Economists can already count the money going into AI infrastructure. What they cannot yet say with confidence is how much that spending will become lasting productivity growth, who will benefit, or how quickly the gains will spread. AI can support investment and output while the broader economy still slows; those are not contradictory forecasts.
The challenge, as The Conference Board’s Erik Lundh told Computerworld on January 8, 2026, is estimating the relationship between AI and productivity. An updated view requires separating observable activity from modeled expectations—and treating forecasts as conditional scenarios, not promises.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
Principles of Economics | $270.79 | Buy on Amazon |
| 2 |
|
Basic Economics: A Common Sense Guide to the Economy | $25.49 | Buy on Amazon |
| 3 |
|
Economics For Dummies: Book + Chapter Quizzes Online | $17.49 | Buy on Amazon |
| 4 |
|
Principles of Economics | $177.99 | Buy on Amazon |
| 5 |
|
Understanding Economics, Student Edition (ECONOMICS PRINCIPLES & PRACTIC) | $49.02 | Buy on Amazon |
What economists mean when they forecast AI’s economic effect
“AI will boost the economy” can refer to several different things. A forecast may be about AI-related investment, the revenue of AI businesses, output per worker, total-factor productivity, employment, or total GDP. Those measures are connected, but they are not interchangeable.
- GDP growth measures the change in total production over a period. It reflects many forces, including consumption, investment, government activity, trade, labor supply, and productivity.
- Labor productivity measures output relative to labor input. It is commonly expressed as output per hour or output per worker. The two differ when average hours change.
- Total-factor productivity (TFP) is the part of output growth not explained by measured changes in labor and capital. It can capture technology and organizational improvements, but also measurement errors and other influences.
- Capital deepening occurs when workers have more or better capital to work with—for example, computing equipment or software. It can lift output per worker even if TFP itself does not rise.
- Potential output is an estimate of how much an economy can produce sustainably. It is not the same as actual GDP in a particular year.
- AI-sector revenue or AI investment measures activity associated with AI suppliers and users; neither alone establishes that the whole economy has become more productive.
A forecast can assume AI raises productivity and still project modest total GDP growth if labor supply, demographics, energy costs, trade, or other conditions weaken. “AI contributes positively to productivity” is therefore a narrower claim than “AI causes total GDP growth to accelerate.”
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →#1 Best Overall
How AI enters an economic model
A simplified production function is Y = A × F(K, L): output (Y) depends on capital (K), labor (L), and the efficiency with which they are combined (A, often associated with TFP).
- Capital: Data centers, chips, networking equipment, software, robotics, and power infrastructure add to or improve the economy’s capital stock.
- Labor: AI can change the number of workers and hours required, the tasks people perform, and the skills firms demand. It may substitute for some tasks while complementing workers on others.
- Productivity: AI could let firms make more with the same inputs, or produce the same output with fewer inputs. Whether either result appears in measured productivity depends on adoption, workflow changes, and reliable measurement.
Attribution is hard even at one company. If output rises after AI adoption, the cause might also include new management practices, better data, process redesign, more staff, higher demand, or another technology. Economists need a counterfactual—what would have happened without AI—to estimate AI’s contribution.
What economists can measure now—and what those measures do not prove
Investment and deployment leave visible traces before their eventual payoff is known. Economists and businesses can track:
- Purchases of servers, GPUs, networking equipment, and other computing hardware;
- Data-center construction, related power projects, and electricity demand;
- Corporate research and development, AI-service revenue, and investment flows;
- AI adoption rates by industry, and prices for model access or inference;
- Changes in output per hour and employment in occupations exposed to AI; and
- Firm-level experiments that compare particular AI-supported tasks or processes.
These are inputs, activity measures, or proxies—not proof of broad productivity gains. The Conference Board’s July 16, 2026 global update linked growth in some economies to spending on ICT equipment, R&D, and other AI-adjacent goods and services. That supports the conclusion that such spending is economically active; it does not establish how much durable economy-wide productivity it will generate. (See The Conference Board’s global forecast update.)
Why AI’s productivity contribution is hard to see
National accounts do not label every AI-assisted activity
Economic statistics classify production by industries and products, not by whether a worker or process used AI. AI-related value can be spread across software, computing, manufacturing, utilities, finance, professional services, and other industries. A rise in one industry’s output does not neatly isolate the value AI added across them.
Rank #2
Quality improvements complicate price measurement
A model may become more capable while the price of using it falls. If statistical methods do not fully capture that quality change, measured real output can miss some improvement. But a dramatic quality-adjusted estimate depends on how researchers define the product, quality, and price adjustment; it should not be read as an official GDP growth rate.
Valuable intangible assets may be difficult to count
Training data, organizational know-how, proprietary workflows, and the effort needed to integrate a model can create value without resembling conventional physical capital. Some of those costs may appear in business spending, but the economic asset and its contribution can be difficult to identify consistently.
Productivity can arrive after investment
Buying equipment is not the same as using it effectively. Firms may need to retrain workers, redesign workflows, connect systems, and change management practices. Those complementary steps take time, so measured productivity can lag the initial investment.
Jobs conceal task-level changes
A job can remain in place even as AI takes over some tasks, changes the skills required, or lets a worker handle more cases. Headcount alone may therefore miss meaningful changes in work, hours, pay, and output.
Technology and adoption assumptions can go stale
Model capabilities, inference costs, hardware, regulation, and deployment practices can change quickly. A long-range forecast necessarily rests on assumptions about a future that may not resemble the conditions used to construct it.
Rank #3
To illustrate the measurement challenge, Peterson Institute for International Economics (PIIE) researchers have proposed an experimental “AI GDP” framework. They estimated nominal US AI GDP at about $250 billion in 2025, while a separate working paper estimated quality-adjusted AI GDP growth at roughly 2,600% per year. These are preliminary, framework-dependent research estimates—not official national-account figures or directly comparable measures of ordinary GDP growth. See PIIE’s discussion of AI in GDP statistics and its working paper on measuring the AI economy.
Why investment may show up before productivity
- Businesses and governments commit money to computing capacity, data centers, chips, and related infrastructure.
- Construction and equipment purchases are recorded as investment and can support measured demand before the assets produce much output.
- Companies experiment with AI and bear the costs of integrating it into existing systems and processes.
- They redesign work, train staff, and adjust operations; some organizations do this more successfully or quickly than others.
- If AI enables more output from the same inputs, productivity gains may emerge later and unevenly.
- Some projects may not earn adequate returns, and some capacity may be underused or made obsolete by newer technology.
Lundh’s infrastructure comparison is useful: building infrastructure contributes to economic activity during construction, while efficiency benefits depend on putting it to productive use. AI capital expenditure can therefore raise current investment even if its future productivity payoff disappoints. The Conference Board interview sets out this distinction in its discussion of AI and forecasting uncertainty.
Will AI replace workers, augment them, or create new work?
Economists do not yet know whether firms will mainly use AI to reduce headcount while maintaining revenue, or keep workers and raise output per worker. Several outcomes can occur at once, across different firms and tasks:
| Scenario | Firm response | Possible economic effect |
|---|---|---|
| Replacement | AI performs tasks previously done by workers. | Demand for labor in exposed tasks may weaken, putting pressure on some jobs or wages. |
| Augmentation | Workers use AI to do more or improve the quality of their work. | Output per worker can rise even if employment remains stable. |
| Expansion | Lower costs support new products or increase demand for existing services. | New demand can create work in complementary activities, though the net employment effect is uncertain. |
| Restructuring | Firms reorganize roles and processes around AI. | Gains may take longer to appear but can become more durable if the redesign works. |
| Concentration | A small group of leading firms captures much of the benefit. | Profits may rise without gains spreading broadly to workers, consumers, or less productive firms. |
| Diffusion | Affordable tools spread across companies, including smaller businesses. | Productivity gains may reach more industries, though adoption and complementary investment still matter. |
These paths are not mutually exclusive. A firm can automate routine tasks, expand a service because it has become cheaper, and retain workers for judgment-intensive work. Aggregate productivity can rise even while particular workers or regions lose income or bargaining power.
Why services may feel the effects before physical industries
Many service tasks involve digital information, language, and repeatable processes, and can be tested using existing computers and software. That makes early applications plausible in customer support, accounting, legal research, paralegal work, software development, marketing, insurance claims, administration, and financial analysis. Because services make up a large share of the US economy, changes there could matter even before widespread physical automation.
Rank #4
Manufacturing, logistics, construction, agriculture, and warehousing may also benefit from AI and robotics, but physical deployment involves equipment, safety, reliability, capital, and regulatory constraints that do not apply in the same way to a digital workflow. In the Computerworld interview, Lundh describes services as a likely earlier site of disruption than work that requires physical robots.
Free tools Windows power users keep installed
One-click scans. No signup required.
Why more efficient AI research could mean more R&D spending
AI could lower the cost of finding or testing ideas, but that does not settle what happens to total research spending. If a company can obtain the same result with fewer researchers or less expenditure, unit costs fall. If cheaper research raises the expected return or makes more ambitious projects viable, the company may increase its overall R&D budget. AI could also accelerate simulation and design while physical testing, regulation, manufacturing, or clinical trials remain bottlenecks. Lower cost per experiment and higher total spending can coexist.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What current forecasts say—and what they do not say
Current institutional forecasts treat AI-related activity as one influence on growth, alongside other forces. They are not forecasts of an automatic AI boom.
| Institution and date | Forecast | How to read it |
|---|---|---|
| The Conference Board, July 16, 2026 | Global GDP growth of 2.8% in 2026 and 3.0% in 2027. | The Board said AI-adjacent spending was supporting growth but not fully offsetting the effects of war. The figures are overall growth forecasts, not estimates of AI’s isolated contribution. Source. |
| PIIE, spring 2026 | Global growth of 3.0% in 2026 and 3.1% in 2027; US real GDP growth of 2.0% in 2026 and 1.9% in 2027. | These forecasts incorporate assumptions about war, energy, labor, and policy as well as AI-related investment and consumption. Source. |
The Conference Board also said in its July 16, 2026 US forecast that AI productivity gains are visible at the macro level, while their magnitude for firms and workers remains uncertain. That assessment is compatible with cautious projections: evidence of a contribution does not establish its size, distribution, or future rate. (See The Conference Board’s US forecast.)
These updated figures should not be confused with the longer-range projection reported in the January 2026 Computerworld interview: an average annual US GDP-growth projection of 1.9% for 2025–2039, compared there with average growth of 2.4% from 2000–2024. Those are the interview’s long-horizon figures, not the latest annual forecasts. The Conference Board’s 2025–2039 Global Economic Outlook addresses the long-range horizon.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Why countries may experience AI differently
Emerging economies: access opportunity, industrialization risk
AI tools could make translation, tutoring, software, medical information, and business expertise cheaper to access. Countries may gain productivity without building frontier models themselves, and smaller firms may reach digital customers or export services more easily.
The risk is that automation and robotics could reduce the advantage of low-cost labor in manufacturing, a traditional route from lower incomes to higher-value production. Vietnam, Bangladesh, Kenya, and parts of sub-Saharan Africa are examples raised in the Computerworld interview in connection with this tension; the outcome is not predetermined. Countries’ ability to benefit depends on more than access to an app:
- Reliable electricity, connectivity, cloud access, and computing infrastructure;
- Access to chips and the ability to afford AI services;
- Education, technical skills, and workers able to complement AI;
- Domestic research and business capacity to adapt tools to local needs; and
- The ability to capture value from AI-enabled services rather than only purchase them from foreign providers.
United States and China: different constraints and advantages
The US and China are both described in the interview as near the leading edge of AI development, but their economic outcomes depend on more than research talent. Investment intensity, energy supply, cloud and data infrastructure, industrial structure, government support, and commercialization all matter. For China, access to advanced chips, domestic alternatives, and geopolitical restrictions are additional uncertainties. A forecast that assumes the same deployment path for both countries would miss those differences.
What could make an AI growth forecast wrong?
Upside and downside cases should identify the conditions behind them. A forecast may disappoint if adoption is slower than expected, businesses do not redesign workflows, enterprise or consumer demand is weak, or investment builds more compute capacity than customers will use. Energy constraints, chip restrictions, regulation, and rapid hardware or model changes can alter costs and deployment. On the other hand, faster capability gains, falling prices, or broader diffusion could make productivity effects larger or earlier than a baseline assumes.
Recommended Free Tools
Even if output rises, the distribution matters. Gains concentrated among a handful of firms may produce a different experience for workers and consumers than productivity improvements that spread across industries. Labor displacement can coexist with higher aggregate output, and higher productivity does not automatically translate into higher wages.
How to judge an AI forecast
Before relying on a forecast for business planning, investment decisions, or economic commentary, ask:
Quick Recap
- What is being measured? Investment, AI-company revenue, output per hour, employment, potential output, or GDP?
- What is the horizon and counterfactual? Is the claim about the next year or the next decade, and what is the comparison without AI?
- Which assumptions are observed? Separate current spending and adoption data from modeled expectations about future productivity.
- How fast must adoption occur? Does the forecast account for training, integration, and organizational redesign?
- How are prices and quality handled? Is a figure nominal spending or a quality-adjusted estimate?
- Are the bottlenecks included? Check assumptions about chips, power, infrastructure, regulation, and physical deployment.
- Who captures the gains? Aggregate GDP can conceal differences among workers, firms, consumers, and countries.
- Does the forecast show a range? Look for baseline, upside, and downside scenarios, sensitivity analysis, and clear conditions that would change the estimate.
- Will assumptions be revised? In a fast-changing field, transparent updates are more useful than false precision.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
