🔍 Read the full analysis: What’s Behind The Gap Between AI Spending And Returns? on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Apollo economist Torsten Slok argues that AI has not produced a visible financial payoff for corporate America outside the technology sector, based on measures of revenue per worker, margins and productivity. Meanwhile, Meta’s filings show about $825 billion in future commitments, while research and company examples point to gains in some specific workflows. The aggregate evidence does not prove that AI creates no value; it shows returns remain uneven and difficult to identify.
Apollo chief economist Torsten Slok said a review of corporate and productivity measures showed no visible financial payoff from AI outside the technology sector, sharpening questions about whether rapidly growing investment in data centers and computing capacity is producing returns. The conclusion, published in an Apollo note on October 24, describes aggregate trends; it does not establish that AI has delivered no benefits at individual companies or in particular tasks.
Apollo’s analysis compares the seven largest U.S. technology companies with the other 493 companies in the S&P 500. Its charts show revenue per worker rising for the seven while remaining broadly flat for the rest, and profit margins increasing for the technology group while holding near 10% for the others. The note also points to San Francisco Federal Reserve measures of productivity, saying they show no sign of an AI-driven productivity boost so far. These indicators track broad economic outcomes; they do not isolate AI as the cause of changes.
The note also uses spending data from Ramp, a corporate card and spending platform with more than 85,000 U.S. business customers. It reports that the top 10% of firms account for nearly all recorded spending on model serving and inference, neocloud services, and OpenAI and Anthropic. The figures describe Ramp’s customers, not all U.S. businesses. Apollo says many leading AI spenders are venture-backed, raising questions about whether some demand reflects sustained commercial use or investment activity.
On the spending side, Meta’s October 21 quarterly release reported $34.91 billion in capital expenditures and negative free cash flow of $2.12 billion for the third quarter, compared with positive free cash flow a year earlier. Its filing the next day listed 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. Those amounts total roughly $825 billion in future commitments; the filing includes items beyond AI, including some Reality Labs hardware. Meta forecast capital spending of $135 billion to $145 billion for 2026.
Infographic 01 · The “zero” in numbers
What “zero financial payoff” actually measures.
Apollo’s headline rests on five charts. Here are their numbers, and what they can and cannot prove.
Who pays for AI? Share of spending by the top 10% of firms
Ramp customer data via Apollo · non-AI categories in grey for comparison
Spending is concentrated in every category, not only AI; AI is the most extreme.
What the five charts do and do not show
shown not shown
- Shown: revenue per worker and margins rising for the 7 tech giants, flat for the other 493
- Shown: no AI lift yet in US productivity statistics (San Francisco Fed TFP, output per hour)
- Shown: AI spending sits in few, mostly venture-funded firms (Ramp, 85,000+ US businesses)
- Not shown: that AI caused the gap; rates, tariffs and wages move the same lines
- Not shown: gains inside single functions, which can cancel out in an average
- Not shown: the S&P 493 itself in the Ramp data; those are Ramp’s customers
“The AI story will be decided in the S&P 493, not the Mag 7.”Torsten Slok, Apollo chief economist, 24 Oct 2026 (opinion)
Sources: Apollo Daily Spark (24 Oct 2026) with charts by Bloomberg, Macrobond, SF Fed, BLS and Ramp; Apollo (20 Oct 2026); Ramp AI Index. Values with ≈ are read off Apollo’s charts. Assessment: Thorsten Meyer.
Powered by Thorsten Meyer AI
The Test Is Moving From Spend to Returns
The dispute matters because AI infrastructure spending is large, long-term and increasingly tied to financing, while many businesses cannot yet point to broad gains in revenue or productivity. If returns remain concentrated among technology companies or a small share of firms, other businesses may face pressure to justify their own spending, and investors may scrutinize whether planned capacity can earn enough to cover its costs.
There are signs that AI can improve specific workflows. The source material cites research finding 14% more customer-support issues resolved per hour, with a 34% gain for novice workers, and retail experiments ranging from no sales lift to a 16.3% increase depending on the workflow. Those results concern defined settings and should not be treated as proof of economy-wide gains. They do show why aggregate averages may miss practical benefits that have not yet translated into higher company-wide productivity or margins.
The distinction is important for workers and customers as well as investors. A tool can save time in one task without changing a company’s total costs or output, particularly if adoption requires new software, training, computing and changes to established processes. The economic question is not only whether a model can perform a task, but whether the full cost of deploying it is lower than the value it creates.
Products Worth Considering
Jackery HomePower 3600 Plus Portable Power Station, 3584Wh, 3600W
Anker SOLIX C200 DC Portable Power Station, 192Wh, 200W
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Black
As an affiliate, we earn on qualifying purchases.
What Apollo’s Measures Capture
Slok’s October 24 note, titled “No Signs of AI Payoff Outside the Tech Sector,” drew attention to comparisons between the Magnificent Seven and the other S&P 500 companies, alongside productivity data and Ramp spending patterns. Its conclusion followed comments from Slok on October 20 and October 22 about compute costs and venture-funded spending. The note’s “zero” wording is an interpretation of aggregate indicators, not a direct experiment measuring AI’s effect on every company.
The investment figures also need to be read carefully. Meta’s roughly $825 billion combines different kinds of obligations over multiple years, including leases and contractual purchases; it is not a single-year cash bill or an AI-only figure. Separately, forecasts cited for large technology companies’ capital spending are projections, not commitments already made. The source material also reports that SpaceX borrowed $40 billion to buy AI chips, illustrating how financing has become part of the debate, but borrowing alone does not establish whether the investment will generate a return.
“No Signs of AI Payoff Outside the Tech Sector”
— Apollo chief economist Torsten Slok, in an October 24, 2026, research note
Where the Return Evidence Falls Short
A flat average does not mean every company has seen no gains. Results among the 493 non-technology S&P 500 companies could vary, with increases and declines canceling each other out. The figures cited do not separate AI’s contribution from other forces such as labor costs, interest rates or tariffs, and productivity data may take time to reflect changes in how firms work.
The Ramp figures cover its own customer base and do not represent a census of businesses. The source material also cites a survey in which 83% of nearly 9,000 executives reported no productivity impact from AI so far, with an average estimated effect of 0.2% a year; those are survey responses, not a direct measurement of national productivity. It remains unclear how quickly reported improvements in specific tasks will spread, how much they will reduce total costs, and whether AI-related revenue will cover the cost of computing and infrastructure.
Meta’s commitments do not all relate to AI, and their timing varies. The source material reports higher insurance costs for debt associated with Meta and CoreWeave, but does not provide enough detail to determine how those costs will affect either company’s investment plans. Nor does the reported $40 billion SpaceX borrowing, by itself, show whether the chips it finances will be profitable.
Evidence to Watch in Coming Quarters
The clearest tests will come from company disclosures over the next several quarters: whether spending on computing capacity is followed by measurable revenue growth, lower operating costs or improved output per worker. Meta’s 2026 capital-spending range and its schedule of future commitments will help investors track the scale and timing of its obligations, though the company’s filings do not assign every expense to AI.
Researchers and executives will also need to distinguish task-level results from company-wide performance. More consistent reporting on deployment costs, adoption rates and productivity outcomes could show whether early gains spread beyond a small group of firms. Until then, the available evidence supports a narrower conclusion: AI investment is expanding, but broad financial returns remain difficult to identify.
Key Questions
Did Apollo prove that AI has produced no financial returns?
No. Apollo’s conclusion is based on aggregate trends in revenue per worker, profit margins and productivity. Those measures do not isolate AI’s effect or rule out gains at individual companies and in specific tasks.
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 commitments span multiple years and include spending that is not all AI-related.
Are there measured examples of AI improving work?
The source material cites a customer-support study reporting 14% more issues resolved per hour, with a larger gain for novice workers, and retail experiments with sales results ranging from zero to 16.3%, depending on workflow. These findings apply to specific settings, not the whole economy.
Why might company-wide productivity gains be hard to see?
Task-level time savings may not immediately raise total output or margins. Companies may face computing, software, training and process-change costs, and gains at some firms can be obscured by flat or falling results elsewhere.
What should readers watch next?
Look for company disclosures that connect AI investment to revenue, costs or output per worker, as well as updates on spending plans and infrastructure commitments. Those measures can help show whether reported task-level gains are becoming durable business returns.
Source: ThorstenMeyerAI.com
