How AI Spending Keeps Growing While Returns Stay Uneven

🔍 Read the full analysis: How AI Spending Keeps Growing While Returns Stay Uneven on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the latest gadgets 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

AI spending is expanding even as broad financial returns remain hard to measure: a new Apollo analysis says revenue per worker and profit margins have risen for major technology companies but stayed broadly flat for the other S&P 500 firms, while Meta’s filings show about $825 billion in future commitments, mostly tied to data centers, servers and cloud capacity. The figures capture a growing divide between investment in AI infrastructure and evidence of business-wide gains; they do not establish that AI has produced no benefits in individual companies or tasks.

Apollo chief economist Torsten Slok set out the analysis in an October 24 note titled “No Signs of AI Payoff Outside the Tech Sector,” which was reported by several outlets over the following days. The charts compare the seven largest technology companies—Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla—with the other 493 S&P 500 companies. Apollo’s figures show revenue per worker and profit margins rising for the seven, while the broader group has remained comparatively flat.

The source report reads Apollo’s chart as showing revenue per worker near $780,000 for the seven in the third quarter of 2026, against about $525,000 for the other companies. It puts the technology group’s margins at roughly 30%, compared with around 10% for the rest. Those gaps describe group-level outcomes; they do not by themselves show that AI caused the difference. The report also cites San Francisco Federal Reserve productivity measures that show no detectable boost to productivity growth from AI so far.

The spending picture is more directly visible in corporate disclosures. Meta reported $34.91 billion in capital expenditures in the third quarter and narrowed its 2026 spending outlook to $135 billion–$145 billion. Its operating cash flow was $32.79 billion, while reported free cash flow was negative $2.12 billion, compared with positive $10.6 billion a year earlier. Meta’s October 22 filing lists approximately $355.73 billion in leases that have not started and $454.52 billion in non-cancelable contractual commitments, plus as much as $14.72 billion in contingent cloud purchases. The amounts cover commitments extending over time and are not all exclusively for AI.

The Cost of AI’s Uneven Payoff

The gap matters because companies are committing large sums to computing capacity before evidence of broad productivity gains has emerged. Those costs can affect cash flow, margins and financing needs, particularly if expected business benefits arrive slowly or are concentrated in a small number of firms. Meta’s negative third-quarter free cash flow and large future obligations offer a concrete example of how investment plans can place pressure on a company’s finances even when revenue is growing.

But the aggregate picture is not a verdict on every AI deployment. The source material cites research finding that customer-support workers resolved 14% more issues per hour with AI assistance, with a 34% gain for novice workers. Retail experiments showed sales effects ranging from no increase to 16.3%, depending on the workflow. These task-level findings suggest that returns may depend on how a tool is integrated, what work it changes and how outcomes are measured.

The contrast also matters to investors and business leaders evaluating whether spending can be sustained. If returns remain concentrated among technology firms and venture-backed companies, broad corporate spending may not yet justify the scale of infrastructure plans. If useful gains spread to more functions, current aggregate statistics could take time to reflect them. Neither outcome is established by the data cited here.

What Apollo’s Figures Compare

Apollo’s headline conclusion comes from aggregate indicators, including revenue per worker, profit margins and productivity measures. A flat average for 493 companies can conceal individual businesses that gained or lost; gains and declines may offset one another. It also cannot isolate AI’s effect from other influences such as labor costs, interest rates or tariffs. Slok’s “zero” is an interpretation of the broad pattern, not a direct measurement showing that every non-technology business received no financial benefit.

The report also cites spending data from Ramp, a corporate card and spending platform with more than 85,000 US business customers. Ramp’s figures indicate that the top 10% of firms account for most spending on model inference, neocloud services and OpenAI and Anthropic. Those customers are not the same sample as the S&P 493, so the figures offer a separate view of concentrated adoption rather than a direct explanation for the index-wide results.

Meanwhile, a $40 billion borrowing by SpaceX to buy AI chips, cited in the source report, points to the scale of financing needed across the infrastructure market. The report also notes that forecasts for capital spending by major technology companies are projections, not confirmed outlays. The distinction matters: announced plans, contractual obligations and completed spending are different measures of the eventual bill.

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

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

How Much AI Pays Back

The available aggregate figures do not establish how much AI itself caused changes in company revenue, margins or productivity. Apollo’s comparison is not a controlled test, and the cited productivity measures have not yet detected a broad AI-driven improvement. The source report also does not provide a full accounting of AI-related revenue against spending for each company.

Meta’s roughly $825 billion figure combines different kinds of future obligations, including leases, contracts and contingent purchases; some items are not solely for AI. It is not an amount paid in a single year. The report says Meta’s commitments rose by about $125 billion, but the exact contribution of AI versus other infrastructure and hardware plans is not fully broken out in the supplied material.

It is also unclear whether task-level productivity gains will spread across industries, persist at scale or translate into higher profits. The claim that 83% of nearly 9,000 executives have yet to see a productivity impact, also cited by the source, is a survey result and not a measure of every firm’s experience. Its interpretation depends on the survey’s sample and timing.

The Next Test Is Measurable Returns

The next evidence will come from companies’ quarterly disclosures: capital spending, cash flow, margins and productivity will show whether infrastructure investments are producing financial results. Investors will also watch whether Meta’s planned spending remains within its $135 billion–$145 billion 2026 outlook and how the company’s large contractual obligations develop.

For businesses adopting AI, the more informative tests are likely to be tied to specific workflows: whether tools reduce costs, increase output or improve sales enough to exceed their operating and computing expenses. The source material does not identify a single upcoming milestone that will settle the wider debate. For now, spending plans and commitments are measurable, while the scale and distribution of returns remain developing questions.

Key Questions

What does Apollo mean by “zero financial payoff”?

It is Torsten Slok’s description of aggregate trends: Apollo’s analysis shows revenue per worker and margins rising among seven major technology companies while remaining broadly flat for the other 493 S&P 500 firms. It does not prove that no individual company or task has gained from AI.

What makes up Meta’s roughly $825 billion in commitments?

Meta’s filing lists about $355.73 billion in leases that have not started and $454.52 billion in non-cancelable contractual commitments, plus up to $14.72 billion in contingent cloud purchases. The categories cover infrastructure and other needs over time, and the source says not all of the amount is AI spending.

Has AI improved productivity?

The productivity measures cited by Apollo show no broad boost detected so far. Separate studies cited in the source report show gains in some customer-support and retail tasks, which do not establish an economy-wide effect.

Why do some businesses report gains while the overall numbers are flat?

Benefits may be concentrated in particular companies or workflows, while other firms see little effect or incur added costs. An aggregate average can hide both winners and losers, and Apollo’s comparison cannot isolate AI from other economic factors.

Source: ThorstenMeyerAI.com