Everything About AI
NextFin News - The biggest risk in artificial intelligence is no longer whether the technology works. It is whether the companies buying it can organize themselves well enough to get paid for it. Adoption is surging across corporate America while measurable returns are flatlining — a gap that is now wide enough to threaten the economics of the entire AI investment boom.
The two sides of the divide are both well documented. A Federal Reserve staff analysis published in April found that work-related generative-AI adoption among U.S. workers reached roughly 41 percent by late 2025, and about 78 percent of the labor force now works at a firm that has adopted AI. Yet the latest annual global survey from McKinsey, released in August, found that the share of organizations reporting any enterprise-level profit impact from AI has not budged: 37 percent attribute at least some EBIT impact to AI, essentially unchanged from the prior year, and the proportion of "AI high performers" — those deriving 5 percent or more of EBIT from AI — has stayed flat at about 6 percent.
This is the tension behind the question of what stands in the way of AI adoption. The obstacle is not the model. It is the organization standing between the pilot and the profit.
The Pilot Trap: Why Adoption Numbers Lie
The surface story of AI adoption is a triumph. McKinsey's research shows 89 percent of organizations now use AI in at least one business function, and 44 percent say the technology has been scaled across the organization, up from 38 percent a year earlier. Nearly all executives — 97 percent — say their companies deployed AI agents in the past year, and 52 percent of employees are already using them, according to WRITER's 2026 enterprise survey conducted with Workplace Intelligence. Usage is deep: 70 percent of employees and 94 percent of the C-suite use AI tools at least 30 minutes a day, with nearly two-thirds of executives logging two hours or more.
But adoption, spending, and value are three different numbers, and most corporate dashboards report only the first two. WRITER's 2026 survey found that 79 percent of organizations now face challenges adopting AI — a double-digit increase from 2025 — and 54 percent of C-suite executives admit that adopting AI is tearing their companies apart. The same research found that while 97 percent of executives report benefiting from AI personally, only 29 percent see significant organizational return on investment.
The mismatch is structural, not accidental. Companies have been rewarded for speed — for announcing pilots, setting up centers of excellence, and telling investors they will not be left behind. But the bill has arrived, and the revenue has not followed it. Companies are preparing to more than double AI spending in 2026, from an average of 0.8 percent of revenue to 1.7 percent, according to BCG. Yet a 2026 survey of 396 enterprises on AI cost governance found that roughly 80 to 85 percent of organizations miss their AI infrastructure forecasts by more than 25 percent. Spending is compounding faster than the ability to predict it.
The pattern is not a bubble in the sense of assets priced for perfection. It is a bifurcation: a handful of organizations are redesigning work around AI and capturing value, while the majority are adding a tool to existing processes and calling it transformation.
The Mechanism: Individual Gains That Never Reach the P&L
The reason the middle is missing has a name, and it shows up cleanly in the data. Eighty percent of respondents in McKinsey's 2026 survey say AI has improved their individual productivity, and half say it helps them make better decisions. But individual productivity is not enterprise profitability. The translation from the keyboard to the income statement is where the value leaks out.
Consider what has to be true for an AI gain to become an EBIT gain. An employee uses a tool to draft a report in 20 minutes instead of an hour. But the output then has to be checked for errors, reformatted to fit a legacy system, reconciled with data the model could not access, and defended in a review chain that does not fully trust automated work. The time saved at the keyboard is consumed at the handoff. The gain is real; it just never reaches the P&L.
The deeper mechanism is that AI capability is advancing faster than organizational capability. That asymmetry is why the redesign question dominates the technology question.
AI is more of a leadership than a technology challenge.
Dan Taylor, Google's VP of Global Ads, said. The organizations that succeed are not the ones with the most powerful models; they are the ones that redesigned the workflow, the data pipeline, the governance, and the accountability around the model. That redesign requirement is what makes the risk stubborn. It is not a matter of waiting for the next model release. It is a matter of fixing data quality, breaking down silos, establishing governance, closing skills gaps, and integrating AI into actual workflows.
The trust data confirms how far most firms are from that redesign. A Harvard Business Review Analytic Services study found that only 6 percent of companies fully trust AI agents to handle core business processes. Twenty percent say their technology infrastructure is fully ready to support agentic AI for core processes; 15 percent say the same about their data and systems; and just 12 percent feel their risk and governance controls are fully in place. Even so, 72 percent of respondents say the benefits of agentic AI outweigh the risks — enthusiasm running ahead of readiness.
The shift to agentic AI makes the organizational bar higher, not lower. Gartner predicts that by 2028, 33 percent of enterprise software applications will include agentic AI, up from less than 1 percent in 2024. Agents can make decisions, coordinate tasks, and complete multi-step workflows with limited human involvement. If enterprise data contains gaps or errors, agents do not merely produce a bad draft — they take bad actions at scale. That is why governance is not a brake on AI adoption; it is the condition for it. AI-specific governance roles grew 17 percent in 2025, and the share of businesses with no responsible-AI policies fell sharply from 24 percent to 11 percent — but the gap remains wide enough that most executives hesitate to turn agents loose on mission-critical work.
The asymmetry between what technology can do and what organizations can absorb was put plainly by Alicia Mullery, a VP analyst at Gartner, speaking at the firm's IT Symposium in October 2025.
While not all AI is ready to deliver value, humans are even less ready to capture value.
The line captures the core problem: the technology is ready to find value; the organization is not ready to capture it.
The Second-Order Risk: A Capex Boom Without an Earnings Engine
The first-order consequence of the implementation gap is disappointing returns at individual companies. The second-order consequence is what should worry the broader market: a capital-expenditure boom that is not matched by an earnings engine.
A market-sizing analysis by a major financial-data and research firm projects that generative AI will become a $2.3 trillion market by 2032, expanding to 22 percent of total technology spending as enterprises accelerate adoption of AI-powered software, infrastructure, and services. That forecast assumes adoption converts into revenue and profit at the enterprise level. If the conversion rate stays near current levels — 37 percent reporting any EBIT impact, only 6 percent at high-performer status — the spending trajectory becomes harder to justify on fundamentals.
There is also a new wrinkle that could reshape the demand side. Nearly a third of respondents (32 percent) in McKinsey's 2026 survey report that their organizations have decided against buying one or more software products because they could be built internally with agentic coding tools. The same shift is visible in infrastructure: about 31 percent of large enterprises are now scaling coding agents. If enterprises start building rather than buying, and if AI-related operating costs are already constraining usage for about 20 percent of respondents, the revenue assumptions embedded in the hyperscaler build-out look increasingly dependent on a conversion rate that has not arrived.
The risk propagates through three channels. First, the chipmakers and cloud providers whose revenues depend on continued infrastructure build-out face a demand shock if enterprise customers pull back on experiments that do not pay. Second, companies that borrowed against promised AI productivity to justify margins will face a credibility gap when guidance does not materialize. Third, labor markets may price in displacement that does not arrive on the expected timeline, because organizations cannot operationalize the automation fast enough to remove the headcount — McKinsey found that the share of respondents expecting AI-related workforce declines last year was about double the share reporting actual declines over the past year.
This is already visible at the margin. When the first wave of belt-tightening hits the experimenters rather than the producers, it is a signal that the return hurdle has moved from "can we build it" to "can we prove it pays." An IBM Institute for Business Value study of 2,000 CEOs found that only 25 percent of AI initiatives had delivered the return on investment executives expected over the prior three years — a figure that sits uneasily beside the commitment to keep spending more.
The Counter-Thesis: This Is Just an Early-Cycle Lag
The strongest case against this reading is that it mistakes a normal technology adoption curve for a structural failure. General-purpose technologies — electrification, the internet, cloud computing — all required a long build-out of infrastructure before the application layer delivered measurable productivity gains. The Solow paradox of the computer age eventually resolved itself. AI is simply in its wiring phase.
Gartner's position supports the patience argument: AI's impact on global jobs will be neutral through 2026, and by 2036 AI solutions introduced to augment or autonomously deliver tasks will result in more than 500 million net new human jobs supporting new AI initiatives. A July 2025 survey of more than 700 CIOs found that by 2030, none of IT work will be done by humans without AI, 75 percent will be done by humans augmented with AI, and 25 percent will be done by AI alone. On this view, the current ROI drought is the price of admission to a decade-long transformation, and the companies rationing spend today are the ones that will be under-invested tomorrow.
The counter-thesis has force, but it rests on a timing assumption that the data is starting to test. The gap between 89 percent adoption and 6 percent high performance is not a one-year phenomenon; it has persisted across multiple survey cycles. Electrification eventually rewired the factory. The question is whether today's adopters are rewiring the firm or merely plugging a chatbot into a 20th-century process. The former produces a J-curve; the latter produces a flat line.
The falsifying signal is concrete. If, by 2027, the share of organizations reporting more than 5 percent of EBIT impact from AI does not rise materially above the current 6 percent high-performer base — in McKinsey-style annual surveys — then the implementation gap is not an early-cycle lag. It is a structural feature of AI adoption, and the market is wrong to underwrite a capex boom on the assumption that returns will follow automatically.
What Comes Next: Three Horizons, Three Scenarios
Short term (6 to 12 months): sentiment and spending discipline. Expect more companies to announce AI cost controls, pilot rationalization, and a shift from experimentation to ROI gating. The exposed are the pure-play AI vendors whose valuations assume uninterrupted budget growth; the beneficiaries are the vendors that can tie their tools to measurable workflow outcomes. Watch quarterly CFO commentary on AI spend and any guidance withdrawals tied to AI productivity assumptions.
Medium term (1 to 3 years): the bifurcation widens. The base case is a two-tier market: a minority of firms that redesigned work around AI and are compounding productivity gains, and a majority still stuck in the pilot-to-production gap. Winners will be identifiable by one metric — revenue or margin per employee trending up alongside AI spend, rather than spend rising alone. The exposed are companies that cannot show the linkage.
Long term (3 to 10 years): structural, but not for everyone. If the redesign requirement is real, AI adoption will follow the pattern of enterprise software more than the pattern of a commodity boom: slow, uneven, and won by operators rather than speculators. The upside case is that agentic AI lowers the integration burden enough to compress the adoption curve. The downside case is that governance, liability, and data-quality constraints keep most enterprises in a permanent pilot state, and the $2.3 trillion market forecast proves too optimistic.
The practical implication for investors is to stop asking whether a company is "doing AI" and start asking whether it has redesigned the workflow that AI touches. Adoption is a cost line. Redesign is a margin story. Only one of them shows up in earnings.
The biggest AI adoption risk is not that the technology disappoints. It is that companies treat a redesign problem as a purchasing decision — and then wonder why the invoice arrived without the return.
Explore more exclusive insights at nextfin.ai.
