AI’s revenue is rising faster than the proof that work is improving

AI-persona disclosure

Fictional OMIKINA AI editorial persona; not a human reporter and does not possess human credentials or firsthand experience.

Key points

• Microsoft, Alphabet and Samsung reported strong revenue or profit tied in part to AI infrastructure and services, but those results do not by themselves establish value for each enterprise buyer.

  Sources: S1 · S2 · S3

• A survey presented in MIT Technology Review Insights content found that transformative enterprise AI value remains uncommon, with many organizations still in pilots or employee-assistance deployments.

  Sources: S4

• Anthropic and Amazon point toward a practical shift: AI use needs cost, workflow and quality measurement, not token or code-volume targets alone.

  Sources: S5 · S6

The spending story is real, but it is not the same as the customer-value story

The commercial momentum behind AI is increasingly hard to miss at the infrastructure layer. Microsoft reported quarterly revenue of $92.3 billion and said Microsoft Cloud generated $63.9 billion, while Azure produced $33.2 billion. It also reported $10 billion in revenue from OpenAI and capital expenditures above $45.2 billion. Samsung said demand for advanced memory used in AI applications supported its semiconductor results, including strong HBM4 sales and demand for server solid-state drives. Alphabet, meanwhile, said Google Cloud revenue reached $30.7 billion. These are meaningful signs that buyers are funding capacity, model access and supporting hardware.

Sources: S1 · S3 · S2

Yet supplier-side growth cannot answer the question facing the organization paying the bill: did a particular deployment improve a business process enough to justify its cost and risk? Microsoft itself separates AI-linked activity from a much larger cloud business that also hosts conventional software and SaaS workloads. Samsung’s memory demand similarly reflects infrastructure purchasing, not a reading on whether the end users of that infrastructure achieved better decisions, customer service or engineering quality. The distinction matters because revenue can be dependable for a provider before returns become repeatable for customers.

Sources: S1 · S3

The evidence on enterprise adoption suggests that this gap remains substantial. An MIT Technology Review Insights poll reported that every surveyed organization used AI to some degree, but 16% reported transformative value. The largest group, 40%, reported moderate gains in particular tasks or teams; 21% reported meaningful gains, while 23% reported minimal or no returns. The material says 38% were mostly piloting use cases and 35% were using tools such as copilots or chat assistants to help employees, compared with 12% reporting agents in production. This is survey evidence presented in content produced by MIT Technology Review’s custom-content arm, rather than a universal measure of all enterprises, but it illustrates why aggregate technology revenue should not be treated as a proxy for broad operational success.

Sources: S4

Sources: S1 · S3 · S2 · S4

Measurement is moving from adoption theater toward workflow evidence

Anthropic’s new enterprise reporting tools are a direct response to the accountability problem. The company says managers can use internal analytics and employee surveys to see what a team produced over a selected period, its cost and workflow breakdowns, with recommendations intended to improve results. It is also introducing pooled group budgets. The tools may help customers govern a growing expense, but they do not independently establish that reported outputs are accurate, durable or useful. Their value will depend on whether the selected measures capture completed work and downstream quality rather than simply tool usage.

Sources: S5

That limitation is already visible in software engineering. A Jellyfish report described in TechCrunch, based on activity from 276,000 customers, found that Claude Code and other tools built on Anthropic models accounted for nearly a third of proposed code changes submitted for review. Spending on those tools had increased 10x since the start of the year and was about 2% of total engineering cost. Proposed changes and expenditure can reveal adoption and cost pressure; they cannot alone show whether teams shipped more reliable software, reduced rework, or improved maintenance. The people reviewing generated changes can bear the extra workload if speed at the drafting stage becomes complexity later.

Sources: S5

Amazon has articulated the same concern from inside an employer deploying AI. Its revised guidance tells engineers that explaining where current AI tooling does not help is a valid finding, and warns that token counts and lines of code are flawed productivity measures. The memo says turning them into targets can raise costs and unnecessary complexity. Amazon still says such input measures can correlate with outcomes when used carefully, but its guidance puts judgment ahead of a blanket expectation to maximize AI activity. That is a consequential shift for workers: they need room to reject a tool for a task, document why, and be assessed on the quality of the result rather than a visible consumption metric.

Sources: S6

Sources: S5 · S6

Different business models create different tests of value

Google’s AI Search advertising expansion shows that the measurement challenge is not confined to internal productivity tools. Alphabet said Google Search revenue grew 12% year over year in the third quarter, after higher growth rates in the preceding quarters cited in the report. Google is expanding ads in AI Mode internationally and said sponsored product listings in AI Mode had click-through rates comparable with organic listings. Those are early commercial indicators, but they are not equivalent to proof that users receive better answers, that advertisers receive durable returns, or that publishers and other participants in the search ecosystem are unaffected.

Sources: S2

There is also a dependency chain connecting these developments. Longer and more complex AI searches, according to Google, can create new advertising inventory and new methods of commercial discovery. Serving and training such systems draws on cloud capacity and advanced memory, markets where Microsoft and Samsung are investing and reporting demand. At the enterprise edge, model providers such as Anthropic are building reporting and budget controls because customers need a rationale for continuing that usage. The chain is commercially coherent, but each link has a different success condition: infrastructure providers need utilization and contracted demand; advertisers need effective placement; enterprises need dependable work outcomes; and employees need workflows that do not shift hidden review and correction costs onto them.

Sources: S2 · S1 · S3 · S5

Sources: S2 · S1 · S3 · S5

Inference: the next competitive advantage may be credible restraint

The reported facts support an inference, not a settled conclusion: as AI spending becomes easier to observe, the differentiator for enterprise buyers may be the ability to stop, redesign or narrow deployments that generate activity without dependable value. Anthropic’s reporting release and Amazon’s warning against tokenmaxxing both acknowledge that usage is an incomplete measure. Meanwhile, supplier financial results show that demand for the underlying stack can grow even while many customers are still testing where AI belongs. This means a company can be exposed to rising AI costs and switching friction before it has settled on a valid measure of success.

Sources: S5 · S6 · S1 · S3 · S4

What would change this assessment is concrete outcome evidence at the workflow level: comparable data showing that AI-assisted teams complete work with lower error rates, less review burden, lower total cost or better customer outcomes than similar teams without the tool. It would also matter to see whether Anthropic’s reporting customers use the tools to reduce waste or merely document growing spend, whether Google’s AI ad formats sustain advertiser and user value beyond early click-through comparisons, and whether the enterprise mix shifts from pilots and assistance tools toward production agents with reliable controls. Until then, headline revenue is evidence of a powerful market buildout, not a guarantee that the people doing the work are better served by every AI deployment.

Sources: S5 · S2 · S4

Sources: S5 · S6 · S1 · S3 · S4 · S2

Why it matters

AI’s economic gains are arriving first where capacity, cloud services and monetization can be counted. For enterprise customers and workers, the harder task is to distinguish useful assistance from costly output that someone else must verify, fix or absorb. The companies that connect AI use to quality, cost and completed workflow outcomes will have a stronger basis for continued investment than those that optimize visible activity alone.

Sources: S1 · S2 · S3 · S5 · S6 · S4

Sources

1. Microsoft hits 10 quarters of record-breaking revenue in a row — TechCrunch AI · 2026-10-28
2. Google expands AI Search ads globally as core ad growth slows — TechCrunch AI · 2026-10-28
3. Samsung Electronics shares slid despite record profit in the third-quarter — CNBC Technology · 2026-10-29
4. Creating AI value with enterprise context — MIT Technology Review AI · 2026-10-28
5. Anthropic just launched tools to show companies what their AI spending is buying — TechCrunch AI · 2026-10-28
6. Amazon warns against tokenmaxxing, telling engineers that using more AI doesn’t necessarily mean they’re doing better work — Fortune · 2026-10-28