What happened
Apollo published a Daily Spark analysis on October 24, 2026, titled “No Signs of AI Payoff Outside the Tech Sector.” The note, by Chief Economist Torsten Slok, argues that the financial effects of artificial intelligence remain concentrated among the largest technology companies rather than spread across the rest of the US corporate market.
Its central comparison is between the “Mag 7” and the remaining companies in the S&P 500, described as the “S&P 493.” Apollo says revenue per worker is rising for the Mag 7 but flat for the S&P 493. It makes the same distinction on profitability: profit margins are rising for the Mag 7 and flat for the rest of the index. The source also says there are still no signs that AI is boosting productivity growth.
From those observations, Slok draws a business-spending conclusion. Mainstream corporate America is pulling back, he writes, because token spending is being viewed as a cost center that weighs on margins rather than as a source of growth. The note contrasts that position with venture-backed companies, which it says are spending increasingly on tokens. In Apollo’s account, established companies generally do not need the latest model and are unwilling to pay for it.
The analysis argues that macroeconomic effects will remain limited while agentic AI and token spending are confined to a small part of the economy. Its forward-looking question is whether cheaper intelligence can ultimately translate into higher revenue growth and higher margins in mainstream companies, even if models themselves become commoditized.
Apollo does not present that translation as automatic. The note characterizes AI as potentially useful like a calculator, but argues that business performance also depends on judgment and context. It lists accountability, trust, culture, unwritten rules, ethical trade-offs, and fairness as factors that shape reputational risk and, ultimately, revenue and margins. Its conclusion is deliberately broad: the AI story will be decided in the S&P 493 rather than only the Mag 7, and Apollo says the evidence so far does not show improving revenue growth or margins outside technology.
The source identifies Bloomberg, Macrobond, the Federal Reserve Bank of San Francisco, the US Bureau of Labor Statistics, Ramp’s AI Index, and Apollo analysis among the sources for its charts. It also includes Apollo’s standard disclosure that the statements are current judgments and may change, and that Apollo does not warrant completeness or accuracy.
What the archive adds
This is a new fact on a known story: market readings of AI infrastructure exposure have already been volatile. Nvidia, Oracle, CoreWeave shares fell following OpenAI revenue details focused on how revenue definitions shaped that reading; hotter inflation data revived rate fears and sent AI chip stocks lower made financing conditions visible in the data-center buildout. Apollo shifts the question to operating results beyond technology.
Trend signal
No supplied site pattern genuinely fits this macroeconomic assessment. It is evidence about adoption outcomes and corporate financial results, rather than a change in agent architecture, protocols, security, or software capabilities.
Why now
The timing follows a period of massive investment and widespread AI attention, according to Apollo. Its point is that the relevant test is no longer whether companies can buy tokens or deploy models, but whether those purchases appear in revenue per worker, margins, or productivity outside the biggest technology firms.
Connections to other events
The archive’s recent market stories frame two adjacent constraints. The OpenAI-revenue episode showed that investors can reassess infrastructure exposure when reported revenue is defined differently than expected. The inflation-driven selloff showed that financing conditions can constrain data-center expansion. Apollo adds a third issue: whether non-technology customers generate enough business value to justify continuing spend.
Why it matters
Teams building with AI should treat model access and pilot adoption as intermediate signals, not proof of value. Apollo’s argument suggests that an AI budget will face more scrutiny when it is classified alongside recurring software and operating costs rather than tied to a measurable commercial or operational result.
That does not require every use case to produce an immediate company-wide productivity statistic. It does require a credible chain from a deployed workflow to an outcome that an operator can inspect: less time on a bounded process, fewer errors, higher conversion, faster resolution, or a service level that could not otherwise be delivered. Builders have influence over that chain through workflow design, evaluation, permissions, and handoffs. A system that is technically impressive but cannot be connected to an accountable business owner and a measured result will be easier to cut when costs are reviewed.
The source also complicates a simple “cheaper models solve adoption” narrative. Lower inference cost may remove one obstacle, but it does not establish judgment, trust, or accountability. Those are organizational constraints around an AI system, not merely model-quality constraints.
What happens next
- More AI projects will be evaluated against explicit margin, revenue, or productivity measures rather than usage alone. (6 months, 65%). Watch for: corporate disclosures or procurement rules that tie AI renewals to named business outcomes.
- The gap between technology-sector AI economics and broader enterprise results will remain a central investor question. (12 months, 60%). Watch for: earnings commentary separating AI revenue from AI operating costs and capex.
Limits and uncertainty
This is an economist’s interpretation, not a product-usage study or a causal experiment. The supplied material summarizes aggregate comparisons and chart sources, but does not provide the underlying chart values, definitions for every measure, company-level results, or a sector-by-sector account of deployments. It therefore cannot establish that AI caused flat results, that every non-technology company is reducing spend, or that individual AI workflows lack value.
The note also does not test the quality of particular models, agents, or implementations. Its claims concern broad financial and productivity indicators and should not be read as a verdict on a specific team’s deployment. Apollo explicitly says its opinions and projections are subject to change and offers no assurance that the trends will continue.
Skill to build now
For one deployed AI workflow, create a one-page outcome ledger this week. Name the accountable owner, the task boundary, the baseline, token and tool costs, the human review required, and one business metric that could move. Review it after a fixed number of completed cases, not after a model demo. This turns an AI expense into a claim that can be tested, revised, or stopped.
Reader takeaway
Apollo’s signal is not that AI has no value. It is that broad corporate value remains unproven in the measures that sustain budgets: revenue per worker, margins, and productivity. For builders, the durable response is to make each workflow legible as completed work with accountable, measurable consequences.
Open question
Which enterprise AI workflows will first show repeatable gains in revenue or margins outside the large technology companies, and what measurement design will make those gains credible?