Assessing Elon Musk’s 4% U.S. Real GDP Growth Claim for 2027

Review period: 2023-01-01 to 2026-09-19 Scope: Assessment of the plausibility that AI could lift U.S. real GDP growth to about 4% in 2027, covering forecasts, investment, productivity, adoption, and scenario requirements. Source policy: Primary government, central-bank, international-institution, and academic sources were prioritized; private forecasts were used for comparison and clearly identified.

Executive summary

Elon Musk wrote on September 18, 2026: “My guess is that AI roughly doubles US GDP growth next year from ~2% to ~4%. Maybe even more.” This report interprets the claim as referring to real, not nominal, GDP growth. (247wallst.com)

Bottom line: 4% real growth in 2027 is possible as an upside-tail outcome, but it is not a plausible central forecast—and attributing most of it to AI would require an unusually rapid, front-loaded productivity breakthrough on top of an already enormous investment boom.

• The principal public and professional forecasts cluster around 1.8%–2.4% growth in 2027. Their rough midpoint is about 2.1%–2.2%, leaving a 1.8–1.9 percentage-point gap to Musk’s 4% scenario. Even Vanguard’s unusually bullish 3% forecast falls one percentage point short. (federalreserve.gov)
• AI is already contributing materially through software, computing equipment, data centers and associated power infrastructure. The IMF estimates AI-related technology investment added about 0.5 percentage point to U.S. growth in 2025. A broader Federal Reserve proxy produced quarterly contributions ranging from 0.26 to 1.40 points during 2025 and 1.36 points in 2026:Q1, but these are volatile, include non-AI activity and are not measures of sustainable productivity growth. (imf.org)
• A Federal Reserve project-level model puts the import-adjusted 2027 contribution from data-center investment at –0.08, +0.20 or +0.53 percentage point, depending on future project announcements. Thus, even its optimistic case covers less than one-third of the gap between consensus and 4%. (federalreserve.gov)
• The evidence for task-level productivity benefits is real but heterogeneous: studies find gains of 14% in customer service, 40% faster completion of selected writing tasks and over 25% faster performance on suitable consulting tasks, while a randomized study found experienced open-source developers were 19% slower with early-2025 AI tools. Translating such results into economy-wide productivity is the central uncertainty. (nber.org)
• Ten-year macro estimates range from roughly 0.1 percentage point of extra annual productivity growth in Acemoglu’s conservative framework to 0.4–1.3 points in OECD scenarios for highly exposed economies. These estimates generally describe an average over a decade, not an immediate 2027 jump. (nber.org)

For 4% growth to be driven primarily by AI, the United States would probably need, simultaneously: a net domestic AI-investment contribution near the top of current estimates; an abrupt AI-driven increase of roughly 1–1.5 percentage points in economy-wide labor productivity growth during 2027; continued growth in labor hours; limited import leakage and energy bottlenecks; and no offsetting contraction from interest rates, inflation, tariffs or weaker household demand. That combination is economically coherent, but substantially more aggressive than the timing embedded in mainstream research.

1. What exactly is the claim?

Four distinctions are essential.

GDP level versus GDP growth rate

If AI raises the level of real GDP by 5% over ten years, the economy is permanently 5% larger than otherwise, but the average addition to annual growth is only about 0.5 percentage point during the transition. Once adoption matures, growth normally returns toward its prior trend unless AI continually generates new innovations.

By contrast, raising the annual growth rate by two percentage points—from 2% to 4%—is much more demanding. If sustained for ten years, it would leave GDP approximately 22% above a 2% growth path.

Investment-driven growth versus productivity-driven growth

Building a data center, installing servers or developing capitalized software directly raises current investment and GDP. But investment contributes to the growth rate only while real spending continues to increase. A permanently high but flat level of spending supports the level of GDP without repeatedly adding to growth.

Productivity-led growth is different: AI allows more real output from the same labor and capital. This is the mechanism that can sustainably raise potential growth, real incomes and productive capacity.

Real versus nominal GDP

The claim is meaningful only if interpreted as real GDP growth. Nominal GDP includes inflation, and a 4% nominal growth rate would not represent a doubling of the roughly 2% real-growth baseline used in current forecasts.

Annual-average versus fourth-quarter growth

The Federal Reserve and CBO commonly report fourth-quarter-to-fourth-quarter growth, while the IMF, OECD and Philadelphia Fed Survey of Professional Forecasters often emphasize annual-average growth. These measures are not perfectly interchangeable. A strong late-2026 starting point could raise 2027 annual-average growth through carryover without requiring 4% growth throughout 2027.

2. Current 2027 forecasts

Sources: Federal Reserve, CBO, OECD, IMF, Philadelphia Fed, BofA and Vanguard. (federalreserve.gov)

The five main public/institutional forecasts—Fed, CBO, OECD, IMF and SPF—have a median of 2.2% and an average of about 2.1%. Musk’s scenario is therefore approximately 1.8–1.9 percentage points above the mainstream center.

The comparison also shows that mainstream forecasts are not simply ignoring AI. CBO explicitly incorporates stronger total-factor-productivity growth partly from generative-AI adoption, while the OECD forecast recognizes strong AI-related investment. Yet both still project only 1.8% real growth in 2027 because slower labor-force growth, fading fiscal support, tariffs and softer consumption offset some of the technology boost. (cbo.gov)

Vanguard provides the strongest credible support for Musk’s direction of travel. It forecasts 3% growth and argues that the AI investment cycle will continue through 2027. But Vanguard also says the shift from investment to broad productivity enhancement will unfold over years rather than quarters. (corporate.vanguard.com)

3. How much is AI already contributing?

Hyperscaler capital expenditure

Aggregate capital expenditure by Amazon, Google, Meta, Microsoft and Oracle reached $412 billion in 2025, equivalent to about 1.31% of U.S. GDP. This demonstrates the exceptional scale of the current buildout, although the total includes some non-AI logistics, cloud and general infrastructure spending. (federalreserve.gov)

The investment includes several GDP categories:

• data-center structures;
• servers, networking equipment and semiconductors;
• electrical-generation, transmission and cooling infrastructure;
• purchased and internally developed software;
• research and development; and
• complementary communications equipment.

Official statistics do not yet contain a clean “AI investment” line. Federal Reserve and BEA researchers therefore construct proxies from broader categories, which necessarily capture some non-AI activity. (federalreserve.gov)

Estimated contribution to recent growth

The IMF estimates that AI-related technology investment added approximately 0.5 percentage point to U.S. GDP growth in 2025, before counting any generalized productivity spillovers from companies using AI. (imf.org)

A broad Federal Reserve presentation—including software, data centers, high-tech equipment and power investment—estimated quarterly annualized growth contributions of:

• 1.40 points in 2025:Q1;
• 0.85 in Q2;
• 0.26 in Q3;
• 0.84 in Q4; and
• 1.36 in 2026:Q1. (federalreserve.gov)

These figures show that AI-associated capital formation can move quarterly GDP materially. They should not, however, be read as causal estimates of AI’s net contribution: some included investment would have occurred without generative AI, while imported servers and chips subtract from U.S. GDP through net exports. The Federal Reserve specifically warns that failing to account for imported equipment can significantly overstate the domestic contribution. (federalreserve.gov)

Data centers: gross versus domestic contribution

A Federal Reserve working paper using individual data-center projects estimated the following 2027 contributions:

The import adjustment assumes only about 44% of data-center investment is domestically produced, reflecting the high imported content of advanced computing equipment. (federalreserve.gov)

The negative low-case contribution does not mean data centers disappear. It means the level of investment remains substantial but declines from an exceptionally high prior-year pace. This is the clearest example of why a capital boom can lift GDP temporarily without permanently raising its growth rate.

Semiconductors and software

Software and computing equipment have been the largest positive components in the Federal Reserve’s AI-investment proxy. Domestic semiconductor production adds to U.S. value added, but imported processors, memory, networking hardware and manufacturing equipment reduce the net GDP contribution. BEA also counts purchased and internally developed software as investment, so some AI model development and implementation enters GDP before its eventual productivity payoff is known. (federalreserve.gov)

4. What does the productivity data show?

Nonfarm business labor productivity was 2.2% higher in 2026:Q2 than a year earlier. Productivity grew at a 2.1% annual rate from the end of 2019 through 2026:Q2, compared with 1.5% during the preceding business cycle. This is consistent with an improving productivity environment but does not establish that AI caused the improvement. (bls.gov)

AI adoption is widespread under some measures but still shallow under others. At the end of 2025, approximately 18% of businesses told the Census Bureau they used AI in at least one business function, while 41% of workers reported work-related generative-AI use. An Atlanta Fed survey found that 78% of workers were employed by firms reporting some AI adoption, illustrating how adoption rates depend heavily on whether the unit measured is a firm, worker or employer-weighted workforce. (federalreserve.gov)

Firm-level surveys remain cautious about realized effects. A multinational survey of executives found that roughly 69% of firms used AI, but about 89% reported no productivity effect so far; U.S. firms generally expected modest rather than transformative near-term gains. (richmondfed.org)

Microeconomic evidence supporting large potential gains

• A study of 5,179 customer-service agents found AI assistance raised issues resolved per hour by 14%, with gains of approximately 34% for novice and lower-skilled workers. (nber.org)
• An experiment involving professional writing tasks found a 40% reduction in completion time and an 18% improvement in quality. (doi.org)
• A field experiment with consultants found workers completed suitable tasks more than 25% faster, with quality gains above 40%. (aiinstitute.hbs.edu)

Evidence challenging easy extrapolation

• Benefits are highly task- and worker-specific; experienced workers often gain less than novices. (nber.org)
• In a randomized trial involving experienced developers working in familiar repositories, access to early-2025 AI tools increased completion times by 19%. (metr.org)
• A 10% improvement on one task does not imply a 10% increase in a firm’s output. Validation, coordination, customer demand, organizational redesign, error correction and non-AI bottlenecks can absorb much of the initial saving. Federal Reserve researchers identify this task-to-firm aggregation problem as a key explanation for why striking experiments have not yet produced an obvious macroeconomic acceleration. (federalreserve.gov)

5. Published estimates over one, five and ten years

Next year: through 2027

The clearest quantified short-run estimate is the Federal Reserve data-center model: a central import-adjusted contribution of 0.20 percentage point, with a range from –0.08 to +0.53 point. Broader software, hardware and power investment could make the total AI-capital contribution larger, but some of that support is already incorporated into the 1.8%–2.4% baseline forecasts. (federalreserve.gov)

Goldman Sachs has long forecast that AI’s measurable productivity effect would begin around 2027, but its published analysis does not imply a two-point GDP-growth jump in that first year. It estimates a medium-term U.S. GDP-growth boost of about 0.4 percentage point by 2034, despite a much larger 1.5-point potential productivity effect under widespread adoption. (goldmansachs.com)

The Penn Wharton Budget Model expects the annual productivity contribution to peak at only 0.2 percentage point in 2032, suggesting that adoption and organizational adjustment, rather than technical capability alone, limit the near-term macro effect. (budgetmodel.wharton.upenn.edu)

Assessment for 2027: Current research supports a meaningful AI contribution, especially through investment, but offers little direct support for a sustainable AI-driven addition approaching two percentage points in a single year.

Approximately five years

Five-year estimates generally imply a meaningful increase in the level of output rather than a discontinuous rise in annual growth. An IMF model estimates that world GDP would be about 0.8% to 2.4% higher after five years, depending on productivity assumptions; its U.S. effects are larger than the global average because the United States combines high occupational exposure with relatively strong readiness. (imf.org)

PWBM’s path rises toward a 1.5% U.S. GDP-level gain by 2035, with the largest annual growth contribution occurring in the early 2030s. (budgetmodel.wharton.upenn.edu)

Thus the five-year literature is compatible with an AI growth dividend of a few tenths of a percentage point annually, and possibly more in aggressive scenarios, but not necessarily with a permanent doubling of trend growth.

Approximately ten years

Sources: Acemoglu, PWBM, IMF, Goldman Sachs and OECD. (nber.org)

The range is wide because the studies make different assumptions about:

• how many tasks become technically feasible to automate or augment;
• how quickly firms adopt and reorganize around AI;
• whether saved labor is redeployed productively or displaced;
• whether AI accelerates research and innovation itself;
• capital, energy and semiconductor availability;
• errors, safety constraints and regulatory costs; and
• competitive pass-through of savings into prices and demand.

6. What would have to happen for 4% growth in 2027?

Starting from a mainstream forecast near 2.1%–2.2%, approximately 1.8–1.9 additional percentage points are needed.

Illustrative growth-accounting requirements

This table is an illustrative decomposition, not a published forecast. Its investment assumption is above the Federal Reserve data-center central case and requires sizable contributions from software, power, semiconductor fabrication and other domestic infrastructure as well. Its productivity assumption requires the high end of decade-average OECD estimates to arrive almost immediately rather than gradually. (federalreserve.gov)

More specifically, the following conditions would be needed:

1. Investment must keep accelerating. Maintaining 2026 spending levels is insufficient; real data-center, server, power and software investment must continue rising sharply through 2027.
2. A larger share must be domestically produced. If most GPUs, networking equipment or fabrication machinery is imported, gross investment figures exaggerate the addition to U.S. GDP. (federalreserve.gov)
3. Productivity gains must spread beyond technology firms. Professional services, finance, health care, retail, manufacturing and government would need rapid workflow redesign rather than superficial chatbot adoption.
4. Labor savings must become additional output. Firms must redeploy saved hours into higher-value work or lower prices enough to expand demand. Layoffs without offsetting demand or investment would weaken the growth effect.
5. Power and construction bottlenecks must not bind. Electricity generation, transmission, grid connections, cooling systems, skilled construction labor and semiconductor supply must expand quickly enough to support deployment.
6. The macro environment must cooperate. Higher energy prices, restrictive interest rates, tariffs or weaker real household income could offset much of the AI impulse. The OECD and Conference Board identify several of these as downside risks for 2027. (oecd.org)
7. The gains must appear in measured real output. Free services, quality improvements and consumer surplus can produce genuine welfare gains without adding proportionately to measured GDP.

With labor hours recently growing only about 0.2% year over year, 4% output growth sustained without a major increase in hours would require labor-productivity growth near 3.5%–4%. That would be a substantial acceleration from the latest 2.2% year-over-year reading, though not historically unprecedented. (bls.gov)

7. Evidence supporting Musk’s view

1. The capital boom is already macroeconomically large. Hyperscaler capital expenditure exceeds 1% of GDP, and official estimates find AI-associated investment added around half a percentage point to 2025 growth. (federalreserve.gov)
2. The project pipeline could produce further upside. Under a doubling of new data-center plans, the Federal Reserve model estimates a 0.53-point domestic contribution in 2027 from data centers alone. (federalreserve.gov)
3. Recent productivity performance is stronger than the pre-pandemic cycle. Nonfarm business productivity has grown at 2.1% annually in the current cycle, compared with 1.5% in the previous one. (bls.gov)
4. Micro studies show genuinely large gains in well-matched tasks. The 14%–40% improvements observed in customer service, writing and consulting demonstrate substantial technical potential. (nber.org)
5. Credible high-end macro scenarios exist. The OECD estimates up to 1.3 points of additional annual labor-productivity growth in rapid-adoption scenarios, while Vanguard already forecasts 3% real growth in 2027. (oecd.org)
6. AI could accelerate invention itself. Federal Reserve research finds suggestive evidence that generative AI may be both a general-purpose technology and an “invention in the method of invention,” potentially extending gains beyond a one-time automation wave. (frbsf.org)

8. Evidence challenging the claim

1. No major mainstream forecast is close to 4%. Even the highest individual Federal Reserve participant projection for 2027 is 2.9%, and Vanguard’s bullish forecast is 3%. (federalreserve.gov)
2. Forecasts already include AI. Adding projected AI investment mechanically to current forecasts would double-count a major component of their business-investment and productivity assumptions. (cbo.gov)
3. Investment growth is vulnerable to base effects. The central Federal Reserve data-center scenario contributes only 0.20 point in 2027 because the growth rate of spending matters, not just its elevated level. (federalreserve.gov)
4. Import leakage is large. The same Federal Reserve work reduces estimated data-center contributions by more than half after accounting for imported high-tech equipment. (federalreserve.gov)
5. Realized productivity gains remain limited at many firms. Executive surveys find broad experimentation but mostly small or zero effects on output per worker to date. (richmondfed.org)
6. Micro results do not aggregate mechanically. Some tasks speed up dramatically, others do not, and errors or workflow disruptions can make experienced workers slower. (metr.org)
7. The conservative macro literature implies a much slower transition. Acemoglu and PWBM estimate roughly 1%–1.5% cumulative GDP gains over a decade, not a two-point annual-growth surge in 2027. (nber.org)
8. Labor-force growth is a structural constraint. Aging and reduced immigration limit potential output growth, requiring an even larger productivity acceleration to reach 4%. (cbo.gov)

9. Overall assessment

Most likely interpretation

AI is already preventing U.S. growth from being weaker by sustaining business investment, and it may be beginning to support productivity. A reasonable reading of the evidence is that AI could add several tenths of a percentage point to annual growth in the near term and more over the medium term as adoption broadens.

Plausibility of 4% in 2027

• 4% as an observed one-year outcome: Possible. Forecast errors, a strong late-2026 starting point, fiscal demand, financial conditions and an exceptionally strong investment cycle could combine with AI to produce it.
• 4% primarily because of AI investment: Difficult. Even aggressive import-adjusted data-center estimates cover only a fraction of the roughly 1.8–1.9-point forecast gap.
• 4% primarily because of sustainable AI productivity: Currently unsupported as a central 2027 forecast. It requires decade-scale high-end productivity effects to arrive almost immediately.
• 4% sustained for multiple years: Much less plausible without evidence that AI is continually accelerating innovation and potential output rather than merely raising the GDP level during an adoption transition.

Verdict: Musk’s forecast should be treated as an aggressive upside scenario rather than a forecast supported by the current consensus. The investment component is already large enough to matter and could surprise further. The decisive—and currently unproven—assumption is that broad, measurable productivity gains emerge in 2027 quickly enough to add roughly another percentage point or more to growth. Without that productivity step-change, current evidence points closer to 2%–3% than 4%.

Key indicators to watch

The claim would become materially more credible if, by early or mid-2027, the data show:

• nonfarm business productivity accelerating above 3% year over year;
• productivity gains broadening from information and finance into retail, manufacturing, health care and other large sectors;
• continued growth—not merely a high level—in real software, computing-equipment and data-center investment;
• rising domestic semiconductor and electrical-equipment production relative to imports;
• firms reporting measurable output-per-worker gains rather than only experimentation;
• falling unit labor costs without weaker aggregate demand; and
• upward revisions toward 3% or more in the Fed, CBO, IMF, OECD and SPF forecasts.

Sources

• Federal Reserve, September 2026 Summary of Economic Projections. (federalreserve.gov)
• Congressional Budget Office, The Budget and Economic Outlook: 2026 to 2036. (cbo.gov)
• OECD, 2026 U.S. Economic Outlook. (oecd.org)
• IMF, July 2026 World Economic Outlook Update. (elibrary.imf.org)
• Philadelphia Fed, Third Quarter 2026 Survey of Professional Forecasters. (philadelphiafed.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 Labor Statistics, second-quarter 2026 Productivity and Costs. (bls.gov)
• OECD, Macroeconomic Productivity Gains from Artificial Intelligence in G7 Economies. (oecd.org)
• Acemoglu, The Simple Macroeconomics of AI. (nber.org)
• Penn Wharton Budget Model, The Projected Impact of Generative AI on Future Productivity Growth. (budgetmodel.wharton.upenn.edu)
• IMF, The Global Impact of AI: Mind the Gap. (elibrary.imf.org)
• Goldman Sachs Research, AI May Start to Boost US GDP in 2027. (goldmansachs.com)
• Brynjolfsson, Li and Raymond, Generative AI at Work. (nber.org)
• Noy and Zhang, Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence. (doi.org)