealized Productivity: Measures of revenue per worker across non-tech S&P 500 members have shown zero acceleration, contradicting predictions that generative AI assistants and agentic workflows would immediately lift worker output.
- The Judgment Gap: Slok emphasizes that while AI excels at rapid computation and data aggregation, enterprise value in traditional industries depends heavily on judgment, accountability, organizational trust, and navigating unwritten institutional rules—areas where current AI models struggle.
- Capital Expenditure Divergence: While capital expenditure on AI hardware and cloud services continues to soar, the financial return remains concentrated exclusively within the technology providers supplying the infrastructure.
What This Means
Slok's findings expose a growing divide between AI hype and financial reality across traditional corporate America. For the past three years, corporate leadership teams have rushed to deploy copilots, automated customer service agents, and LLM-powered internal search engines, eager to demonstrate technological forward-thinking to shareholders. However, integrating AI models into legacy enterprise workflows has proven far more complex and costly than originally anticipated.
The fundamental challenge stems from the difference between task automation and economic value creation. Replacing or augmenting discrete tasks—such as drafting internal memos or generating boilerplate code—does not automatically translate into top-line revenue growth or bottom-line margin expansion. Unless companies can structurally restructure their operations or reduce headcount without degrading quality, the upfront license fees and cloud computing overhead of AI deployments can actually compress operating margins rather than expand them.
Technical Breakdown
The disparity between enterprise expectations and measured ROI highlights structural technical limitations in how current frontier models are deployed:
- Context Fragmentation: Legacy enterprise data structures remain fragmented across disparate legacy databases, making it difficult for Retrieval-Augmented Generation (RAG) systems to deliver accurate, actionable insights without extensive manual data cleaning.
- Inference Cost Overhead: Continuous API usage fees and dedicated cloud compute instances required to run advanced reasoning models create recurring operational expenses that offset marginal efficiency gains.
- Human-in-the-Loop Bottlenecks: Due to hallucination risks and compliance requirements, most non-tech enterprises require human oversight for AI outputs, effectively preventing full automation and maintaining existing labor costs.
- Lack of Institutional Context: Off-the-shelf generative AI models lack deep domain knowledge regarding specific corporate cultures, customer relationships, and nuanced risk management frameworks.
Industry Impact
The economic reality outlined in Apollo's research is expected to trigger a shift in corporate procurement strategies. CFOs outside the tech sector are increasingly demanding clear proof-of-concept metrics and explicit cost-reduction milestones before renewing expensive enterprise AI subscriptions. The era of blank-check experimentation with AI technology appears to be drawing to a close.
Furthermore, this dynamic creates risk for AI software vendors and enterprise SaaS providers whose valuations rely on expanding seat licenses across non-tech corporations. If client companies fail to measure tangible productivity improvements, renewals will contract, forcing software providers to pivot toward outcome-based pricing rather than seat-based models.
Looking Ahead
As corporate leaders re-evaluate their technology expenditures, the focus will likely pivot from broad conversational AI deployments to highly targeted, domain-specific automation solutions with unambiguous ROI. Companies that succeed in extracting genuine value from AI will be those that re-engineer core operational processes from the ground up rather than overlaying conversational interfaces onto legacy software architectures.
In the long run, productivity gains from general-purpose technologies historically require time to materialize as organizations adapt their workflows. Whether AI will eventually yield the promised efficiency boom for the broader economy remains an open question, but for now, the financial rewards of the AI boom remain almost exclusively confined to Silicon Valley.
Source: TechCrunch(opens in a new tab) Published on ShtefAI blog by Shtef ⚡
