NextFin News - Elon Musk says artificial intelligence will probably lift the global economy by 20% to 30%, or roughly $20 trillion to $30 trillion a year - a forecast that would add, in a single year, an amount roughly equal to the current annual output of China, or up to the entire U.S. economy. The Tesla, SpaceX and xAI chief made the estimate Monday at the G20 Innovation Ministerial in Chapel Hill, North Carolina, where he appeared virtually alongside former White House AI adviser David Sacks as finance and technology officials from the Group of 20 gathered to debate how to govern the technology even as they race to capture its output.
Markets took the comments in stride. S&P 500 futures hovered near the flat line and Nasdaq-100 futures dipped 0.1% in early trading on Monday, as the forecast landed on an investor base that has already spent years pricing AI-driven growth into semiconductor and megacap platform stocks.
The scale of the claim is what separates it from the usual boosterism. Global gross domestic product is on track for about $105 trillion this year, according to International Monetary Fund figures. A 20% to 30% increase means adding $21 trillion to $31.5 trillion to annual world output - roughly two-thirds to one full U.S. economy, or about one to one-and-a-half times China's. Yet the IMF's own World Economic Outlook, updated in July, puts global growth at 3.0% for 2026 and 3.4% for 2027, below the prepandemic average. Musk's number is not a forecast for next year; it is a claim about the terminal size of the AI economy once the technology matures. The tension between those two timelines - the slow growth of today and the exponential promise of tomorrow - is where the real debate begins.
The Arithmetic Behind the 20% to 30% Claim
Musk framed the estimate as a rough scale calculation rather than a point forecast, telling the G20 gathering:
I think, just to give you some sense of scale here, AI will probably increase the global economy by 20 to 30%. That is my rough estimate. Meaning, on the order of $20 to $30 trillion per year.
The math is internally consistent with the current baseline: 20% of a $105 trillion world economy is $21 trillion, and 30% is $31.5 trillion.
But the claim rests on an implicit assumption about what AI becomes. In Musk's telling, AI is not merely a productivity tool layered on top of existing work. It is a new factor of production - a form of labor that can be replicated at near-zero marginal cost. He told the ministerial that AI software will become "stockfish-level good," meaning it will be "impossible for a human to compete in writing software with AI," and that the technology will become extremely capable at all forms of engineering and digital work within 12 to 18 months. Stockfish, the open-source chess engine, calculates millions of positions per second and has been unbeatable by humans for two decades; the analogy is that coding, design and analysis will follow the same path.
This is not Musk's first time drawing a line from AI to aggregate growth. At the World Economic Forum in Davos in January, he argued that "if you have ubiquitous AI that is essentially free or close to it and ubiquitous robotics, you will have an explosion in the global economy that is truly beyond all precedent." He has also predicted double-digit U.S. GDP growth within 12 to 18 months, tying economic output directly to "applied intelligence." The G20 estimate is the global version of that thesis: if intelligence is the binding constraint on growth, and AI removes that constraint, then growth itself gets re-priced.
The Mechanism: How AI Becomes a Factor of Production
The transmission channel matters more than the headline percentage. Historically, general-purpose technologies - electrification, the internal combustion engine, the personal computer - raised productivity through a slow diffusion process. Firms had to reorganize workflows, retrain workers and rebuild capital stock before the gains showed up in the data. That is why measured productivity lagged the invention of the transistor by decades.
AI differs in one critical respect: it is a general-purpose technology that can itself accelerate the adoption of general-purpose technologies. An AI system can write the software, design the circuit, optimize the logistics network and draft the regulatory filing that would otherwise each require specialized human labor and months of calendar time. The bottleneck shifts from skilled labor to compute, energy and deployment permission. That is why the capital expenditure numbers now matter as much as the productivity forecasts.
Musk's own companies illustrate the scale of the bet. SpaceX reported second-quarter AI revenue of $2.56 billion, up 247% from a year earlier and 213% sequentially, driven by new cloud service agreements and growth in Grok and X subscriptions. That single segment now accounts for roughly a third of the company's $7.81 billion in quarterly revenue. But the spending side is starker: AI capital expenditure in the quarter was $15.83 billion, which the company's report puts at about 86% of total capital spending, while the AI segment posted an operating loss of $1.26 billion. The company ended the quarter with $100 billion in cash and has announced a pending $60 billion acquisition of AI coding platform Cursor.
There is a logic to spending ahead of revenue. Musk has said SpaceX is targeting 10 gigawatts of AI computing capacity by the end of next year, valued at roughly $30 to $50 per watt - which, by his math, implies $300 billion to $500 billion in annual AI revenue by 2028. That is a five-year revenue trajectory that would put AI alone ahead of almost every national economy. The bet is that demand for intelligence will absorb whatever supply gets built, and that the marginal cost of an inference will keep falling faster than the price.
Cyclical or Structural: The Call That Decides the Trade
This is the judgment that separates a trading view from an investment thesis. The AI capex cycle itself is cyclical - it will revert. History offers three comparable episodes. The fiber-optic buildout of the late 1990s left enough dark cable to carry the internet for a decade after the bubble burst. The shale boom of the 2010s overshot demand and produced a decade of subpar returns for energy producers even as oil output kept rising. The cloud buildout of 2020-2021 saw hyperscalers over-provision, then spend the next two years digesting capacity. In each case, the infrastructure survived the investors.
But the underlying driver here is structural. Electrification did not stop being transformative because railroad stocks crashed in 1929. If AI genuinely reduces the marginal cost of cognitive labor toward zero - and the evidence from coding benchmarks, image generation and scientific models suggests it is moving in that direction - then the productivity regime has changed regardless of who owns the chips. The distinction is clean: the capex cycle will mean-revert; the technology will not. An analysis that blends the two - either the bubble-bear who concludes AI is worthless because valuations compress, or the booster who concludes valuations can never compress because AI is real - gets the conclusion backwards.
The market leg of this story is already well advanced. Semiconductor names and megacap platform companies have rallied for years on the same premise Musk is now stating at a G20 ministerial. When a thesis reaches the level of a heads-of-states forum, the first-order trade is crowded. That does not make the thesis wrong; it means the excess returns, if they come, will come from the second-order effects that the crowd has not priced.
The Second-Order Question the Market Is Not Asking
The first-order effect is the one everyone states: AI raises productivity, profits rise, growth accelerates. The second-order effect runs through the fiscal and monetary channels, and it cuts against the simple version.
If AI does deliver a genuine productivity boom, the near-term consequence is not unambiguously bullish for risk assets. A productivity-driven expansion allows central banks to hold policy rates higher for longer without triggering a recession, because potential growth itself has risen. The discount rate applied to those future earnings may not fall as much as the bull case assumes. Meanwhile, governments facing structurally higher growth will see tax revenues rise - and the Congressional Budget Office has already begun counting on that: in February it added an AI productivity adjustment to its baseline projections for the first time. Those assumed gains are now being used to offset other policy drags on growth. If the productivity arrives slower than modeled, the fiscal math breaks; if it arrives on time, the deficit narrative changes but the rate path stays restrictive.
There is also a distributional channel that feeds back into demand. A technology that substitutes for cognitive labor faster than it creates new labor categories compresses the wage share before it lifts the level of employment. In the transition, aggregate demand can soften even as measured productivity rises - the same paradox the U.S. saw in 2025, when GDP growth ran strong while monthly job growth averaged barely above zero. That gap between output and employment is exactly what the Yale Budget Lab flagged when it warned against declaring a productivity boom before the data is clearer.
The third-order effect is the expectation gap itself. Musk's 20% to 30% number sets a bar that no quarterly earnings cycle can clear. The companies building the infrastructure must now deliver terawatt-scale deployment, double-digit revenue growth and expanding margins simultaneously - while the macro data will only confirm the boom years after the stocks have already moved. The mismatch between the speed of equity pricing and the speed of GDP measurement is where the volatility will live.
The Strongest Case Against the Thesis
The bear argument is not that AI is useless. It is that the productivity data has not shown up yet, and that history says it may take decades to arrive. A 2026 survey of economists found that while respondents assign a 61% probability to moderate or rapid AI progress by 2030, their unconditional GDP forecasts barely move above recent trends: a median of 3.3% annualized growth by 2030 and 3.5% by 2050 even under the rapid-progress scenario. Their productivity forecasts are even more restrained - a 2% annualized increase in labor productivity by 2030 in the unconditional case, 3.2% under rapid progress, against a 2025 baseline of 1.94%. For context, the postwar U.S. boom averaged about 4% a year.
The mechanism the skeptics point to is diffusion lag. The Penn Wharton Budget Model estimated that generative AI will raise productivity and GDP by about 1.5% by 2035 and nearly 3% by 2055 - real gains, but an order of magnitude below Musk's 20% to 30% and spread over decades rather than a single cycle. Ray Dalio, founder of Bridgewater Associates, has compared the rapid gains in AI-related stocks to previous technology bubbles, arguing that the market is pricing a golden age before the economic data confirms one.
The strongest version of this critique attacks the core of Musk's mechanism: it assumes AI complements labor and capital fast enough to show up in aggregate output. If instead AI substitutes for labor while the institutional friction of deployment - regulation, liability, enterprise sales cycles, energy permitting - slows adoption, then the GDP effect arrives on the skeptic's timeline, not the booster's. In that world, the $15.8 billion in quarterly AI capex is not a down payment on a $30 trillion economy; it is the latest entry in the ledger of infrastructure bubbles where the builders lost money even though society eventually benefited.
The falsifying signal is specific and observable. If measured U.S. labor productivity growth does not sustain at least 2.5% annualized over the next four quarters - well above the 1.94% baseline and the 2% unconditional forecast - then the structural-productivity thesis is wrong and the boom remains a capex cycle. A second signal: if the AI segment revenue growth of the major builders decelerates below 50% year-over-year while capex remains above 70% of total spending, the demand-absorption assumption has broken. Either print would shift the base case from structural transformation to cyclical overshoot.
What Comes Next: Beneficiaries, the Exposed, and the Scenarios
Translating the mechanism into impact produces an asymmetric map. The beneficiaries are the owners of the scarce complements to cheap intelligence: compute capacity, power generation and transmission, advanced packaging, and the platforms that control distribution and data. Energy is the cleanest read - an AI economy measured in tens of trillions of dollars of output runs on gigawatts, and permitting cycles make new supply slow regardless of how fast chip fabs can build. The exposed are the owners of purely cognitive labor with no proprietary data or deployment moat, and the equity holders of infrastructure builders who paid peak-cycle multiples assuming uninterrupted demand growth.
The forward look splits by horizon. In the short term - the next two to four quarters - the trade is dominated by the capex cycle and the earnings bar. Every data-center announcement and every inference-cost benchmark will move names more than the GDP data will. In the medium term - two to five years - the question is whether revenue growth at the infrastructure layer can keep pace with capacity additions, or whether utilization rates roll over the way they did in fiber and shale. In the long term - a decade and beyond - the structural question resolves: either AI becomes a general-purpose technology that lifts the productivity trend, or it joins the list of transformative inventions whose aggregate economic payoff arrived a generation later than the hype.
Three scenarios frame the path. The base case: AI delivers meaningful productivity gains but on the economists' timeline rather than Musk's - global growth lifts toward 3.5% to 4% by the early 2030s, capex normalizes, and the winners are the scarce-complement owners rather than every AI proxy. The upside case: deployment friction falls faster than expected, inference costs drop an order of magnitude, and measured productivity breaks above 3% - in that world Musk's 20% to 30% becomes a floor rather than a ceiling. The downside case: capex stays elevated while utilization rolls, margins compress, and the fiscal assumptions built on AI productivity fail to materialize - a 2000-style digestion period in which the technology survives but the valuations do not.
The G20 setting is itself a signal. When the world's largest economies convene their finance officials to write principles for a technology - the so-called Carolina Principles, under which Washington is urging member nations to refrain from creating new AI regulatory bodies or rules - they are acknowledging that the technology has moved from the laboratory to the balance sheet of the global economy. Musk's number will be wrong in its timing, as his double-digit-growth call almost certainly will be. But the direction of the claim - that AI is a regime shift in the cost of intelligence, and therefore in the ceiling on output - is the hypothesis the next decade of data will test.
The market is not being asked to decide whether AI is real. It is being asked to decide whether the boom is priced for a structural shift or a cyclical wave - and those two prices do not converge.
Explore more exclusive insights at nextfin.ai.
