How To Read The Numbers Behind AI’s Expanding Spend

🔍 Read the full analysis: How To Read The Numbers Behind AI’s Expanding Spend on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get hardware and tech essentials delivered free with Prime

• Fast, free delivery on millions of items
• Prime Video, Amazon Music and more included
• Member-only deals all year

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

Apollo economist Torsten Slok says recent corporate data show no visible financial payoff from artificial intelligence outside the technology sector, even as companies make large commitments to data centers, chips and cloud computing. His analysis, published October 24, puts a widening gap between measurable spending and uncertain returns at the center of the debate over AI’s business value.

Apollo’s charts compare the Magnificent Seven — Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla — with the other 493 companies in the S&P 500. In Apollo’s reading, inflation-adjusted revenue per employee and profit margins have risen for the seven technology companies, while the broader group has remained comparatively flat. The charts put the seven near $780,000 in revenue per employee in the third quarter of 2026, against about $525,000 for the other companies; the latter figure has remained within a similar range since 2021.

The note also points to productivity data from the San Francisco Federal Reserve, which Apollo says show no clear AI-driven boost in utilization-adjusted total factor productivity or output per hour. Slok’s broader conclusion is that mainstream companies are beginning to treat spending on AI tokens as a cost that can weigh on margins, rather than as a demonstrated source of growth. That is an interpretation of aggregate trends, not a direct measurement of AI’s effect on every company.

On the spending side, Meta’s October 22 filing lists about $355.73 billion in leases not yet started and $454.52 billion in non-cancelable contractual commitments, plus up to $14.72 billion in contingent cloud purchases. The commitments largely involve data centers, cloud capacity, servers and networking, though they are not all AI-related. Meta separately forecast $135 billion to $145 billion in capital spending for 2026. In the third quarter, it reported $34.91 billion in capital expenditures and negative free cash flow of $2.12 billion, compared with positive $10.6 billion a year earlier.

The Gap Between AI Costs and Returns

The figures matter because companies and investors are weighing large, long-term infrastructure bills against returns that are not yet apparent across the broader economy. Meta’s lease and contract totals extend well beyond a single year, while the company’s near-term capital spending is already substantial. The commitments indicate planned obligations, not money already spent, and they do not establish that all the capacity will be used for AI.

Evidence from individual business processes is more mixed than the aggregate picture. The supplied source cites studies in which customer-support workers resolved 14% more issues per hour, with a 34% gain among novice workers. Retail experiments cited in the source ranged from no sales increase to a 16.3% increase, depending on the workflow. Those results suggest AI can help in particular settings, but they do not show that the same gains have spread across companies or offset the costs of infrastructure and software.

The difference between firm-level examples and economy-wide averages is central to interpreting the debate. A handful of successful deployments can coexist with flat overall productivity if adoption is limited, benefits are concentrated, or costs offset gains elsewhere. Conversely, a flat average cannot establish that every company has received no benefit. Readers should treat spending totals, measured workplace outcomes and broad productivity indicators as different evidence, not interchangeable proof of success or failure.

What Apollo’s Charts Compare

Apollo’s October 24 Daily Spark note, titled “No Signs of AI Payoff Outside the Tech Sector,” used five charts to make its case. Alongside revenue per worker, margins and productivity, it cited Ramp data on AI-related spending by its business customers. In that dataset, the top 10% of firms account for 99.5% of spending on model serving and inference, 99.0% of neocloud spending and 95.6% of spending on OpenAI and Anthropic. Those figures describe Ramp’s more than 85,000 U.S. business customers; they are not a census of the S&P 500 or all U.S. companies.

Apollo also reported that 80.4% of the top 1% of AI spenders in the Ramp data were venture-backed, compared with 13.1% of median spenders. The concentration suggests that a relatively narrow group is driving much of the recorded spending. It does not, by itself, establish whether those firms are earning a return, or whether the spending will eventually create benefits for other companies.

Meta’s quarterly results show why infrastructure spending attracts attention. The company reported $64.60 billion in third-quarter revenue, up 26% year over year, alongside $32.79 billion in operating cash flow. Its filing’s roughly $825 billion in combined leases and commitments is a measure of future obligations across varying timelines, not a single-year bill. Separately, projections cited in the source put capital spending by Amazon, Alphabet, Meta, Microsoft and Oracle near $780 billion in 2026. That figure is a forecast, not a total of binding commitments.

“No Signs of AI Payoff Outside the Tech Sector”

— Apollo chief economist Torsten Slok, in the October 24 Daily Spark note

What the Aggregate Data Cannot Show

Apollo’s analysis does not isolate AI as the cause of changes in revenue per worker, margins or productivity. Those indicators can also be affected by interest rates, labor costs, tariffs, demand and other business changes. A flat average among 493 companies may conceal both firms that gained from AI and firms whose results weakened for unrelated reasons. The charts support the narrower conclusion that a broad payoff is not yet visible in these measures; they do not prove that AI has delivered zero value in every non-technology business.

The Ramp spending data have a separate limitation: they cover Ramp customers, not the full corporate sector, and spending concentration does not measure business outcomes. It is also unclear from the supplied figures how much of Meta’s future commitments will ultimately be used for AI rather than other products and infrastructure. The source does not provide a common measure of AI-related spending and returns across companies, so a direct comparison remains difficult.

Evidence to Watch in Coming Reports

The next test will be whether companies report measurable gains that can be compared with their costs: higher output, lower expenses, improved margins or revenue tied to specific AI uses. Future earnings reports and productivity data may help show whether the investments are translating into results beyond the technology sector. Companies’ updates on spending plans, infrastructure use and financing will also matter as commitments come due over different periods.

For now, the available evidence points to a clear difference between rapidly growing investment and unevenly documented returns. Apollo’s figures are a snapshot of current aggregate measures, not a final verdict on AI’s long-term economic effect. Investors and businesses will need to evaluate both the quality of reported gains and the costs required to produce them.

Key Questions

Does Apollo prove that AI has delivered no financial payoff?

No. Apollo’s statement describes aggregate trends outside the technology sector, including flat readings in selected productivity and company measures. The analysis does not test every AI deployment or establish that no individual company has benefited.

What does Meta’s roughly $825 billion figure represent?

It combines about $355.73 billion in leases not yet started, $454.52 billion in non-cancelable contractual commitments and up to $14.72 billion in contingent cloud purchases. The obligations cover multiple years and are not all specifically for AI.

Are companies already seeing AI productivity gains?

The source cites some task-level results, including a 14% increase in issues resolved per hour in customer support, with a larger gain for novice workers. It also reports that 83% of nearly 9,000 executives said they had seen no productivity impact so far, while the average reported effect was 0.2% a year. These results measure different things and should not be treated as a single sector-wide estimate.

Why can company examples look positive while broad productivity remains flat?

Benefits may be limited to particular tasks or firms, and adoption may not yet be broad enough to move economy-wide measures. Costs can also offset gains, while broad averages can conceal differences between companies. The available data do not identify one explanation as the cause.

Source: ThorstenMeyerAI.com