Could AI Lift U.S. Real GDP Growth to 4% in 2027?

Review period: September 19, 2026–September 19, 2036
Scope: Assessment of whether AI could lift U.S. real GDP growth to roughly 4% in 2027, covering forecasts, investment, productivity, constraints, and one-, five-, and ten-year effects.
Source policy: Priority to official U.S. statistics, Federal Reserve and government research, international institutions, peer-reviewed studies, and clearly identified private-sector forecasts.

Executive summary

Elon Musk’s September 18, 2026 claim—that AI could lift U.S. growth from roughly 2% to 4% in 2027—is possible as an upside scenario but substantially more optimistic than the evidence-based central case. Musk’s 4% figure exceeds current 2027 forecasts by 1.6 to 2.2 percentage points: the Federal Reserve projects 2.4%, the IMF and Philadelphia Fed forecasters 2.2%, the OECD and CBO 1.8%, and major private forecasts are generally around 2.1%–2.2%. (investing.com)

AI investment is already economically material. A Federal Reserve decomposition estimates that software, data centers, power facilities and computer equipment—after an adjustment for related imports—added about 0.73 percentage point to annualized GDP growth in 2026:Q1. But that contribution is volatile, and imported chips, servers and components can offset much of the gross investment boost. (federalreserve.gov)

The stronger part of Musk’s case is the unprecedented scale of the buildout. Data-center construction, computing-equipment investment, semiconductor output and hyperscaler capital expenditure have risen rapidly. The weaker part is the leap from investment to economy-wide productivity: official data show respectable but not explosive productivity growth, adoption intensity remains shallow at many firms, and Federal Reserve researchers found limited evidence as of mid-2026 that task-level AI gains had become a broad aggregate productivity transformation. (federalreserve.gov)

Bottom line: Reaching 4% in 2027 primarily because of AI would probably require the simultaneous realization of the upper end of investment estimates, a rapid productivity acceleration well beyond CBO’s baseline, strong domestic rather than imported supply, adequate power and grid capacity, continued consumer demand, and little offset from inflation or tighter monetary policy. That combination is credible enough not to be dismissed, but it is a bull case rather than the most plausible outcome. Sustaining growth near 4% after the investment surge would be considerably harder and would require persistent productivity growth, not merely more data-center construction.

1. What exactly is the claim?

Musk wrote on September 18, 2026 that his “guess” was that AI would roughly double U.S. GDP growth in 2027 from about 2% to 4%, potentially more. He did not specify whether he meant calendar-year average growth or fourth-quarter-over-fourth-quarter growth, nor did he provide an economic decomposition. (investing.com)

That ambiguity matters:

• Calendar-year average growth compares average output in 2027 with average output in 2026.
• Q4/Q4 growth compares output in the fourth quarter of 2027 with the fourth quarter of 2026.
• Federal Reserve and CBO projections generally use Q4/Q4 growth, whereas IMF and OECD headline forecasts normally use annual averages. The figures are therefore close comparisons, but not perfectly identical. (federalreserve.gov)

2. Mainstream 2027 growth forecasts

Sources: Federal Reserve, IMF, Philadelphia Fed, Goldman Sachs, U.S. Bank, OECD and CBO. (federalreserve.gov)

The forecasts cluster around 1.8%–2.4%, with an approximate midpoint of 2.2%. Musk’s scenario therefore requires roughly 1.8 percentage points of additional growth relative to the center of the mainstream range.

The Federal Reserve’s details put the divergence in sharper perspective. Its 2027 participant projections range from 2.0% to 2.9%; no participant projects 4%. However, the Fed’s historically based 70% forecast-error interval extends from 0.6% to 4.2%. Thus, 4% is within a broad statistical uncertainty interval, but outside policymakers’ actual modal judgments. (federalreserve.gov)

The Conference Board’s September outlook also emphasizes tighter financial conditions and downside risks for 2027, while acknowledging continued AI investment as an upside risk. (conference-board.org)

3. Four concepts that should not be conflated

3.1 AI raising the level of GDP

A level effect means the economy becomes permanently larger than it otherwise would have been, even if its growth rate eventually returns to normal. For example, an IMF model estimates that U.S. output could be 1.8% to 5.4% higher after ten years, depending on productivity assumptions. A 5.4% cumulative increase over ten years corresponds to an average annual growth increment of only about 0.5 percentage point, not 5.4 percentage points every year. (imf.org)

3.2 AI raising the annual growth rate

To raise annual GDP growth from 2% to 4%, AI must produce substantially more incremental output in 2027 than in 2026. Merely maintaining an already high level of AI activity does not accomplish this.

A sustained two-percentage-point growth increase would be historically large. The Federal Reserve’s estimate of longer-run U.S. growth is 2.0%, while CBO projects potential GDP growth averaging 2.1% during 2026–2030. (federalreserve.gov)

3.3 Temporary growth from capital investment

Building a data center, installing servers or constructing a power facility counts as current investment and directly raises GDP. But the growth contribution depends primarily on the change in investment, not simply on its high level.

If AI investment rises from $400 billion to $600 billion, that increase boosts growth. If spending then remains at $600 billion, it continues supporting the level of GDP but provides little additional year-over-year growth. If it falls to $500 billion, investment becomes a drag even though the absolute level remains historically high.

This dynamic appears in Federal Reserve data-center simulations. After adjusting for imports, the estimated 2027 growth contribution ranges from –0.08 percentage point in the pessimistic scenario to +0.53 percentage point in the optimistic scenario. Before the import adjustment, the range is much wider, from –0.17 to +1.21 points. (federalreserve.gov)

3.4 Sustainable productivity-driven growth

Sustainable growth requires AI to enable workers and capital to produce more output per hour or per unit of input. This can arise from:

• automating or accelerating tasks;
• improving decisions and product quality;
• reorganizing production;
• speeding research, software development and innovation;
• creating new products and services; and
• allowing capital and labor to move toward more productive activities.

Productivity improvements can persist after the construction boom ends, but typically require complementary investment in workflows, databases, worker training and organizational change. Federal Reserve research notes that general-purpose technologies have historically exhibited a productivity “J-curve,” with measured productivity gains lagging the initial investment phase. (federalreserve.gov)

4. How much is AI already contributing?

4.1 A Federal Reserve investment-accounting estimate

The Federal Reserve constructed a broad proxy covering software, data centers, power facilities, computers and peripherals, and associated net exports. The estimated contributions to annualized quarterly GDP growth were:

The totals are calculations from the Federal Reserve’s component data. They demonstrate both the significance and volatility of AI-adjacent investment, as well as the importance of imported equipment. The Fed cautions that these are broad categories, not clean AI line items, and therefore can include ordinary software and non-AI computing activity. (federalreserve.gov)

An alternative presentation used by Federal Reserve Vice Chair Philip Jefferson, which does not make the same net-import adjustment, estimated that capital expenditure likely related to AI contributed 1.40 points to annualized GDP growth in 2025:Q1, 0.85 in Q2, 0.26 in Q3, 0.84 in Q4 and 1.36 in 2026:Q1. The difference illustrates why gross spending and domestic GDP contribution should not be treated as interchangeable. (federalreserve.gov)

4.2 Scale of the buildout

Federal Reserve data show:

• Private data-center construction rose from an annualized $9.3 billion in January 2021 to $59.3 billion in May 2026, an increase of more than sixfold. (federalreserve.gov)
• Real private investment in computers and peripheral equipment rose from about $143.7 billion in 2021:Q1 to $384.2 billion in 2026:Q1, at seasonally adjusted annual rates. (federalreserve.gov)
• Semiconductor and electronic-component industrial production reached an index level of 188 in May 2026, up from approximately 139 at the end of 2023. (federalreserve.gov)
• Amazon, Alphabet, Meta, Microsoft, Oracle and CoreWeave together reported approximately $156 billion of capital expenditure in 2026:Q1, more than three times the comparable total in 2024:Q1; this sum is calculated from Federal Reserve-compiled SEC data. (federalreserve.gov)

A 2026 NBER paper reports that the five largest U.S. technology firms spent $380 billion on capital expenditure in 2025 and were expected to spend roughly twice that in 2026. (nber.org)

4.3 Latest aggregate data

Real GDP grew at a 1.5% annualized rate in 2026:Q2, down from 2.1% in Q1. Nonresidential investment remained strong, and real final sales to private domestic purchasers grew 4.2%, but rising imports and weaker government spending restrained headline GDP. (bea.gov)

The data therefore support the proposition that AI investment is an important growth engine, but not the proposition that it has already pushed total U.S. growth close to 4%.

5. What do productivity data show?

Aggregate productivity

Nonfarm business labor productivity increased 1.4% at an annualized quarterly rate in 2026:Q2 and 2.2% from a year earlier. Year-over-year productivity had accelerated to 2.9% in 2026:Q1 before moderating. (bls.gov)

Those figures are healthier than the weak productivity performance of much of the 2010s, but they are not yet an AI-driven breakout consistent with a sustained 4% economy. The OECD argues that recent U.S. strength may partly reflect AI-related capital deepening, while Federal Reserve researchers find that productivity trends across high-, medium- and low-AI-exposure sectors have not yet diverged enough to demonstrate a broad aggregate effect. (oecd.org)

Task-level evidence

Controlled experiments and field studies generally find that generative AI can raise output on particular tasks, including customer support, writing and programming. A peer-reviewed customer-service field study is among the strongest causal evidence of such gains, with especially large benefits for less-experienced workers. (academic.oup.com)

However, task productivity is not automatically firm productivity. A worker completing a task 10% faster does not necessarily make the whole firm 10% more productive if approval processes, customer demand, data quality, legal constraints or other production stages remain bottlenecks. Federal Reserve researchers explicitly identify this task-to-firm aggregation problem. (federalreserve.gov)

Firm surveys

A 2026 survey of almost 6,000 executives found that 69% of firms used AI, but executive use averaged only 1.5 hours per week; roughly nine in ten reported no measurable employment or productivity effect over the preceding three years. The same executives expected AI to raise firm productivity by an average of 1.4% over the next three years, output by 0.8%, and reduce employment by 0.7%. (nber.org)

A separate survey of nearly 750 executives found positive but heterogeneous productivity gains, concentrated in finance and high-skill services, while also documenting that perceived benefits exceeded measured benefits—possibly because revenue and organizational effects arrive with a lag. (nber.org)

The surveys support future gains, but challenge the idea that a sufficiently broad productivity surge is already locked in for 2027.

6. Estimated effects over one, five and ten years

The following ranges distinguish sourced institutional estimates from this report’s synthesis.

One-year evidence

CBO’s baseline attributes only about 0.1 percentage point per year of additional TFP growth to generative AI. Federal Reserve data-center scenarios imply a 2027 import-adjusted contribution between –0.08 and +0.53 point from data centers alone. Software, power infrastructure and domestic semiconductor production could add more, but some of those effects are already embedded in mainstream forecasts. (cbo.gov)

Consequently, a 0.3–1.2-point AI lift in 2027 is plausible as an analytical range, but it would still leave growth below 4% from most starting forecasts. A two-point lift requires both exceptionally strong investment acceleration and productivity effects arriving much sooner than the central institutional assumptions.

Five-year evidence

Executives’ average expectation of a 1.4% firm-productivity increase over three years is directionally consistent with a modest economy-wide level effect, although firm-level survey results cannot be directly aggregated into GDP. (nber.org)

CBO’s 0.1-point annual TFP increment would cumulate to approximately half a percent in nonfarm business output by around 2031. OECD and IMF estimates permit stronger effects if adoption broadens rapidly and AI capabilities continue improving. (cbo.gov)

Ten-year evidence

The principal estimates include:

• CBO: AI adds 0.1 percentage point to annual TFP growth during 2026–2036 and raises nonfarm business output by about 1% in 2036. (cbo.gov)
• Acemoglu-based estimate: roughly 0.12 percentage point of additional annual U.S. labor-productivity growth over ten years. (oecd.org)
• OECD: approximately 0.4–0.9 percentage point added to annual U.S. labor-productivity growth over a ten-year horizon in its earlier U.S. exercise; newer G7 scenario estimates span about 0.2–1.3 points. (oecd.org)
• IMF model: U.S. GDP 1.8% higher after ten years in a low-productivity scenario and 5.4% higher in a high-productivity scenario. (imf.org)
• Goldman Sachs estimate, as summarized by the IMF: about 1.5 points of additional labor-productivity growth, but only around 0.4 point of extra medium-term GDP growth after broader macroeconomic adjustments. (elibrary.imf.org)

A highly speculative NBER model calibrated to current AI investment produces additional cumulative GDP growth ranging from 5 to 58 percentage points by 2030. That result demonstrates the theoretical upside if investment correctly anticipates a rare technological boom, but it is not comparable to a conventional macroeconomic baseline and depends on large assumed AI-sector productivity shocks. (nber.org)

7. What must happen for 4% growth in 2027?

Starting from a roughly 2.2% mainstream midpoint, approximately 1.8 percentage points of additional growth is needed.

An illustrative—but deliberately optimistic—decomposition would be:

This is an analytical scenario, not an independent forecast. The components are difficult to measure and can overlap.

Growth-accounting hurdle

At a high level, real GDP growth is approximately the sum of growth in labor hours and output per hour. If aggregate hours grow only 0.5% in 2027, 4% GDP growth requires labor productivity growth near 3.5%. Even with 1% hours growth, productivity must rise about 3%. Current year-over-year nonfarm productivity growth is 2.2%, while professional forecasters expect relatively modest payroll gains and a broadly stable unemployment rate in 2027. (bls.gov)

Thus, Musk’s scenario requires both a clear productivity acceleration and continued positive labor input. If AI reduces entry-level hiring or total hours before the output benefits fully arrive, the productivity requirement becomes even larger.

Infrastructure conditions

Power availability is a binding constraint. The Department of Energy’s updated reference scenario projects U.S. data-center electricity consumption rising from 192 terawatt-hours in 2024 to 464 TWh in 2028 and 649 TWh in 2030, with data centers accounting for about 11.8% of total U.S. electricity use by 2030. DOE also identifies generation, transmission and interconnection constraints as barriers to rapid deployment. (energy.gov)

For the 4% scenario, projects already announced would need to move into service on schedule; electricity generation and transmission would have to expand quickly; chip, memory and networking-equipment supply could not become binding; and construction costs could not crowd out too much investment elsewhere.

Macroeconomic conditions

The productivity gains would also need to be sufficiently supply-enhancing that they do not simply generate inflation and a monetary-policy response. The September 2026 Fed projections show 2027 PCE inflation at 2.3%, core inflation at 2.5%, and a year-end federal funds rate of 4.1%, suggesting policymakers do not currently anticipate an accommodative environment designed to produce 4% demand growth. (federalreserve.gov)

8. Evidence supporting Musk’s claim

1. Investment is already large enough to affect national growth. Broad AI-adjacent categories have added between roughly 0.1 and 1.0 percentage point to quarterly annualized GDP growth after import adjustment since early 2025. (federalreserve.gov)

2. The project pipeline is unusually large. Federal Reserve simulations put 2027 data-center investment at an annualized $357 billion to $921 billion, depending on assumptions about future project announcements. (federalreserve.gov)

3. Recent productivity performance has improved. U.S. nonfarm productivity was 2.2% above its year-earlier level in 2026:Q2, stronger than much of the post-financial-crisis era. (bls.gov)

4. Task-level evidence is consistently positive in important knowledge-work settings. The strongest studies indicate that AI can improve speed and quality, particularly for less-experienced workers. (academic.oup.com)

5. Credible research contains a substantial upside tail. OECD estimates reach 0.9 point of annual U.S. labor-productivity growth in an optimistic ten-year case, while more speculative models permit much larger outcomes if AI becomes a genuinely transformative general-purpose technology. (oecd.org)

9. Evidence challenging the claim

1. Every major baseline forecast remains far below 4%. The mainstream range is 1.8%–2.4%, and the highest individual FOMC participant forecast for 2027 is 2.9%. (federalreserve.gov)

2. A meaningful part of the investment leaks into imports. Imports of chips, servers and related equipment can largely offset gross domestic capital spending in GDP accounting, as happened in 2025:Q1, 2025:Q4 and 2026:Q1. (federalreserve.gov)

3. Capex growth is not indefinitely repeatable. Once investment reaches a high level, maintaining that level no longer generates the same growth contribution. The Fed’s pessimistic 2027 data-center scenario produces a small negative contribution. (federalreserve.gov)

4. Broad productivity evidence remains limited. Federal Reserve analysis describes the economy as being in a buildout phase, with aggregate transformation and labor-market displacement still limited and concentrated. (federalreserve.gov)

5. Firm adoption is often shallow. High headline adoption rates coexist with limited usage intensity and little reported historical impact at most surveyed firms. (nber.org)

6. Power and transmission can constrain deployment. Data-center electricity demand is rising much faster than the grid has historically expanded, and individual proposed sites can request several gigawatts. (energy.gov)

7. Demand-side offsets are possible. If AI reduces hiring or shifts income toward capital owners with lower propensities to consume, the productivity benefit may not translate one-for-one into near-term aggregate demand. Executive surveys already anticipate some employment reduction, although the evidence is highly uncertain. (nber.org)

10. Overall assessment

Plausibility of 4% total growth in 2027

Possible, but outside the central forecast range. Ordinary forecast uncertainty alone is wide enough to include an outcome near 4%; a favorable combination of productivity, investment, consumer demand and other shocks could produce it. The Fed’s historical 70% forecast-error band reaches 4.2%. (federalreserve.gov)

Plausibility of 4% primarily because of AI

Low-plausibility bull case. Current evidence supports an AI growth contribution that is already material and could become larger, but the most defensible near-term range remains below the approximately 1.8-point increment required to move the consensus midpoint to 4%.

Musk’s scenario becomes plausible only if:

• AI investment remains near the upper end of project forecasts;
• domestic production captures substantially more of the hardware and infrastructure value chain;
• productivity effects broaden from individual tasks to whole firms within months rather than years;
• output gains exceed any reduction in labor hours and household demand;
• grid, generation, chip and construction constraints do not bind; and
• inflation, interest rates, trade or geopolitical shocks do not offset the AI boost.

Plausibility of sustainably maintaining 4% growth

Much lower. Investment can temporarily lift growth to unusually high rates, but sustained 4% growth would require a persistent productivity-growth increase approaching the top of—or beyond—mainstream research estimates, combined with continued labor-force or hours growth. Most institutional estimates instead imply AI will make GDP progressively larger while raising the annual growth rate by a few tenths of a percentage point, not permanently doubling it.

11. Key uncertainties to monitor through 2027

1. Quarterly contributions from software, information-processing equipment and related imports in BEA data.
2. Whether data-center construction and hyperscaler capex continue accelerating or merely remain high.
3. Domestic semiconductor, server and electrical-equipment production relative to imports.
4. Labor-productivity growth in AI-exposed industries versus less-exposed industries.
5. Firm survey evidence on measured—not perceived—revenue, output and cost gains.
6. Growth in aggregate labor hours, entry-level hiring and real household income.
7. Electricity generation, transmission interconnection and data-center completion schedules.
8. Whether the Fed accommodates AI-led supply growth or offsets demand and inflation pressures.

Sources

• Federal Reserve, September 2026 Summary of Economic Projections. (federalreserve.gov)
• Congressional Budget Office, Budget and Economic Outlook: 2026 to 2036. (cbo.gov)
• IMF, July 2026 World Economic Outlook Update and AI model research. (elibrary.imf.org)
• OECD, 2026 U.S. outlook and AI productivity research. (oecd.org)
• Federal Reserve, The AI Buildout and the Economy. (federalreserve.gov)
• Federal Reserve, Estimating Aggregate Data Center Investment with Project-level Data. (federalreserve.gov)
• Bureau of Economic Analysis, 2026:Q2 GDP release. (bea.gov)
• Bureau of Labor Statistics, 2026:Q2 productivity data. (bls.gov)
• Philadelphia Fed, Third Quarter 2026 Survey of Professional Forecasters. (philadelphiafed.org)
• NBER executive surveys on firm AI adoption and productivity. (nber.org)
• Department of Energy and Lawrence Berkeley National Laboratory, data-center electricity projections. (energy.gov)