AI spending is rising. Where are the returns? : Conferences That Work


There’s a growing disconnect between the money being spent on artificial intelligence and the evidence that it’s delivering significant economic benefits to businesses.

On October 24, Torsten Slok, Chief Economist at Apollo Global Management, published an analysis titled “No Signs of AI Payoff Outside the Tech Sector.” Its conclusions are striking.

Revenue per worker is increasing among the Magnificent Seven (Mag 7) technology companies, but remains flat across the other 493 companies in the S&P 500 (S&P 493).

Chart illustrating rising revenue per worker in select industries (the Magnificent Seven tech companies), while the revenue per worker in the rest of the S&P remains unchanged. Inflation adjusted to 2019 Q4. Sources: Bloomberg, Macrobond, Apollo Chief Economist.
Sources: Bloomberg, Macrobond, Apollo Chief Economist.

Profit margins show the same pattern: rising for the Mag 7, flat for the S&P 493.

Chart illustrating profit margins remaining flat in the S&P 493, with profit margins rising solely in the Magnificent Seven tech companies. Sources: Bloomberg, Macrobond, Apollo Chief Economist.
Sources: Bloomberg, Macrobond, Apollo Chief Economist.

Apollo also finds no evidence that AI is boosting productivity growth.

Chart depicting a lack of improvement in productivity due to AI, with consistent low levels between 1985 - 2025. Total Factor Productivity (TFP) is the utilization-adjusted series from the San Francisco Fed (Fernald, 2014), adjusted for cyclical variation in factor utilization. Sources: Federal Reserve Bank of San Francisco, US Bureau of Labor Statistics, Apollo Chief Economist.Chart depicting a lack of improvement in productivity due to AI, with consistent low levels between 1985 - 2025. Total Factor Productivity (TFP) is the utilization-adjusted series from the San Francisco Fed (Fernald, 2014), adjusted for cyclical variation in factor utilization. Sources: Federal Reserve Bank of San Francisco, US Bureau of Labor Statistics, Apollo Chief Economist.
Sources: Federal Reserve Bank of San Francisco, US Bureau of Labor Statistics, Apollo Chief Economist.

These are aggregate economic indicators. They don’t tell us what is happening inside individual companies, and they don’t prove that AI has delivered no benefits outside the technology sector. But they do suggest that, so far, the enormous investment and enthusiasm surrounding AI have not translated into measurable improvements across most of corporate America.

Who is spending on AI?

Apollo’s analysis includes another finding I think deserves particular attention.

Venture-capital-backed companies are spending more and more on AI tokens. Meanwhile, mainstream corporate America is pulling back, increasingly treating token spending as a cost center that weighs on margins rather than as an investment in growth.

A structured table illustrating that AI spending is extremely concentrated in a small number of firms. Sources: Ara Kharazian, Chief Economist, Ramp, Apollo Chief Economist.A structured table illustrating that AI spending is extremely concentrated in a small number of firms. Sources: Ara Kharazian, Chief Economist, Ramp, Apollo Chief Economist.
Sources: Ara Kharazian, Chief Economist, Ramp, Apollo Chief Economist.

AI spending is also highly concentrated in a relatively small number of companies. The data Apollo presents suggest that much of the current demand for AI is being driven by venture-funded businesses, rather than by established companies generating demonstrable returns from their AI investments.

Bar chart displaying the financial backing of businesses by AI spend intensity. VC backed businesses are 80.4% of the top 1% of AI spenders, compared to 13.1% of non-VC backed companies. Sources: Ramp AI Index, Apollo Chief Economist.Bar chart displaying the financial backing of businesses by AI spend intensity. VC backed businesses are 80.4% of the top 1% of AI spenders, compared to 13.1% of non-VC backed companies. Sources: Ramp AI Index, Apollo Chief Economist.
Sources: Ramp AI Index, Apollo Chief Economist.

In other words, substantial spending on AI is happening, but much of it isn’t yet justified by measurable improvements in business performance.

More capable models, more consumption, more complexity

There’s a plausible reason to expect AI economics to change.

Models are improving rapidly. Open-weight models can increasingly match the performance of much more expensive proprietary models. A recent Mozilla report, covered by Ars Technica, reports that the performance gap between leading open-weight models and frontier proprietary models had narrowed to about four months. In some comparisons, open models deliver similar results at a fraction of the cost.

This is an important development. The cost of accessing a given level of AI capability is falling, even as the capabilities themselves improve.

But cheaper models can also encourage much greater use. Organizations can run more queries, deploy more agents, automate more processes, and build increasingly elaborate AI-powered systems. Lower costs per token don’t necessarily mean lower total spending.

And even if the cost of generating intelligence keeps falling, other costs remain.

Security and exposure

The more systems, services, models, APIs, data stores, and external dependencies we connect, the more opportunities there are for failures and compromises. On October 29-30, Binance disclosed a security incident involving unauthorized withdrawals from its hot wallets, with an estimated exposure, at the time of writing, of $6.3 billion. The incident wasn’t AI-related, but it illustrates the potentially enormous consequences of security failures in complex technical infrastructures. Binance’s investigation is ongoing, but the incident is a warning of the potentially enormous consequences of security failures in complex technical systems.

Maintenance and reliability

People still have to test, understand, update, debug, secure, and maintain AI-generated software. Dependencies change. Services fail. Vulnerabilities are discovered. Systems that nobody fully understands can be particularly difficult and expensive to repair when something goes wrong. I’ve explored this problem in more detail in The Coming Crisis of Unmaintainable AI-Generated Software.

Human oversight and accountability

Businesses still need people who can evaluate what systems are doing, determine whether their outputs are appropriate, manage exceptions, and take responsibility for consequential decisions. Automating code production or task execution doesn’t eliminate these responsibilities. It may make them harder.

There’s also a question of whether organizations can realize the potential savings from cheaper models while managing the growing complexity of the systems they build around them.

An important distinction remains: the cost of generating software and the cost of operating it successfully over time are very different things.

The business case remains unproven

I think we’re at a critical point in the development of AI.

While capabilities are improving rapidly and the cost of accessing those capabilities is falling, AI usage and spending by some organizations are increasing.

Yet we still have little evidence of broad improvements in productivity, revenue per worker, or profit margins outside the largest technology companies.

Perhaps those benefits will emerge as businesses learn how to use increasingly capable and inexpensive models effectively.

But the question remains whether cheaper intelligence will translate into higher revenue and better profit margins once you include the full costs of deploying and operating AI systems.

I think that we’re underestimating the costs of increasingly complex and opaque dependencies, consequential security exposure, software maintenance, and the continuing need for human oversight.

Spending more on AI, or using more AI, isn’t the same as creating more value.

I’d like to see convincing evidence that we’re getting better at the latter.

We will be happy to hear your thoughts

Leave a reply

Som2ny Network
Logo
Register New Account
Compare items
  • Total (0)
Compare
0
Shopping cart