Everything About AI
NextFin News - Nine out of ten companies report no measurable productivity gain from their AI spending after three years, yet the same executives expect AI to lift productivity by 1.4% and cut employment by 0.7% over the next three. That gap between what boards are seeing today and what they are betting on tomorrow is the defining tension of the 2026 AI cycle - and it is producing a dangerous conclusion: that waiting for proof of payoff is the prudent course. It is not. By the time the productivity data turns, the competitive order will already be fixed.
The Invisible Payoff: Why the Paradox Is Not a Reason to Wait
The numbers are stark enough to make any chief executive hesitate. A survey of nearly 6,000 CFOs, CEOs and senior executives across the United States, United Kingdom, Germany and Australia - conducted by the Federal Reserve Bank of Atlanta, the Bank of England, the Deutsche Bundesbank and Macquarie University between November 2025 and January 2026 - found that more than 80% of firms detected no impact on either employment or productivity from AI over the previous three years. Average executive AI use sits at just 1.5 hours a week, and a quarter of executives report not using AI at all.
Adoption, meanwhile, looks close to universal. Stanford's 2026 AI Index puts organizational adoption at 88%, up from 78% a year earlier, with generative AI in use in at least one business function at 70% of organizations. WRITER's 2026 enterprise survey, covering 1,200 C-suite executives and 1,200 employees, found 79% of organizations facing challenges in adopting AI - a double-digit increase from 2025 - and 54% of C-suite executives admitting that adopting AI is "tearing their company apart." Nearly half, 48%, call AI adoption a "massive disappointment," up from 34% a year earlier. Only 29% see significant return on investment from generative AI, and just 23% from AI agents.
Against that backdrop, the capital being poured in looks almost irrational. The four largest hyperscalers - Amazon, Google, Meta and Microsoft - plan up to $630 billion in capital expenditure in 2026, a 62% increase from the record $388 billion spent in 2025, with Amazon alone guiding to $200 billion. Broadening to the five largest builders including Oracle, 2026 capex lands between $660 billion and $725 billion, with roughly three-quarters tied directly to AI infrastructure. Wall Street's consensus estimate reaches $930 billion by 2028 and $4.7 trillion across 2025-2030. Investors have noticed the mismatch: shares of Google, Amazon and Microsoft sold off after their earnings calls as the spending guidance landed.
So the picture is contradictory: adoption is everywhere, measured returns are almost nowhere, and the money keeps accelerating. The natural response - wait until the returns show up before committing - is exactly the response this moment punishes.
The Paradox Is Real, but the Reading of It Is Wrong
The first question is whether the absence of measured productivity is a signal about the technology or about the measurement. The evidence points to the latter, and to timing.
Robert Solow's 1987 observation - "You can see the computer age everywhere but in the productivity statistics" - was not resolved by computers suddenly becoming productive. It was resolved by measurement and reorganization catching up to investment. Total factor productivity growth in the United States accelerated only in the early 1990s, years after IT investment had become widespread, and the information technology industry was among the fastest-productivity-growth sectors from 2000 through 2022. The lag was not a verdict on the technology; it was the time required for firms to stop bolting new machines onto old workflows.
The same mechanism is visible today, and the future-tense nature of the bet is now quantifiable. Research from the Federal Reserve Bank of St. Louis, analyzing nearly 490,000 earnings-call transcripts from 5,198 publicly traded U.S. firms between 2000 and 2025, found that approximately 95% of AI-related productivity sentences describe gains executives expect in the future rather than gains already realized - a share that has held steady since 2023. The Atlanta Fed paper's own forward-looking data shows firms expect AI to add 1.4% to productivity over the next three years - a modest number, but one that implies executives see the payoff as ahead of them, not behind them. The paradox is not that AI fails to deliver; it is that delivery arrives on a different clock than the quarterly reporting cycle.
Task-Level Gains Are Real; Firm-Level Gains Require Redesign
The second question is why individual gains fail to scale. Here the data is unambiguous: the gains exist at the task level and evaporate at the firm level because most firms have adopted tools without redesigning the work around them.
WRITER's survey found AI super-users are five times more productive than colleagues slow to adopt - yet only 29% of organizations see significant ROI from generative AI. Boston Consulting Group's 2026 Global AI at Work report, surveying nearly 12,000 frontline employees, found 42% reporting eight hours of saved time a week - a full workday - while 66% said they received limited or no guidance on what to do with the hours they saved, and half said they were not using that saved time for more strategic work. The time is being freed; the organization does not know what to do with it.
The historical analogy is precise. When factories first electrified in the early 20th century, managers simply replaced the central steam engine with an electric motor and kept the line-shaft-and-belt layout intact. Productivity barely moved for more than three decades. Only when manufacturers reimagined the factory around decentralized power - single-floor layouts, flow-based organization, new management models - did the payoff arrive. Diffusion of electric lighting and motors did not reach the 50% mark until the 1920s, and factory electrification could explain half, at most, of the roughly 5-percentage-point acceleration in the average annual rate of aggregate industrial productivity growth during that decade. Economist Paul David's 1990 paper "The Dynamo and the Computer" made the point explicit: managers had overlaid "one technical system upon a preexisting stratum."
That is the 2026 condition in one sentence: companies are running AI through organizational layouts designed for a world without it. The productivity is not missing; it is deferred behind a redesign that most firms have not started.
Layoffs are not a viable AI strategy. The leaders who are putting in the work to radically redesign operations with human-agent collaboration at the center are the ones compounding their advantage in ways competitors can't replicate. AI transformation is ultimately about people, and the future belongs to the companies putting agent-building power directly into the hands of people closest to the work.
- May Habib, CEO and co-founder of WRITER
The Invisible Phase Is Where Advantage Compounds
This is the heart of the argument, and the part most boards get backwards. Competitive advantage in a general-purpose technology does not accrue when the returns become visible. It accrues while they are still invisible, because that is when the inputs that later become unreplicable are cheapest to acquire.
Three assets compound during the invisible phase, and all three are harder to buy later. First, proprietary data and the workflows that structure it: an organization that embeds AI into its processes now generates the usage data, error logs and refinement loops that make its systems better than any off-the-shelf model a laggard can purchase in 2028. Second, organizational learning: the super-users, the internal playbooks, the governance routines, the agent infrastructure. WRITER found 92% of the C-suite is already cultivating an "AI elite" - employees who are three times more likely to have received a raise or promotion - while 60% plan layoffs for non-adopters. That is not just a training program; it is the formation of a two-tier workforce whose capability gap widens with every quarter. Third, cost basis: compute, power contracts and talent are priced for uncertainty today; they will be priced for proven demand tomorrow.
The asymmetry is brutal. A firm that waits for the productivity data to turn is not buying the same thing later at a slightly higher price. It is buying a commodity - the model - after the scarce inputs have already been claimed. The technology commoditizes; the advantage does not.
The second-order implication runs through the equity market, and it cuts against the surface reading of the capex numbers. The firms with visible revenue today are not the AI adopters waiting for productivity to show up - they are the infrastructure suppliers: semiconductors, data centers, power equipment, cooling. Their order books are already full because the buildout is real regardless of whether the payoff arrives on schedule. The companies most exposed are the incumbent adopters whose strategies are "more for show" - 75% of executives admit as much - and whose margins will compress twice over: first from the cost of AI they cannot yet justify, then from competitors who have already locked in the data flywheel and the reorganized workflow. The capex cycle and the firm-level strategy are not the same bet, and the market is pricing them as if they were.
The Counter-Thesis - and Why It Is Partly Right
The strongest case against this view is not sentimental; it is financial, and it deserves its due. The hyperscaler capex buildout - nearly $690 billion in combined 2026 guidance from the major builders, with consensus pointing to $4.7 trillion across the decade - is underwritten entirely by a future productivity wave that has not yet materialized. If that wave does not arrive, the write-downs will be historic, and the firms that moved early will be holding stranded infrastructure and stranded talent. The St. Louis Fed has noted that technology diffusion across regions, occupations and firms is "extremely slow," typically unfolding over 20 to 30 years, and that compressing AI's lag to three to five years "would be a huge change" from historical precedent. A board that waits is not necessarily cowardly; it may be demanding evidence before underwriting a decade-long J-curve.
That caution is correct as a view of the aggregate capex cycle. It is wrong as a guide to firm-level strategy. The two are not the same bet. The capex thesis requires the macro productivity wave to arrive on schedule; the early-adopter thesis only requires that adoption, once it does arrive, rewards those who have already reorganized. Even in the computer era, the winners were not the firms that spent the most on hardware in the 1980s; they were the firms that had redesigned their operations to exploit the hardware when the productivity data finally turned. Waiting for the data is a hedge against a capex bubble, not a strategy for competitive survival. The board that sits out the invisible phase protects its balance sheet from a possible bubble and forfeits its market position for certain.
What to Watch: Scenarios and the Signal That Breaks the Thesis
The cyclical-versus-structural call is this: the absence of measured productivity is cyclical - a diffusion and measurement lag that will revert as firms reorganize, on the historical timetable of decades, not quarters. But the distributional consequence is structural. The gap between adopters and laggards, once established during the invisible phase, does not self-correct when the data improves; it hardens.
Who benefits and who is exposed follows directly. Beneficiaries are firms using the quiet period to redesign workflows end-to-end rather than sprinkle AI across existing processes, those building proprietary data flywheels, and the infrastructure suppliers whose revenues are already visible even while the productivity payoff is not. The exposed are firms treating AI as a cost-cutting lever alone, those whose strategy is "more for show," and any organization that assumes it can compress a decade of organizational learning into a single late procurement cycle.
The forward view splits by horizon. In the short term, expect continued disappointment at the aggregate level: the more-than-80% of firms reporting no impact will not flip quickly, and capex skepticism will produce volatility in the builders' shares. Over the medium term, the split will widen - the 29% of firms already seeing significant ROI will pull away from the 48% calling adoption a disappointment, and the gap will show up in margins and market share before it shows up in economy-wide productivity data. Over the long term, the electrification precedent suggests the payoff arrives not when the technology improves, but when the organization finally matches it.
Three scenarios frame the path. The base case: productivity impact remains concentrated in roughly a third of firms through 2027, then broadens as reorganization spreads - winners are already identifiable by their workflow redesign, not their model choice. The upside case: agent deployment, currently in the single digits across nearly all business functions, crosses into mainstream enterprise use by 2027, pulling the laggards into the measured-gain column faster than the electrification precedent implies. The downside case: measured impact stays flat while capex stays above $500 billion annually - the bubble thesis wins, and the early movers write down assets even as their market positions hold.
The falsifying signal is specific: if by 2028 the share of firms reporting measurable AI productivity impact remains below roughly 40% - that is, if it fails to approximately double from current levels - while annual AI-related capex stays above $500 billion, the structural-advantage thesis breaks and the bubble thesis wins. Watch that ratio, not the capex absolute.
The productivity paradox is not a reason to wait. It is the reason the window is open. Advantage is built while the data is still ambiguous; by the time the numbers justify the investment, the investment is no longer what separates winners from laggards. "I told you so" will be true - and it will be too late.
Explore more exclusive insights at nextfin.ai.
