AI Spending Keeps Exploding: What The Returns Show

🔍 Read the full analysis: AI Spending Keeps Exploding: What The Returns Show on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get privacy and security gear 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 chief economist Torsten Slok says corporate data show no clear financial payoff from AI outside major technology companies, even as firms commit heavily to data centers, computing capacity and chips. The conclusion, published in an October 24 Apollo note, describes broad economic indicators; it does not establish that AI has produced no gains at individual companies or in specific tasks.

Apollo compared revenue per worker and profit margins at the seven largest technology companies—Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla—with those of the other 493 S&P 500 companies. The note’s charts show revenue per worker rising for the seven while remaining broadly flat for the rest, and a similar separation in margins. Apollo also cited San Francisco Federal Reserve productivity measures that, it said, show no visible AI boost to productivity growth.

At the same time, companies are taking on substantial infrastructure costs. Meta’s October 22 quarterly filing lists about $355.73 billion in leases not yet started and $454.52 billion in non-cancelable contractual commitments, largely for cloud capacity, servers and network equipment. Including up to $14.72 billion in contingent cloud purchases brings the total to roughly $825 billion. The commitments extend over time, are not all AI-related, and are not the same as cash already spent.

Meta reported $34.91 billion in capital expenditures for the third quarter and narrowed its 2026 capital-spending outlook to $135 billion to $145 billion. 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. Separately, the source reports that SpaceX borrowed $40 billion to buy AI chips, illustrating how financing is becoming part of the debate alongside chip supply and power availability.

Spending Rises Faster Than Visible Returns

The gap between large infrastructure bills and uncertain economy-wide returns matters to investors, businesses and customers. If expensive computing capacity does not translate into higher output or revenue, it could weigh on corporate margins and make continued investment harder to justify. Apollo’s argument is that the payoff is not yet evident in broad measures for most large public companies—not that no company benefits.

The evidence also points to a concentration problem. Apollo cites Ramp data from more than 85,000 U.S. business customers: the top 10% of firms accounted for 99.5% of model-serving and inference spending, 99.0% of neocloud spending and 95.6% of spending on OpenAI and Anthropic. Those figures describe Ramp’s customers, not the whole economy, but they suggest adoption and spending are uneven. If only a small group is buying substantial AI services, broad productivity effects may take time to appear in national or corporate averages.

That does not rule out practical gains. The source cites customer-support research reporting 14% more issues resolved per hour, rising to 34% for novice workers, and retail experiments with sales effects ranging from no increase to 16.3%, depending on workflow. These task-level results can matter to individual employers, but they do not by themselves show that AI has lifted overall corporate productivity or delivered a positive return after costs.

What Apollo’s Figures Capture

Slok’s October 24 note, titled “No Signs of AI Payoff Outside the Tech Sector,” drew attention to a contrast between technology companies and the rest of the S&P 500. Its measures include revenue per worker, profit margins and productivity indicators. The source says the seven technology companies were near $780,000 in inflation-adjusted revenue per employee in the third quarter of 2026, while the other 493 were around $525,000 and had remained in a similar band since 2021. Apollo’s chart put the seven’s margins near 30%, compared with roughly 10% for the other firms.

Those comparisons are observations about groups over time, not a controlled test of AI’s effect. A flat average can conceal companies whose gains and losses offset one another, and it cannot isolate AI from other factors such as labor costs, interest rates or tariffs. The Ramp spending statistics also cover Ramp customers, rather than the same set of S&P 500 companies.

The spending figures require similar care. Meta’s commitments include capacity and infrastructure scheduled for future periods, with some obligations running through 2036; they should not be read as an immediate bill or as entirely AI spending. Estimates that the five largest technology spenders could invest nearly $780 billion in 2026 and more than $1 trillion annually from 2027 through 2030 are forecasts cited by Investing.com, not amounts already committed or spent.

“Artificial intelligence has delivered zero financial payoff to corporate America outside of tech.”

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

How Much AI Pays Back

The aggregate data do not establish causation. Apollo’s comparisons cannot determine how much of the difference between technology companies and other firms is due to AI, or whether gains in some businesses are being obscured by flat results elsewhere. Productivity measures may also take time to reflect new technology, while the source does not provide a controlled comparison of AI adopters and non-adopters.

The return on the infrastructure spending remains uncertain. Meta’s future commitments cover multiple years and include items that are not exclusively for AI; the source does not establish what share will be used for AI workloads or what revenue those investments will generate. It also cites reports that insuring debt from Meta and CoreWeave became more expensive, but does not provide enough detail here to determine the scale or cause of that change.

The source says 83% of nearly 9,000 executives reported no productivity impact so far, with an average impact of 0.2% a year. The survey’s precise timing, question wording and methodology are not included in the supplied material, so those figures should be treated as reported survey results rather than a definitive measure of AI’s effect across all businesses.

Evidence to Watch in Coming Quarters

Investors and executives will look for evidence that AI spending is translating into measurable revenue, cost savings or output per worker. Company filings can show whether capital expenditure and future commitments keep rising, while results and productivity measures may indicate whether the investment is producing returns. A single quarter or broad market average is unlikely to settle the question.

For Meta, the next reported financial results and updates to its capital-spending plans will help show how its infrastructure commitments develop. Across cloud providers and other large technology companies, financing costs and access to capital will also matter as they build capacity. The available source material does not identify a specific next release or milestone that will directly resolve Apollo’s claim.

For businesses outside the technology sector, the relevant test is more specific: whether an AI tool improves a defined workflow enough to exceed its costs. Apollo’s aggregate conclusion may change as adoption spreads and data accumulate, but the current evidence supports a gap between heavy investment and a clearly visible economy-wide payoff, not a final verdict on AI’s long-term returns.

Key Questions

What does Apollo mean by saying AI has delivered no payoff outside tech?

Apollo’s Torsten Slok is referring to aggregate indicators, including revenue per worker, margins and productivity measures. The claim describes trends in broad groups; it does not prove that every non-tech company has received no benefit.

Does Meta’s roughly $825 billion figure mean it has already spent that amount?

No. The figure combines future lease obligations, contractual commitments and possible cloud purchases. Many obligations extend over several years, and the total is not all AI-related or cash already paid.

Are there examples of AI producing measurable gains?

Yes. The source cites customer-support research showing 14% more issues resolved per hour, with a larger gain for novice workers, and retail experiments reporting sales effects from zero to 16.3%, depending on the workflow. These results do not establish a broad economy-wide return.

Why might company-level gains not show up in productivity statistics yet?

Benefits may be concentrated in particular teams or tasks, offset by costs elsewhere, or too recent to register in broad measures. The source also notes that aggregate averages cannot separate AI’s effects from other economic changes.

What should readers watch for next?

Watch company filings and earnings for spending plans, revenue tied to AI services, cost savings and output per worker. Those indicators can help show whether infrastructure investment is producing returns, though no single result will settle the wider question.

Source: ThorstenMeyerAI.com