AI May Lift Growth and the Neutral Rate at the Same Time

AI, Long-Run Growth, and the Neutral Rate

The chart’s first lesson is humility. Over more than a century and a half, U.S. real GDP per capita has moved through civil conflict, world wars, depressions, oil shocks, inflation regimes, financial crises, pandemics, demographic transitions, and repeated technological revolutions. Yet the long-run line has been remarkably stable. Real output per person has tended to revert toward a trend growth rate near 1.9% a year. That number is not magic, and it is not a policy target. It is a historical summary of an economy that repeatedly absorbs shocks, reallocates capital, adopts new technology, and returns to a compounding path.

That backdrop is essential for the artificial-intelligence debate. The seductive version of the AI story says that higher productivity gives the central bank room to run easier policy: if firms can produce more with the same labor and capital, inflation pressure should fall, growth should improve, and interest rates should be able to move lower. There is a cyclical version of that argument that can be true for a while. A favorable supply shock can lower unit costs and allow stronger real growth without immediate inflation. But the long-run version is more complicated. If AI permanently raises expected growth, raises expected returns on capital, and creates a large new investment cycle in chips, data centers, power, software, automation, robotics, and complementary infrastructure, then the economy’s neutral real interest rate may rise rather than fall.

That is the central issue. AI can be bullish for growth and earnings without being bullish for structurally lower rates. A productivity revolution can raise real incomes and improve corporate margins, yet also increase the demand for capital and lift the equilibrium real return required to balance saving and investment. In that world, AI is not a simple disinflationary gift to bond investors. It is a real-economy shock that changes the entire intertemporal price system.

The distinction matters because markets often compress the AI debate into a single question: does AI reduce inflation? A better question is: does AI shift the economy’s sustainable growth path and the equilibrium price of capital? If the answer is yes, the implications are deeper. The economy may be able to grow faster, companies may generate higher profits, and households may enjoy higher real incomes. But the long-run interest-rate environment may settle above the pre-AI baseline, not below it.

The Chart’s Deeper Message

A 1.9% real GDP-per-capita trend over a very long horizon is a statement about institutional and technological resilience. It says that the United States has repeatedly found ways to convert innovation, capital deepening, education, migration, entrepreneurship, and organizational change into higher living standards. The economy did not grow smoothly; it grew through discontinuities. The railroad, electrification, the internal combustion engine, mass production, antibiotics, computing, the internet, shale energy, cloud software, and mobile platforms all changed parts of the production function. But most innovations did not permanently lift the measured growth trend by several percentage points. They were powerful, but they were absorbed into a broad historical average.

That is why the chart should make investors cautious about heroic AI extrapolation. A technology can transform firms, industries, and daily life without permanently doubling economy-wide growth. The Solow productivity paradox was not a denial that computers mattered; it was a warning that general-purpose technologies require complementary investment, organizational change, diffusion, and measurement before they show up in aggregate productivity. AI may follow the same pattern. The technology can be extraordinary while the macro effect arrives slowly.

At the same time, the chart should also make investors cautious about dismissing AI. Long-run trend stability is not proof that the trend can never change. It is proof that a true trend shift is rare and important. If AI eventually becomes a general-purpose technology comparable to electrification or computing, it could affect not only software productivity but also research, drug discovery, logistics, manufacturing, energy management, education, coding, design, finance, customer service, and managerial decision-making. The aggregate effect would depend on diffusion, but the potential scope is unusually broad.

The right reading is therefore neither hype nor cynicism. The historical trend is an anchor. AI is a possible shock to that anchor. Most shocks revert. A few shocks rewrite the path. Investors should treat AI as a probability distribution around long-run growth, not as a slogan.

Productivity Is Not the Same as Disinflation

A productivity shock lowers the cost of producing a given unit of output. If a firm can produce more with the same inputs, unit labor cost can fall, margins can rise, prices can be lower than they otherwise would have been, or some combination of the three can occur. This is the intuitive reason people associate AI with lower inflation. If AI allows a law firm, software company, bank, hospital, manufacturer, or logistics network to complete more work per employee, then the supply side improves.

But lower unit cost is not the same as lower equilibrium interest rates. Inflation is the rate of change in prices. The neutral real interest rate is the real return consistent with full employment and stable inflation when policy is neither stimulating nor restraining the economy. Productivity can affect both, but through different channels.

In the short run, productivity improvement can be disinflationary if supply rises faster than demand. In the long run, however, stronger productivity can raise expected income growth, increase desired investment, and raise the marginal product of capital. If households expect higher future income, they may save less today because their future resources look stronger. If firms expect higher returns from AI-related capital, they may demand more funds for investment. Lower desired saving and higher desired investment both point in the same direction: a higher real interest rate is needed to clear the capital market.

A simple way to frame the issue is:

`r* rises when desired investment increases relative to desired saving.`

AI can move both sides. It can increase investment demand through data centers, semiconductors, power infrastructure, enterprise software, security, automation equipment, and reorganization of production. It can reduce saving incentives if households and firms expect higher future income. That combination is not a recipe for structurally lower real rates. It is a recipe for a higher equilibrium price of capital.

This is why the AI-rate debate must separate cyclical inflation relief from structural neutral-rate pressure. A positive supply shock can give the central bank room to ease in a downturn. But if that supply shock also raises the economy’s long-run growth and investment opportunity set, the long-run neutral rate may move up.

The Saving-Investment Channel

The neutral real rate is not set by central-bank preference alone. The central bank sets the policy rate, but the neutral rate is shaped by deeper forces: productivity, demographics, fiscal policy, risk appetite, global capital flows, inequality, desired saving, desired investment, and the supply of safe assets. AI matters because it may alter the productivity and investment components at the same time.

Start with investment. A broad AI transition is capital hungry. It requires advanced chips, semiconductor fabrication capacity, cooling systems, cloud infrastructure, data centers, high-voltage power connections, grid expansion, backup generation, model-training clusters, inference capacity, enterprise integration, cybersecurity, proprietary datasets, and human capital. If AI diffuses into robotics, industrial automation, healthcare, finance, defense, and science, the complementary capital stock becomes even larger. The economy must build the physical and intangible infrastructure needed to make AI useful.

This is different from a purely software-only productivity shock. The current AI stack has a heavy physical footprint. Data centers consume land, electricity, water, chips, networking equipment, transformers, and construction labor. Semiconductor supply chains require long-cycle investment. Power grids require permitting and capital. The more AI becomes mission-critical, the more companies will invest in redundancy, security, and proprietary capacity. That raises investment demand.

Now consider saving. If AI raises expected future income, households may feel less need to save out of current income. This is a standard permanent-income logic. A household that expects faster wage growth or stronger asset returns can rationally consume more today. Firms may also retain less cash if expected returns on investment are high and external financing is available. Governments may feel more comfortable running deficits if expected growth improves debt sustainability, though that can itself raise capital demand.

The equilibrium effect depends on magnitudes, but the direction is not obviously lower. A world of higher investment demand and lower desired saving requires a higher real rate to balance the system. That is the core reason a powerful AI shock can be associated with higher `r*`.

This point also helps explain why the bond market may not react to AI the way equity investors expect. Equity investors see higher productivity, higher margins, and larger addressable markets. Bond investors must ask whether the same forces raise real rates, steepen the term structure, increase electricity and infrastructure demand, or keep fiscal and private capital spending elevated. The technology can be good news for equities and ambiguous news for duration.

Historical Analogies: Technology Can Raise Capital Demand

Technological revolutions often require enormous complementary investment. Railroads required steel, land, finance, labor, stations, bridges, and legal rights of way. Electrification required generation, transmission, motors, factories, and rewiring of production. Automobiles required roads, oil distribution, assembly lines, suburbs, logistics networks, and consumer finance. The internet required fiber, servers, data centers, devices, software, and eventually cloud infrastructure. These technologies did not simply make production cheaper; they created new investment frontiers.

AI looks similar in that respect. The model may be software, but the system is not only software. The system is a capital network. It is chips, energy, cooling, data, compute, talent, security, and workflow redesign. If the technology is truly general purpose, the capital cycle will not be limited to a few hyperscalers. It will spread across utilities, industrial equipment, real estate, semiconductor tooling, enterprise software, consulting, cybersecurity, education, healthcare, defense, and manufacturing.

That historical pattern matters for rates. A technology that merely reduces costs without requiring new capital might be strongly disinflationary. A technology that reduces costs while opening a vast new investment frontier has a more mixed rate implication. The supply curve improves, but the demand for capital also rises. The real interest rate is the price that equilibrates those forces.

There is also a timing issue. Capital demand can arrive before productivity gains are fully realized. Data centers can be built before every enterprise has transformed its workflows. Chips can be ordered before applications earn their cost of capital. Power infrastructure can be financed before utilization is optimized. During that phase, AI can be capital-intensive before it is fully productivity-enhancing. That sequence can put upward pressure on real investment demand even if the later productivity payoff is real.

This is one reason investors should be careful with simple analogies to software deflation. AI may reduce the cost of cognition, but it may do so through a very expensive industrial base. The macro effect is not just cheaper tasks. It is the construction of a new production layer.

Why Higher Growth Can Mean Higher Rates

The easiest way to see the rate logic is to think about expected growth. In many macro models, a higher expected growth rate raises the real interest rate because households prefer smoother consumption over time. If future consumption is expected to be much higher, households are more willing to consume today and less willing to save at low returns. To persuade them to save enough to fund investment, the real rate must rise.

In a production economy, faster productivity growth can also raise the marginal product of capital. If each unit of capital becomes more productive because it is paired with AI tools, firms are willing to pay more to finance capital. A manufacturer that can combine robotics, predictive maintenance, AI design, and automated quality control may find new equipment more valuable than before. A pharmaceutical company that can accelerate discovery may find research infrastructure more valuable. A bank that can automate compliance, underwriting, customer service, and risk analytics may find software and data investment more valuable.

This is the positive version of the story: higher rates are not always bad. A higher `r*` caused by stronger productivity is very different from higher rates caused by inflation credibility problems or fiscal stress. The first reflects better investment opportunities. The second reflects macro risk. If AI raises the neutral rate because the economy can productively absorb more capital, then higher real rates would coexist with stronger growth and better earnings.

But markets can still struggle with the transition. Equity valuations depend on both cash flows and discount rates. If AI raises expected profits but also raises real discount rates, the net valuation effect depends on duration, competitive advantage, margin capture, and reinvestment needs. Companies with genuine AI-driven productivity gains may justify higher earnings, but companies priced only on distant optionality may be hurt by a higher discount rate.

This is why the AI trade should not be treated as uniformly bullish for every long-duration asset. The technology can improve fundamentals while raising the hurdle rate. That creates dispersion. Firms that convert AI into near-term cash-flow growth can win. Firms that require endless capital and promise distant payoffs can struggle if real rates stay high.

The Federal Reserve’s Problem

For the central bank, AI creates a measurement problem before it creates a policy conclusion. Productivity is hard to measure in real time. Potential output is unobservable. The neutral rate is unobservable. AI adoption is uneven. Some firms will become more efficient quickly; others will spend heavily without immediate payoff. Some gains will appear as lower prices; others will appear as higher margins, better quality, faster service, or new products that statistics capture imperfectly.

If policymakers overestimate the productivity shock, they may ease too much and allow demand to exceed supply. If they underestimate it, they may keep policy too tight and suppress a genuine supply-side improvement. The right response is therefore not a mechanical rule that AI means lower rates or higher rates. It is a Bayesian process: update estimates of potential growth, unit labor costs, investment demand, labor-market slack, inflation expectations, and financial conditions as the evidence arrives.

The near-term policy implication could still be dovish in a specific environment. If inflation is falling, labor markets are softening, and AI is helping firms control costs, then the central bank can cut rates without reigniting inflation. But that is a cyclical statement. The structural statement is different. If AI lifts trend productivity and investment demand, the long-run policy rate consistent with neutrality could be higher than the pre-AI secular-stagnation baseline.

This distinction also affects how investors should interpret central-bank language. A policymaker saying that AI improves productivity is not necessarily saying that rates should be permanently lower. The same premise can support the opposite conclusion depending on the horizon. Over the next few quarters, better productivity can reduce inflation pressure. Over the next decade, better productivity can raise growth, capital demand, and `r*`.

The central bank will also need to distinguish between good and bad reasons for higher rates. If long-run real yields rise because productivity expectations improve, that is not the same as yields rising because inflation expectations become unanchored. Policy should not automatically fight every rise in real yields. It should ask whether financial conditions are tightening relative to a higher productive capacity or merely reflecting a healthier economy.

Corporate Earnings and Market Structure

AI can support corporate earnings through several channels. It can reduce labor intensity in routine cognitive tasks. It can improve pricing, personalization, fraud detection, customer service, logistics, inventory management, coding, marketing, and research. It can speed product development. It can lower error rates. It can allow smaller teams to scale faster. If these gains are real, margins can expand and return on invested capital can rise.

But the distribution of benefits will be uneven. Some firms will capture AI rents because they own data, distribution, compute, customer relationships, specialized workflows, or regulatory trust. Others will see AI commoditize services they once sold at high margins. In competitive markets, productivity gains often pass through to customers. In concentrated markets, firms may keep more of the surplus. The macro productivity effect can be broad while the profit effect is concentrated.

The capital-spending side is also uneven. Hyperscalers, semiconductor firms, utilities, equipment suppliers, and data-center developers may see enormous demand. But customers adopting AI may face rising software bills, integration costs, security costs, and organizational disruption before they see savings. The market will need to separate genuine productivity adoption from expensive experimentation.

A higher neutral-rate world would sharpen that separation. Cheap capital allows speculative AI projects to survive longer. Higher real rates demand proof. Firms will need to show that AI spending produces measurable revenue growth, cost savings, or strategic control. The phrase “AI investment” will not be enough. Investors will ask whether the investment earns above the new hurdle rate.

This is a healthy discipline. If AI truly raises productivity, it should eventually raise cash flows enough to justify capital. If it does not, higher rates will expose weak projects. The result is likely to be a more selective market rather than a universal AI melt-up.

What Would Confirm the Higher-r* AI Thesis

The first confirming signal would be sustained productivity improvement beyond a few AI-adjacent sectors. Investors should watch nonfarm business productivity, unit labor costs, output per hour in services, software-development productivity, research efficiency, healthcare administration, finance operations, and manufacturing quality metrics. A true macro shock must diffuse beyond the companies selling the tools.

The second signal would be persistent AI-related capital demand. Data-center construction, semiconductor capex, power-grid investment, electrical equipment backlogs, cloud infrastructure spending, and enterprise software implementation should remain strong even after the first hype cycle. If AI capex collapses quickly, the higher-r* argument weakens.

The third signal would be stronger expected income and investment behavior. If households, firms, and markets revise long-run growth expectations higher, saving and investment behavior should change. Consumer confidence about future income, business capex plans, venture formation, labor reallocation, and equity-market earnings revisions would all matter.

The fourth signal would be real yields staying firm even as inflation expectations remain anchored. That combination would suggest the market is pricing stronger real growth or higher capital demand rather than inflation fear. A rise in real yields with stable breakevens is more consistent with the productivity-r* channel than a rise in nominal yields led by inflation compensation.

The fifth signal would be broadened earnings growth. If AI only enriches a narrow group of infrastructure suppliers while the rest of the economy does not improve productivity, the macro thesis is weaker. If AI raises margins and output across sectors, the thesis becomes stronger.

What Would Falsify the Thesis

The higher-r* AI thesis would be weakened if productivity gains remain narrow, if AI spending proves redundant, or if firms cannot convert AI tools into measurable output. A large amount of capex can be wasted. History is full of overbuilt infrastructure cycles that eventually produced value but first destroyed capital. If AI investment runs ahead of useful adoption, the short-run effect could be lower returns rather than higher neutral rates.

The thesis would also weaken if AI primarily substitutes for labor in a way that depresses household income without creating enough new demand. In that scenario, desired saving might rise because households feel insecure, consumption could weaken, and investment could become more cautious after the initial buildout. That would look less like a productivity boom and more like a distributional shock.

Another falsification would be a strong global saving response. If the rest of the world supplies abundant capital to finance U.S. AI investment, domestic rates may not rise as much. Global capital flows matter. The U.S. neutral rate is connected to global saving, reserve demand, fiscal positions, and cross-border investment preferences.

Finally, the thesis would be weakened if regulatory, energy, data, or organizational constraints prevent broad diffusion. AI models may improve quickly, but economic productivity requires implementation. If electricity bottlenecks, legal constraints, data quality, cybersecurity, worker resistance, or management failures slow adoption, the economy may experience high capex without a sustained trend-growth shift.

The correct approach is probabilistic. AI has the potential to raise long-run growth and `r*`, but potential is not realization. The evidence must be tracked.

The Labor Market Transmission

The labor-market channel is especially important because productivity shocks do not affect all workers at once. AI may raise output per worker in some occupations while displacing tasks in others. If the dominant effect is augmentation, wages can rise, profit margins can improve, and demand can expand. If the dominant effect is displacement without sufficient new job creation, household income expectations can weaken and the saving-investment logic changes.

This is why investors should watch not only aggregate productivity, but also labor reallocation. A healthy AI boom should show rising output, new firm formation, new tasks, higher real wages in complementary roles, and manageable transition costs. An unhealthy version would show narrow profit capture, weak labor income, and social pressure that eventually invites regulation or redistribution. The neutral-rate implication is strongest when AI raises both productivity and the expected return to complementary human and physical capital.

Portfolio Implications

The first implication is that investors should separate AI earnings winners from duration winners. AI can raise expected cash flows while also raising real discount rates. That is favorable for companies with near-term earnings upgrades and durable competitive advantages, but less favorable for assets whose value depends mostly on distant optionality.

The second implication is that AI infrastructure should be analyzed as a capital cycle. Semiconductors, data centers, utilities, power equipment, cooling, networking, and construction are not just technology themes. They are macro investment channels. If the AI buildout persists, it can support industrial demand, commodity demand, grid investment, and regional real estate markets. It can also create bottlenecks and cost inflation in specific inputs.

The third implication is that bonds require nuance. AI-driven productivity can be disinflationary at the unit-cost level, but higher trend growth and stronger investment demand can keep real yields elevated. Long-duration bonds may benefit from lower inflation risk, but they may not benefit if `r*` rises. The key is whether nominal yields fall because inflation compensation declines more than real rates rise.

The fourth implication is that equity market breadth matters. A narrow AI boom concentrated in a handful of infrastructure suppliers is not the same as a broad productivity boom. The higher-r* thesis becomes more credible if earnings revisions improve across sectors using AI, not just sectors selling AI.

The fifth implication is that investors should watch the interaction between AI and fiscal policy. If policymakers interpret higher expected growth as permission for larger deficits, public borrowing can amplify upward pressure on real rates. Productivity can improve debt sustainability, but it can also encourage fiscal complacency. The neutral-rate effect depends on private and public capital demand together.

Conclusion: AI Can Be Pro-Growth Without Being Low-Rate

The long-run U.S. growth chart reminds us that the economy is resilient, adaptive, and historically anchored around a stable real GDP-per-capita trend. That history should discipline the AI debate. Most shocks do not permanently change the trend. But the rare shocks that do are enormously important.

AI may be one of those rare shocks. If it lifts productivity broadly, it can support stronger growth, higher real incomes, better margins, and improved corporate earnings. It may also reduce some inflation pressure by lowering unit costs. But none of that automatically implies a structurally lower interest-rate world.

If AI raises expected income growth, reduces the incentive to save, increases the return on capital, and triggers a large investment cycle in infrastructure and complementary assets, it can raise the neutral real rate. In that scenario, the long-run equilibrium level of interest rates would be higher, not lower, even as the economy becomes more productive.

That is the nuance the debate needs. AI can be good for growth and complicated for bonds. It can improve supply while increasing capital demand. It can make companies more profitable while raising the hurdle rate for speculative assets. It can give policymakers cyclical room to ease in some periods while lifting the structural neutral rate over time.

The chart’s message is not that the future must look like the past. It is that changing the long-run path is difficult. If AI truly does it, the result will not simply be cheaper money. It will be a different growth regime, with stronger productivity, larger investment needs, higher real income, and potentially a higher equilibrium real interest rate.