The AI Boom By The Numbers: Costs Rise, Returns Vary

🔍 Read the full analysis: The AI Boom By The Numbers: Costs Rise, Returns Vary on ThorstenMeyerAI.com

Age 18–24?Offer from Amazon

Prime made for students and young adults

• Fast, free delivery for dorm and study essentials
• Prime Video and Amazon Music included
• Member-only deals

Try Prime for Young Adults →Free trial for eligible 18–24 year olds

As an affiliate, we earn on qualifying purchases.Hide

Apollo Global Management chief economist Torsten Slok says recent corporate and productivity data show no visible financial payoff from artificial intelligence outside the technology sector, even as companies commit hundreds of billions of dollars to infrastructure. His October 24 analysis compares large technology companies with the rest of the S&P 500; it does not directly measure AI’s effect on each business, and evidence from specific workplace tasks shows some gains.

Apollo’s analysis finds that inflation-adjusted revenue per employee has risen for the “Magnificent Seven”—Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla—while remaining broadly flat for the other 493 S&P 500 companies. Apollo’s charts put the seven near $780,000 per employee in the third quarter of 2026 and the other companies around $525,000. The comparison describes business performance; it does not establish that AI caused the difference.

The same charts show the seven companies’ profit margins climbing to about 30%, compared with roughly 10% for the other 493. Apollo also cites San Francisco Federal Reserve productivity measures that, according to the note, show no signs yet of AI boosting productivity growth. These economy-wide measures may not capture gains in particular tasks or businesses, and they cannot isolate AI from other forces affecting costs and output.

The spending data point to concentration as well as scale. Figures from Ramp, a corporate spending platform with more than 85,000 U.S. business customers, show the top 10% of firms account for 99.5% of spending on model serving and inference, 99% of neocloud spending, and 95.6% of spending on OpenAI and Anthropic. Ramp’s customer data are not a census of U.S. companies. Apollo says 80.4% of the top 1% of AI spenders are venture-backed, compared with 13.1% of median spenders.

Meta’s filings illustrate the infrastructure bill. Its third-quarter results, released October 21, reported $64.60 billion in revenue, up 26% year over year, and $34.91 billion in capital expenditures. Operating cash flow was $32.79 billion, leaving free cash flow at negative $2.12 billion, compared with positive $10.6 billion a year earlier. Meta set its 2026 capital spending outlook at $135 billion to $145 billion.

A filing the next day listed roughly $825 billion in future commitments: about $355.73 billion for leases not yet started, including data center and energy infrastructure, and $454.52 billion in non-cancelable contractual commitments, largely for cloud capacity, servers and network equipment. Up to another $14.72 billion could be spent on contingent cloud capacity. The total includes commitments that are not all AI-related, and much of the spending is scheduled over multiple years. Meta reported 66,775 employees, down 15% from a year earlier.

The Cost of Building AI Capacity

The gap between large infrastructure bills and unclear broad returns matters to investors, companies and customers. If AI tools do not lift revenue, reduce costs or improve output enough to cover their expense, businesses may cut deployments or face pressure on margins. Slok’s note says mainstream companies are beginning to view token use as a cost center; that is his interpretation of the trends, not a finding that applies to every company.

Financing is another pressure point. SpaceX borrowed $40 billion to buy AI chips, according to the source material. The report also says insurance costs for debt issued by Meta and AI cloud provider CoreWeave rose. These developments suggest lenders and investors are watching how the buildout is funded, alongside demand for computing capacity. They do not, by themselves, show that either company is unable to meet its obligations.

For workers and users, the aggregate figures are only part of the picture. In customer support research cited in the source, AI assistance was associated with 14% more issues resolved per hour, and 34% more for novice workers. Retail experiments reported sales changes ranging from zero to 16.3%, depending on the workflow. Those results suggest where practical benefits may appear, but they do not establish a typical return across industries.

What Apollo’s Comparison Captures

Apollo published its “No Signs of AI Payoff Outside the Tech Sector” note on October 24, 2026. The analysis was reported by TechCrunch, PYMNTS and Bloomberg the following day. Its central comparison is between seven large technology companies and the remaining S&P 500 firms, using revenue per employee, margins and productivity indicators. The title’s “zero payoff” is a shorthand for a lack of visible aggregate gains outside tech, not a controlled estimate of AI’s financial impact.

That distinction matters because a flat average can conceal companies whose gains and losses offset one another. The comparison also cannot separate AI from changes in interest rates, labor costs, tariffs or other business conditions. Ramp’s spending figures cover its own customers, while Apollo’s productivity evidence relies on broad economic indicators. Neither source, as described in the note, tracks the return on every firm’s AI investment.

Some companies are spending at a scale well beyond individual software trials. Meta’s commitments cover leases and supply contracts extending into the future, while forecasts cited for Amazon, Alphabet, Meta, Microsoft and Oracle project nearly $780 billion in 2026 capital spending and more than $1 trillion annually from 2027 through 2030. Those are forecasts, not amounts already spent or contractual obligations. Slok has also described venture-funded AI spending as “circularity,” reflecting concern that investment and spending may reinforce one another without yet producing equivalent returns for the wider economy.

“starting to view token spending not as a source of growth but as a cost center that weighs on margins”

— Torsten Slok, in the October 24 note

How Much AI Returns Are Hidden

The size and timing of economy-wide AI returns remain uncertain. Apollo’s comparisons show no broad productivity boost in the cited measures so far, but they do not prove that AI has generated no returns at individual firms. The note’s aggregate figures cannot distinguish AI effects from other economic factors or show whether gains are concentrated in narrow tasks.

It is also unclear how much of Meta’s roughly $825 billion in commitments will ultimately be used, when the costs will fall due, or what share is directly tied to AI. The total combines leases, contracts and potential purchases, and includes some non-AI activity. Reported capital expenditure is a separate measure from future commitments.

The workplace studies cited in the source show results that vary by task, including no sales change in some retail experiments. Their findings should not be treated as a forecast for every deployment. The source material also offers no complete accounting of returns across industries, nor a verified outcome for the financing pressures facing AI infrastructure providers.

Evidence to Watch in Coming Quarters

Investors and company leaders will be watching upcoming earnings reports for changes in capital spending, cash flow, margins and revenue per employee. Those measures could show whether heavy infrastructure investment is followed by financial returns, though they will not by themselves isolate AI’s contribution. Meta’s spending outlook and its disclosures about lease and supply commitments are among the figures likely to draw continued attention.

Further evidence may come from productivity data and company-level reporting on how AI tools affect specific work. The key test is whether measured gains in tasks such as customer support translate into sustained savings or additional revenue after implementation costs. Until then, the evidence supports a distinction: spending is visible and rising, while returns vary by use and remain difficult to identify across the wider corporate economy.

Key Questions

Did Apollo prove that AI has generated no financial returns?

No. Apollo’s “zero payoff” wording summarizes aggregate trends, including flat revenue per employee among 493 S&P 500 companies and no apparent AI boost in the cited productivity measures. Those comparisons do not directly measure AI’s impact at each company and can hide gains and losses that offset one another.

What does Meta’s $825 billion figure represent?

It is a total of reported future leases, contractual commitments and possible cloud capacity purchases, mainly involving data centers, servers, cloud capacity and related infrastructure. It is not a bill already paid, and the source says not all of the commitments are AI-related.

Are there measurable workplace benefits from AI?

Some task-level studies cited in the report found gains. Customer support workers resolved 14% more issues per hour, with a larger reported gain among novices, while retail experiments recorded sales changes from zero to 16.3% depending on the workflow. These results do not establish a typical return for all businesses.

Why is AI financing receiving more attention?

Building computing capacity requires large investments in chips, data centers and cloud infrastructure. The report cites a $40 billion SpaceX borrowing for AI chips and says insurance costs for debt from Meta and CoreWeave rose. Those signs have drawn attention to funding costs, but do not establish that the companies cannot finance their plans.

Source: ThorstenMeyerAI.com