What’s Really Driving The Surge In AI Spending?

🔍 Read the full analysis: What’s Really Driving The Surge In AI Spending? on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

Apollo chief economist Torsten Slok says aggregate data show no clear financial payoff from artificial intelligence for companies outside the technology sector, even as major firms make large commitments to computing infrastructure. The contrast is sharpening scrutiny of whether the growing AI investment bill will produce returns beyond a small group of technology companies.

Slok’s October 24 Apollo Daily Spark note, titled “No Signs of AI Payoff Outside the Tech Sector,” compares the seven largest technology companies—Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla—with the other 493 companies in the S&P 500. Apollo’s charts show revenue per worker and profit margins rising for the seven, while those measures have remained broadly flat for the rest. The note also cites San Francisco Federal Reserve productivity measures that, it says, show no sign yet of AI lifting productivity growth.

Those figures do not establish that AI caused the gap, or that no company outside tech has benefited. They describe aggregate trends, which can conceal firms or departments with gains and others with losses. The cited Ramp spending data are also drawn from Ramp’s own business customers, not the S&P 500 as a whole. Slok’s “zero financial payoff” is an interpretation of the evidence, rather than a direct measurement isolating AI’s effect.

The investment commitments are easier to count. Meta’s October 22 filing lists about $355.73 billion in leases that have not started and $454.52 billion in non-cancelable contractual commitments, largely tied to cloud capacity, servers and network equipment. It also reports up to $14.72 billion in contingent cloud purchases. Together, those categories total roughly $825 billion; they extend over multiple years, and are not all exclusively for AI. Meta separately forecast $135 billion to $145 billion in capital spending during 2026.

The Growing Cost of AI Capacity

The figures put two questions side by side: who is paying for AI infrastructure, and where the resulting services are generating enough value to justify the expense. Meta reported $34.91 billion in capital expenditures in the third quarter. With operating cash flow of $32.79 billion, its free cash flow was negative $2.12 billion, compared with positive $10.6 billion a year earlier. That one quarter does not determine the long-term return on its investments, but it shows the near-term cash demands involved.

Funding is becoming another part of the story. The supplied report says SpaceX borrowed $40 billion to buy AI chips, while insurance costs on debt issued by Meta and CoreWeave have risen. These details point to growing attention on how infrastructure spending is financed, not just on the availability of chips and power. They do not, by themselves, establish that borrowing is unmanageable or that expected returns will fail to materialize.

For companies adopting AI, broad averages may be a poor guide to whether a particular use case works. The source cites research reporting 14% more customer-support issues resolved per hour, rising to 34% for novice workers, and retail experiments with sales effects ranging from zero to 16.3%, depending on the workflow. It also reports that 83% of nearly 9,000 executives surveyed saw no productivity impact so far, with an average impact of 0.2% a year. These results cover different settings and measures; they should not be treated as a single estimate of AI’s overall economic return.

From AI Hype to Measured Returns

Slok’s note follows a period of exceptionally large investment plans by major technology companies. The supplied source cites projections that Amazon, Alphabet, Meta, Microsoft and Oracle together could spend nearly $780 billion in 2026 and more than $1 trillion annually from 2027 through 2030. Those figures are forecasts, not amounts already spent or contractual obligations. Meta’s roughly $825 billion figure, by contrast, combines several categories of future commitments listed in its filing.

Apollo also points to the concentration of AI-related spending. Based on Ramp data covering more than 85,000 U.S. business customers, its note says 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. The same note says 80.4% of the top 1% of AI spenders are venture-backed, compared with 13.1% of the median spender. These figures describe spending within Ramp’s dataset, not the entire U.S. economy.

There is also evidence of gains in specific tasks, alongside the lack of a clear economy-wide productivity signal. That distinction matters: improvements in a narrow workflow may not yet show up in company-wide margins or national productivity statistics, particularly while firms are paying for implementation and infrastructure. The source reports that Slok has described some venture-funded token spending as “circularity” and has questioned whether the cost of compute exceeds the value it delivers. Those are his assessments, not settled findings about every AI product or business.

“No Signs of AI Payoff Outside the Tech Sector”

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

What the Aggregate Data Cannot Show

The available figures do not isolate AI as the cause of changes in company revenue, margins or productivity. A flat average for 493 S&P 500 companies can mask gains and losses, and the data may also reflect other forces, including labor costs, interest rates and tariffs. The source does not provide a method for separating those effects. Nor does Ramp’s customer spending data establish how AI spending is distributed across all businesses.

It is also not yet clear how much of Meta’s commitments will be used specifically for AI, how much capacity will ultimately be taken up, or what returns those investments will generate. Meta’s listed commitments include infrastructure and other uses, including Reality Labs hardware, and span several years. The cited task-level productivity and sales findings do not show whether gains persist at scale or exceed the cost of deployment.

Finally, the borrowing and debt-insurance details signal financial scrutiny but do not settle whether the sector’s investment is sustainable. The supplied material gives limited detail on SpaceX’s loan terms and does not establish the cause of changing insurance costs for Meta and CoreWeave debt.

Returns Will Be Tested Over Time

The next test is whether infrastructure spending turns into sustained revenue, stronger margins or measurable productivity improvements. Investors and companies will be able to track spending against results through future earnings reports, capital-expenditure guidance and disclosures about capacity and AI-related demand. Meta’s 2026 spending outlook is one near-term marker, but the company’s full commitments stretch much further.

For businesses using AI, the clearest evidence is likely to come from results tied to specific tasks: output, quality, time saved and operating costs measured against the cost of tools and computing. The broad evidence cited by Apollo currently shows no clear payoff across most non-tech companies, while selected use cases show gains. Whether those isolated improvements spread widely—and produce returns large enough to support continued investment—remains an open question.

Key Questions

What does Apollo mean by “zero financial payoff”?

It is Slok’s interpretation of aggregate trends showing rising revenue per worker and margins for seven major technology companies, with broadly flat measures for the other 493 S&P 500 firms. The figures do not measure AI’s effect in isolation or prove that no individual company has benefited.

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

Meta’s filing lists about $355.73 billion in leases that have not started, $454.52 billion in non-cancelable contractual commitments, and up to $14.72 billion in contingent cloud purchases. The amounts cover multiple years and are not all exclusively for AI.

Are there examples of AI improving business results?

The supplied source cites task-level findings, including 14% more customer-support issues resolved per hour and retail sales effects ranging from zero to 16.3%, depending on the workflow. These results do not establish a broad return across companies or industries.

Does the evidence show that AI investment is unprofitable?

No. The cited aggregate data show that broad gains are not yet visible in the measures Apollo examined. They do not establish future returns, account for every company or isolate AI from other economic factors.

Source: ThorstenMeyerAI.com