Apollo economist: AI profit gains outside tech may take years

Key Takeaways

• Apollo's chief economist says AI profit margins outside tech may take 'well beyond' what markets expect to materialize
• Regulated industries face process overhauls and privacy requirements that could delay AI productivity gains for years
• Knowledge work productivity improvements are difficult to measure, meaning gains may not appear on balance sheets

AI productivity gains outside the tech sector could take far longer than Wall Street expects, according to Torsten Slok, chief economist at Apollo Global Management. His analysis suggests that regulated industries like healthcare, banking, and manufacturing face structural barriers that could delay meaningful returns for five years or more.

The warning matters because current AI stock valuations assume those gains will arrive quickly. Slok argues that AI company valuations rest entirely on the promise of rising margins at S&P 493 companies, the index minus the so-called Magnificent Seven tech giants. If the productivity bump takes five years instead of five months, many AI stocks face a painful repricing.

Why regulated industries face longer AI adoption timelines

Healthcare, banking, energy, pharma, and manufacturing share a common problem: compliance overhead. Process overhauls in these sectors require regulatory approval, security audits, and often multi-year implementation cycles. Privacy requirements add another layer. A hospital cannot deploy an AI system that touches patient data without clearing HIPAA review. A bank cannot automate loan decisions without satisfying fair lending rules.

These constraints do not disappear because the technology is ready. They extend the timeline between AI capability and AI profit impact. Slok's point is that markets are pricing in fast earnings growth, but real cash flows could trail far behind.

The measurement problem in knowledge work

There is another issue that Slok highlights. Even where individual employees are already more productive with AI tools, the gains are hard to measure in knowledge work. A software engineer who writes code 30% faster is not necessarily shipping 30% more features. A marketing analyst who drafts reports in half the time may just attend more meetings.

Without clear metrics, management cannot act on productivity improvements. They do not show up on the balance sheet. Instead, they get absorbed into daily operations. The company is not worse off, but it is not capturing the margin expansion that investors expect.

This creates a disconnect. Employees feel AI is helping them. Surveys show high adoption of tools like ChatGPT and GitHub Copilot. But CFOs looking at quarterly results do not see the corresponding uplift in earnings per share.

Falling token costs could squeeze hyperscalers

Slok's analysis also notes that falling token costs could cap hyperscaler revenue. The price of running AI inference has dropped sharply over the past 18 months. This is good for enterprises deploying AI, but it limits how much cloud providers can charge.

If Microsoft, Google, and Amazon cannot maintain AI revenue growth, the second-order effects hit chip suppliers like Nvidia. The entire AI value chain depends on sustained spending. A slowdown in enterprise AI adoption would ripple through.

What this means for AI builders

For teams building AI products, Slok's analysis is a reminder that enterprise sales cycles in regulated industries are long. A proof of concept in Q1 does not become a signed contract in Q2. Budget holders in healthcare, banking, and manufacturing move slowly because they have to.

The implication is strategic. AI startups targeting these verticals need runway measured in years, not months. They also need to build compliance infrastructure, security certifications, and audit trails from day one. These are table stakes, not optional features.

Product teams should also think about measurement. If customers cannot prove ROI internally, they cannot justify expansion. Building analytics and reporting into AI products helps customers show value to their CFO, which makes renewals and upsells easier.

The valuation question

Apollo manages over $600 billion in assets. When its chief economist warns about a mismatch between market expectations and likely outcomes, the audience is institutional investors managing billions more. The implication is that some AI positions may need to be trimmed or hedged.

For startups, this creates a secondary effect. If public market AI valuations compress, private market valuations follow. Fundraising for AI companies in 2025 and 2026 may get harder, not easier. Teams with strong revenue will be fine. Teams running on hype and promises may find investors less patient.

Frequently Asked Questions

What is the S&P 493?

The S&P 493 refers to the S&P 500 index minus the seven largest tech companies (Apple, Microsoft, Nvidia, Amazon, Alphabet, Meta, Tesla). These 493 companies represent about 85% of the index by count but a much smaller share of total market capitalization and recent gains.

Why are AI productivity gains hard to measure in knowledge work?

Knowledge work output is often qualitative and context-dependent. A faster report is not necessarily a better report. Time saved on one task may be absorbed by other activities. Without standardized metrics, companies struggle to quantify AI's impact on margins.

How long could AI adoption take in regulated industries?

According to Apollo's analysis, regulated industries like healthcare, banking, and manufacturing could see AI productivity gains delayed for five years or more due to compliance requirements, process overhauls, and privacy constraints.

What happens if AI productivity gains take longer than expected?

Current AI stock valuations assume rapid margin expansion across the economy. If gains take years instead of months, AI stocks face potential repricing as expected earnings diverge from actual cash flows.

Source: The Decoder / Matthias Bastian