> **来源：[研报客](https://pc.yanbaoke.cn)** # 2026 # Capital Markets # Forecast # Economic # Experimentation Trade, Labor, and Debt Under the Microscope # 2026 # Capital Markets # Forecast 5 THEME01 The Reshoring Experiment Tariffs and Smart Manufacturing 15 THEME02 The Labor Experiment Replacing Workers with AI 23 THEME 03 The Debt Experiment Investing in a Debt Cycle 33 CONCLUSION The Outlook From Lab to Portfolio Reality 01 Ta Tariffs 02 L Labor 0 Outlook 03 De Debt Ta # Tony Roth Wilmington Trust Investment Advisors, Inc. # Investing in a Period of Economic Experimentation Successful investing involves both art and science. The former is the "je ne sais quoi" that comes from an effective mixture of experience, intuition, and humility. You are lucky if you have it and, if you don't, there is little you can do to acquire it anytime soon. The science of investing, by contrast, is available to all through the careful study of events and their antecedents. Fortunately, history rhymes and through the thoughtful framing of key economic factors and study of their effects, we can gain a clearer sense of what economic trends may hold in the future. # Experimentation is normal Periodically, there is a deviation from economic trends or policy changes so significant that it can be analogized only to economic experimentation. We find ourselves in such a time today—making history less relevant and calling on a deeper understanding of macro theory and current observations to predict where the economy and markets may be headed. Critical to our inquiry is appreciating that experimentation is a feature—not a bug—of an evolving economy. Much like science itself, there cannot be progress without testing new combinations of elements and interactions among variables. History is littered with examples. These notably include withdrawing the U.S. from the gold standard in 1971; creating the euro in 1999; founding the World Trade Organization (WTO) in 1995 and granting China's membership in 2001; bailing out banks during the global financial crisis; and the Federal Reserve taking decisive action in times of crisis such as the pandemic that encompassed rock-bottom interest rates, balance-sheet expansion (i.e., quantitative easing), and the Fed's role as lender of last resort. Many economic experiments have elevated nations, moved financial markets forward, and even saved lives. But as with science, not all experiments yield anticipated—or even positive—results. Some offer no material advancement, while others can go very wrong. Often, the results are known only with the benefit of many years of careful observation. While the paradigm of scientific experiment involves controlled tests and the isolation of one experimental variable at a time, the real world is not a laboratory. Economic experiments implicate the unpredictability of human behavior and play out in a dynamic arena of infinite variables. Isolating any single factor at a time is simply not possible. We must combine the science of observation with the art of interpretation to best estimate how today's economic experiments will continue to unfold. # Themes for 2026 Our 2026 Capital Markets Forecast dives into some of the economic experiments occurring today, offers our insights and observations, and draws key connections to investment strategy. We place no normative judgment on the merits of the hypotheses being tested today, but rather strive to draw economic conclusions and identify investment opportunities—where we believe we have sufficient observational data points to do so. <table><tr><td></td><td>01 Ta Tariffs</td></tr><tr><td>02 L Labor</td><td></td></tr><tr><td></td><td>03 De Debt</td></tr><tr><td>04 O Outlook</td><td></td></tr></table> A historically concentrated U.S. equity market could mean recommitting to both a thoughtful allocation to active management and diversification across all asset classes. Our first theme takes on the experiment of tariffs at the highest rates in the post-WWII period, intended to redirect global supply chains back to the U.S. for purposes of economic prosperity and national security. We analyze the tariff policy's current and possible future impact on manufacturing activity, jobs, and budgetary revenue while highlighting some of the investment opportunities we see associated with the new trade regime. The second theme looks at the shrinking of the labor force. Demographics, immigration policy, and corporate investment in artificial intelligence form a powerful concoction with potentially significant implications for long-term economic growth and tomorrow's stock-market leaders. Third, we tackle a familiar topic: the U.S. debt trajectory. This experiment has been years in the making and shows no signs of changing. Can the U.S. borrow far above current debt levels, as is projected, without unleashing disastrous consequences? How can investors attempt to hedge against such risks? Finally, we pull these strands together with what we believe matters most to our clients: a discussion of how the experiments are informing our investment and diversification strategies to start the new year. And as it relates to portfolio construction, we recognize that the equity market's current concentration in the shares of Al-related companies is an experiment in and of itself. To reiterate, we believe that experimentation is not inherently negative, but that it takes us into unfamiliar territory, adding risk and, potentially, volatility to the investment landscape. It also can create a greater need for the art of investing, i.e., experience and inference to help anticipate which investments may prove to be successful diversifiers in the future. # The scientific method at work in portfolios We begin 2026 recognizing that the results of some economic experiments will become clear in the year ahead, but many will take more time. We expect the U.S. economy to recover from its fourth-quarter slowdown, but it could take the balance of the year before we see growth return to a $2\%$ annualized trend. Consumers likely will continue to operate in a two-speed economy wherein high-income consumers continue to spend but low-income consumers face strain. Overall, we see the U.S. expansion continuing in 2026. From our themes, we seek to identify compelling investment opportunities in areas such as industrial capital goods, technology, and private markets.<sup>1</sup> However, a long investment horizon is necessary—and caution is warranted in the short term—as certain areas of the market are trading at frothy valuations. Diversification against global fiscal risks may be partially achievable with opportunistic additions to precious metals and even cryptocurrency. The latter may not be appropriate for all investors, particularly the risk averse. Finally, a historically concentrated U.S. equity market could mean recommitting to both a thoughtful allocation to active management and diversification across all asset classes. THEME 01 # The Reshoring Experiment # Tariffs and Smart Manufacturing 01 # Pm Private markets 01 # C Cybersecurity 01 # A Automation 01 # Ai Artificial intelligence 01 # P Productivity 01 # Sm Smart manufacturing 01 # Tr Trade 01 Tariffs THEME 01 # The Reshoring Experiment # Tariffs and Smart Manufacturing The U.S. drive to reshore manufacturing activity and jobs via the use of tariffs is an experiment that, if successful, would unwind decades of movement to a globally integrated production and supply chain system. Importantly, the common narrative describes manufacturing job losses as a simple zero-sum game in which one country loses a job and another adds that job. The truth, though, is that much of historical manufacturing job losses stem from the ongoing automation of manufacturing processes that increase productivity and reduce the need for labor. That said, overcoming the high relative cost of U.S. labor is a major challenge. In addition, higher-skilled labor is necessary to operate ever-more-sophisticated production methods. As we start 2026, our scientific observation indicates that the use of tariffs as the main tool to spur manufacturing reshoring does not yet appear successful. In the near term, we continue to see more economic risks than rewards from these tariffs as firms and trading partners struggle to adjust in an environment of continuing uncertainty. Over the longer term, if heightened tariffs remain in place, we expect them to encourage some gains in domestic production, though not enough to overcome the labor-cost disparity and generate manufacturing job growth. Instead, we see increased production coming from accelerated capital-intensive methods using AI, robots, and smart manufacturing<sup>1</sup> techniques, all of which present investment opportunities. Figure1 Tariff hikes in 2025 among the largest shocks in U.S. history Largest federal tax increases as $\%$ of GDP Sources: Tax Foundation, Wilmington Trust. Data as of October 31, 2025. Figure 2 Manufacturing jobs hit by outsourcing and productivity U.S. manufacturing jobs and real output Sources: Bureau of Economic Analysis, Bureau of Labor Statistics, Wilmington Trust. Data as of November 30, 2025. # Tariff impacts in 2026 Tariffs created a drag on the U.S. and global economies in 2025 and continue to pose a threat to further expansion in 2026. If the tariffs announced by President Trump in the Rose Garden on April 2, 2025 had been fully implemented, it would have amounted to an eye-watering $30\%$ effective tariff rate—up from $2.5\%$ on January 1 and sufficient to push the economy into recession. For a variety of reasons (e.g., postponements, trade agreements, voluminous product exclusions, implementation lags, rerouting of shipments through lower-tariffed countries, noncompliance), the effective tariff rate reached just $11\%$ by the fourth quarter of 2025.² That is still a substantial tax hike of roughly $275 billion per year—equal to about $0.9\%$ of gross domestic product (GDP) and among the largest peacetime tax hikes in U.S. history (Figure 1). The tariff hikes are somewhat mitigated by tax cuts from the One Big Beautiful Bill Act (OBBBA). Corporate taxes could fall by $137 billion in 2026, with the manufacturing sector as the largest beneficiary.$ That said, firms with substantial imports and low capital investment likely will see far less benefit. Consumers have been stung by higher prices on imported goods, leading to a slowdown in spending. We expect tariffs to continue weighing on growth this year. OBBBA included personal income tax cuts that could help to soften tariffs' impact. The bulk of these cuts only make existing tax rates permanent and do not affect current income. The new cuts (e.g., higher deduction for seniors, deduction for auto loan interest, no tax on tips or overtime pay, and a higher deduction for state and local taxes) amount to roughly $30 billion per year.<sup>4</sup> New manufacturing techniques require highly skilled workers, and the U.S.'s high domestic labor costs are a disincentive for firms to reshore. # Automation and the labor cost challenge Figures 2 and 3 illustrate the challenges facing this experiment. Domestic manufacturing jobs took a hit just after China joined the WTO (Figure 2), as many labor-intensive, low-value industries (such as apparel) outsourced their production overseas. The U.S. lost nearly six million jobs over 10 years and two recessions, and has added about 1.3 million after the losses bottomed out in 2010. However, manufacturing output rose steadily over these recent decades, revealing a shift to more highly valued industries as well as higher productivity for the sector. Total output for the sector increased $17\%$ from 2000 to 2024 even with the dramatic reduction in jobs.[5] Remarkably, the turn to higher efficiency and productivity also has occurred in China, which has reported more than 15 million jobs lost in goods-producing sectors—including manufacturing—over the past decade, while also raising output.[6] These dramatic increases in productivity are not unprecedented. One hundred years ago, about $25\%$ of U.S. workers were farmers. Yet even as that number has shriveled to $1.5\%$ , food production has more than quadrupled.[7] We do not expect manufacturing productivity gains to reverse, which would make the reshoring of jobs challenging. New manufacturing techniques require highly skilled workers, and the U.S.'s high domestic labor costs are a disincentive for firms to reshore. Only Switzerland and Norway have higher total labor costs in the sector<sup>8</sup> (Figure 3). Figure 3 High U.S. labor costs a hurdle for reshoring Total manufacturing costs per employee (annual, USD) Sources: Bureau of Economic Analysis, Bureau of Labor Statistics, Wilmington Trust. Data as of November 30, 2025. Figure 4 Labor-intensive products more challenging to reshore Equating production costs for shoes and semiconductors Shoes Semiconductors Sources: Wilmington Trust and references noted in endnotes 9, 10, 11, and 13. Data as of October 31, 2025. # The reshoring experiment We do not observe or expect that tariffs alone will suffice to induce a significant shift of production, especially for more labor-intensive products. Figure 4 is a stylized example comparing production costs for a pair of sneakers made in the U.S. versus in Vietnam, and the same for a leading-edge logic chip, a critical semiconductor device. The former is an example of labor-intensive textile manufacturing that contributes to large-scale offshore production. Much of sneaker production is either fully automated or machine assisted, but labor still plays a significant role. Shoes like this may retail for $100 but cost far less to produce (around$ 30 when accounting for materials, wages, and other costs such as administrative expenses and profit margins). Tariffs are levied based on the import value of the item, and in this case would apply to the $30 cost that the U.S. firm pays at the shipping dock. The current (as of November 2025) 20% U.S. reciprocal tariff rate on Vietnam would add another$ 6, bringing the total cost to $36, still well below the estimated U.S. production cost. Labor costs are the critical differentiating factor making production in the U.S. more expensive. They have been reported at between $1 and$ 2 an hour for textile workers in Vietnam, while U.S. wages are 10 times higher $^{10}$ —bringing labor costs to $30 in this example and increasing the U.S. production cost to $63, or$ 33 higher than when outsourced (Figure 4). This implies that a 111% tariff would be required to equalize the cost of manufacturing a pair of shoes in the U.S. and Vietnam. It would take a $100\%$ tariff to make reshoring profitable on a per-unit basis, even before considering the higher fixed costs associated with construction of U.S. production facilities. At a reciprocal tariff level of $20\%$ with Vietnam, it makes more sense for importers to pay the tariffs until the cost and scale of automated solutions for the labor-intensive steps of production are economically viable. Until then, these tariffs mainly serve as a source of government tax revenue at the expense of corporate profits and U.S. consumers. Like textiles, semiconductors are a heterogeneous category that includes, for example, logic processors, memory chips, and analog components, each produced at varying scales and for distinct applications. For illustrative purposes, the example reflects the production of logic chips, which today occurs primarily in Taiwan, especially for the most advanced designs. Manufacturing advanced semiconductors is highly automated and requires more skill, with smaller wage differentials between countries. For instance, production staff at TSMC, a leading chip producer, reportedly earn about $30,000 per year in Taiwan, while job postings for related roles at TSMC's U.S. foundry in Arizona pay close to$ 50,000.[11] These numbers place U.S. wages at roughly 67% more than those in Taiwan—a much smaller difference than the 10x wage gap in the textile example. When all costs are factored in, manufacturing semiconductors in the U.S. is more expensive than in Taiwan, but the differential is less pronounced than in the case of textiles. A Goldman Sachs analysis found that operating expenses are $18\%$ higher in the U.S. compared to Taiwan, and all-in expenses are $44\%$ higher when factoring in other items, particularly the cost of capital expenditures (capex). $^{12}$ Use of tariffs could be sufficient to tilt the balance of costs in favor of domestic production, particularly if combined with capex incentives. $^{13}$ The calculus for whether such an investment is realistic hinges on more complex factors (notably the wealth of specialized knowledge that Taiwan has accumulated over decades of producing the most advanced chips, plus the extensive nexus of supply chains developed to support it). It is possible for a similar industry to be built in the U.S., but it likely would be a multi-year, multi-stage process that requires businesses to feel relative certainty about the policy framework going forward. Based on the foregoing, products with very intensive labor input, such as sneakers, probably will remain imported goods and provide tax revenue. There are more capital-intensive products that could be reshored, but we think it would take new investments in smart manufacturing techniques and other innovations to overcome the labor-cost differential. We also expect reshoring to be a gradual process over many years as technology develops and firms replace existing capital. Lastly, we must not overlook the impact that future tariff uncertainty could have on corporate decision making. As with most experiments, if the observed results are not considered favorable, the experiment may be terminated. This possibility, whether within the timeframe of the current administration or beyond, also has a strongly depressive effect on corporate America's interest in reshoring manufacturing. Figure 5 Globalization has supported margin expansion for manufacturers S&P 500 manufacturers' and all others' net profit margins 1952-August 2025 Source: Empirical Research Partners. Data as of August 30, 2025. Excludes financials, real estate investment trusts, and utilities. Based on trailing four-quarter aggregate data smoothed on a trailing six-month basis. The large-cap stock universe is used prior to 1977. Gray bars represent recessions. # Smart manufacturing, productivity, and profitability Manufacturers' net profit margins have accreted over the course of decades of globalization (Figure 5). We expect the clearing of regulatory and labor-cost obstacles for U.S.-domiciled manufacturing to depend heavily on smart manufacturing. Such technologies have taken a huge leap forward with generative and agentic AI. The latter builds on the power of generative AI's large language models (LLMs) by adding the ability to learn, adapt, and be more proactive in accomplishing goals. These are necessary skills for many enterprise applications. The potential improvements in predictive maintenance, quality control, and cooperative work between humans and collaborative robots (cobots), sometimes without any human oversight, could yield double-digit productivity gains for the industry over time.[14] KPMG research projects a $20\%$ reduction in production downtime, $15\%$ increase in product quality, and $25\%$ reduction in defect rates for manufacturers utilizing generative AI.[15] However, investors may need to be patient. Researchers have documented a productivity J-curve-i.e., when initial investments result in a dip in margins before growth reaccelerates—that especially plagues manufacturing companies.[16] # Use cases of AI and robotics in smart manufacturing<sup>17</sup> Collaborative robots (cobots) work alongside human workers to enhance productivity and safety while handling repetitive or physically demanding tasks. Internet of Things (IoT) generates data from sensors, programmable logic controllers, deep learning, $^{18}$ and AI algorithms to constantly update the digital model with live data and perform predictive maintenance. Sensors can perform quality control with new precision. Agentic AI can manage supply chains and inventory, and schedule system optimization. We expect to see the greatest benefit from smart manufacturing accrue to companies that have high capital intensity (i.e., spend more on fixed-asset investment) or high labor costs and complex, customizable products. It helps to already have production facilities in the U.S. that could be expanded to increase space and expertise. Deep balance sheets are a must as well. If companies can profitably bring manufacturing supply chains back to the U.S., consumers could benefit from better, U.S.-made quality and lower shipping costs. Collaborative robots 01 Internet of Things 01 Sensors 01 Agentic AI Percent adoption of smart manufacturing across industries Figure 6 Sophisticated, high-cost industries lead in smart manufacturing adoption Source:2025 Smart Manufacturing and Operations Survey: Navigating challenges to implementation, Deloitte Insights, May 1, 2025. The greatest adoption of smart manufacturing has been in the automotive, electronics, and pharmaceutical industries (Figure 6). National security objectives in the pharmaceutical, energy, chemicals, and rare earths sectors could provide additional incentive to manufacture domestically. # Investing in industry 5.0 There are several interesting investment opportunities associated with the smart manufacturing theme that we are implementing in client portfolios. Industrial capital goods providers present a compelling avenue to invest in automation, robotics, and smart building solutions associated with upgraded domestic manufacturing. In fact, industrial companies may be in the early stages of a multi-year super cycle. Valuations for the sector are not cheap, but industrials are trading at a narrower historical premium to the overall market (Figure 7). AI software, specifically inference applications that focus and direct the computing power of well-trained LLMs, has the potential to streamline inefficiencies in the production process. Agentic AI could usher in a real boost to manufacturing productivity. While generative AI models create content based on learned patterns, agentic AI more proactively applies model output to accomplishing goals or completing more complex tasks. For example, an LLM could create a travel itinerary, but an agentic AI model could book the trip. In the context of manufacturing, agentic AI has the ability to autonomously adjust schedules, diagnose supply chain issues and proactively reroute inventory, and predict maintenance-related issues. We believe we have only begun to scratch the surface of agentic AI applications and their use cases, and that there is a long runway for software companies producing this technology. Cybersecurity stands to benefit from smart manufacturing and AI adoption more broadly. An astounding $91\%$ of respondents to a recent Deloitte survey of manufacturers reported cybersecurity breaches in the last year—yet just $32\%$ of those respondents cited cyber risk assessment as a top priority.[19] The higher the technology intensity in manufacturing or any other business, the greater the need for cybersecurity. Valuations of many cybersecurity companies have skewed rich for years, but it is difficult to understate future growth expectations for the industry. The coming advent of quantum computing also might trigger a super cycle in cyber investment. Lastly, small companies also can benefit from shifting supply chains back home. Often, such firms lack the capital required to take advantage of the latest technology by making the investments to retrofit, enhance, or expand manufacturing facilities. In these cases, the private markets can be a solution. Investors in both private equity and debt could benefit from earlier access to these operational-turnaround stories. Figure 7 Valuations for industrials look reasonable relative to the market Price-to-earnings ratio for industrials sector vs. S&P 500 Sources: Bloomberg, Wilmington Trust Investment Advisors. Data as of October 31, 2025. Presents a premium/discount valuation (measured by P/E ratio based on 12-month forward-loc consensus earnings estimates) relative to the S&P 500. THEME 02 # The Labor Experiment # Replacing Workers with AI 02 Labor THEME 02 # The Labor Experiment # Replacing Workers with AI The move to reduce immigration is perhaps the riskiest experiment underway, as demographic trends are famously challenging to reverse and immigration policies could have lasting impacts. A simultaneous experiment is occurring in the business community, with implementation of AI replacing workers in a handful of job types where automation has been achievable. Labor demand has been hit even harder across the economy by a cyclical slowdown, highlighting the challenges of scientific observation in a noncontrolled experiment. It is early days in these dual experiments, but our observations thus far indicate greatly reduced supply and demand for labor. Technology's impact on labor demand is difficult to pin down and differs across industries and jobs—a truth seen throughout major past innovations, whether the steam engine, the automobile, the personal computer, or the internet. Where a new technology acts as a substitute for labor—as AI has thus far—it puts people out of work. This is typical during the initial adoption phases. But as the new technology is integrated into business processes and new implementations are cooked up, it becomes a complement to workers, driving labor demand higher and increasing productivity. We see the potential for a productivity boom emanating from AI, offering investment opportunities across the technology universe. # Near-term impacts of AI adoption Concerns about AI replacing workers and hollowing out the labor force have run rampant since the release of ChatGPT. In our view, AI is a transformative technology that ultimately will create more jobs than it destroys, but there is mounting evidence that its impact thus far has been negative, especially in highly exposed jobs and for those early in their careers. In particular, AI may have replaced software engineers and customer service workers aged 22-25, reducing jobs by $20\%$ and $10\%$ , respectively, since late 2022.<sup>1</sup> By contrast, there has been little impact for less-exposed occupations such as production supervisors, health aides, and stock clerks. The dynamic of new technologies dampening labor demand in the early stages of adoption translates to a downside risk for job growth and consumer spending in 2026. We maintain a cautious view on labor markets and job growth in the year ahead before considering the AI impact. On the more positive side, we expect the reduction of the labor force, particularly the anticipated lower issuance of high-skilled H-1B visas, to mitigate the labor-market effects of AI-driven job losses for exposed sectors. Figure 1 Low birth rates and aging population lead to a declining native population within a decade Population growth and contributing factors Source: Congressional Budget Office. See www.cbo.gov/publication/61390#data. Data as of September 30, 2025. Population refers to the Social Security area population, which includes all residents of the 50 U.S. states and the District of Columbia, as well as civilian residents of U.S. territories. It also includes federal civilian employees and members of the U.S. armed forces living abroad and their dependents, U.S. citizens living abroad, and noncitizens living abroad who are eligible for Social Security benefits on the basis of their earnings while in the United States. # Dwindling labor force and immigration Prior to the second Trump administration, the U.S. already faced a dramatic slowdown in population and labor-force growth. Net natural increase, which is simply births minus deaths, has been heading downward for years as the population ages and fertility rates fall (Figure 1). In this context, immigration increasingly has been a saving grace as a key source of fresh labor supply, but it is now being sharply restricted. Even work visas for highly skilled workers (especially H-1B holders) could face a sudden decline due to new fees. The latest projections indicate that the U.S. population could dip into decline starting in 2031. $^{2}$ The Congressional Budget Office (CBO) assumes that net immigration will average roughly 900,000 over the next five years, which may not materialize if current policies remain in place—meaning that population growth may turn out even lower than expected. In fact, our calculations show that immigrants have almost entirely accounted for net job growth since 2022. That is not to say immigrants are taking jobs from U.S. nationals. Instead, it highlights the degree of slowdown and decline of the labor force's native-born population.<sup>3</sup> The unemployment rate for the native-born population was just $4.3\%^{4}$ as of September 2025, so there is very little spare capacity for nationals to fill jobs that immigrants otherwise would have taken. A vital concern for the U.S. going forward is the ability to fill high-skilled positions that have long been filled not only by citizens, but also by immigrants through the H-1B visa program (Figure 2). Since 1990, the program has enabled U.S. employers to hire foreign workers with specialized skills on a temporary basis. But the Trump administration instituted a $100,000 fee for new applicants as of September 2025, which could reduce applications by as much as $30\%$ .[5] Strong productivity growth is not only achievable, but also is the most likely outcome. The magnitude of realized productivity growth will significantly depend on the success of AI. We see it as likely to define a new era in growth. Drawing highly talented immigrants has long been a signature characteristic of the U.S. economy, one that has boosted U.S. innovation and productivity. Skilled migrants make up just $5\%$ of the workforce but earn $10\%$ of total labor income, reflecting the higher-productivity positions and industry concentration. Overall, immigrants comprise about $10\%$ of the U.S. population but make up $23\%$ of patent holders. More than $25\%$ of U.S.-based Nobel Prize winners have been immigrants, $^6$ as are four CEOs of the so-called Magnificent Seven, the darlings of U.S. business and innovation. # Productivity boost needed With the labor force slowing or perhaps even in decline, the U.S. economy is in dire need of stronger productivity growth. The simple math of long-term economic growth is that it equals the sum of the growth rates of the labor force and productivity. If immigration pulls back sharply enough that labor-force expansion is flat, it would take consistent $2\%$ annualized productivity growth to achieve the current consensus estimate of $2\%$ long-term economic growth. To put that in perspective, annual productivity growth was roughly $1\%$ in the expansion between the global financial crisis and the pandemic. It was about $3.5\%$ in the late-1990s and early-2000s boom.[7] We believe that strong productivity growth is not only achievable, but also is the most likely outcome. A review of projections of annualized labor productivity growth from AI over the next 10 years yields a range of just $+0.1\%$ to $+3.3\%$ . That compares with the estimated $+1\%$ to $+1.5\%$ attributed to the internet boom of the 1990s. We expect a similar result from AI, boosting annual productivity by $+1\%$ to $+1.5\%$ over the next decade. The magnitude of realized productivity growth will significantly depend on the success of AI. We see the new technology as likely to define a new era in growth, much like the adoption of electricity, the steam engine, the automobile, the personal computer, and the internet. But it also will be disruptive (as it already has been) for many industries and occupations. And we could be wrong, as the technology might fail to achieve its promise. Figure 2 High-skilled immigrants a key source of labor Sources: U.S. Citizenship and Immigration Service, Wilmington Trust. Data as of October 31, 2025. Figure 3 Growing share of firms look to adopt AI Firms that have used AI in the last two weeks and plan to use in next six months Source: "Business Trends and Outlook Survey," U.S. Census Bureau. Data as of September 30, 2025. # Adoption of AI is growing While estimates vary, most signal that a high proportion of consumers have at least tried using AI.<sup>8</sup> On the business side, research focused on large firms found that more than $80\%$ had explored or piloted an AI solution, and $40\%$ had deployed some kind of tool<sup>9</sup> (Figure 3). A Census Bureau survey attempting to represent all firms showed much lower economy-wide usage, but rapid growth. Just $8\%$ of firms had tried using AI as of October 2025, double the $4\%$ from two years ago. Additionally, $14\%$ indicated they plan to use AI at some point in the next six months.<sup>10</sup> The data supports our view that AI is likely to grow quickly, given the doubling of usage in such a short timeframe. But it also points to an experiment that still is in its very early stages, with a long road ahead before it is proven. We observe meaningful productivity gains at the microeconomic level, but do not expect to see economy-wide gains with a measurable macroeconomic impact until further down the road. # Early innings of the AI investment cycle Productivity growth is dependent on AI living up to—or exceeding—the recent hype. The echoes of 1999 are ringing loudly today, and the debate over an AI bubble is even louder. For investors, the most pressing question today is where we are in the AI investment cycle. Unfortunately, this simple question has a complicated answer. We see the AI investment opportunity as multifaceted, with the different layers of the AI ecosystem presenting different degrees of value to investors today. Ultimately, the opportunity is a function of long-term demand for the technology, which is unknowable but hotly debated. We believe that we are still in the early innings of the AI investment cycle while recognizing the risks associated with over-investment and investor exuberance. # The great bubble debate Myriad signs abound of over-exuberance associated with AI investment. Private AI firms like Anthropic and OpenAI command valuations of approximately $350-$ 500 billion $^{11}$ with no profits yet to show. We have seen a flurry of deal announcements amounting to chipmakers and hyperscalers taking stakes in AI platform providers, who then use the financing to increase orders of chips and cloud storage from those same companies. These headlines echo the telecom vendor financing deals of the dot-com era. Hyperscaler<sup>12</sup> capex has more than doubled since 2022,<sup>13</sup> when ChatGPT was first released. It is projected to grow another $50\%$ by 2027 to $520 billion<sup>14</sup> (Figure 4), raising risks of over-investment that could be impossible to pay off. But the opportunities, in our view, justify the risks. Al is not the internet. Adoption of LLMs has been immediate, and applications are barely scratching the surface. The productivity benefit could be larger and faster than that realized in the internet's early days. Al-related capex is still modest compared to the overall economy's $6 trillion fixed investment.[15] It also is funded largely with cash, though debt financing is increasing for some companies. Hyperscalers in the public market continue to generate tremendous free cash flow. The physical infrastructure buildout needed for our technology economy is still in its infancy and has not yet resulted in a debt bubble. For all of the promise of this transformative technology, the stock market's robust performance compels us to approach AI investments with cautious optimism. The faster and higher valuations climb, the more attention and intention are required to maintain appropriate diversification (we discuss this in a later section). For now, we think we are still in the early innings of this super cycle, broadly speaking. Higher productivity could create a virtuous cycle for enterprise demand, but monetization is key. There still is value for investors in various parts of the Al ecosystem, in our view. Figure 4 Mega-cap tech capex has accelerated U.S. capital expenditures ($ billions) for Alphabet, Amazon (AWS)*, Meta, Microsoft, and Oracle Sources: Bloomberg, Wilmington Trust Investment Advisors. Data as of October 31, 2025. * Amazon capex discounted for specific Amazon Web Services spending. # The AI ecosystem and portfolios Infrastructure. The infrastructure layer of the AI ecosystem has been on the frontlines of investor enthusiasm, with capital flowing into semiconductors, hyperscalers, data centers, energy, and utilities. Semiconductors may be among the more extended parts of the AI trade, at least in the near term. Demand is outstripping supply, and chip sales are on track to reach $1 trillion by 2030.[16] Future demand trends, however, are very difficult to predict and technological advancement can unexpectedly render inventory obsolete. The combined market cap of the top 10 global chip companies has risen by a whopping $392\%$ between mid-November 2022 and October 2025. $^{17}$ These large companies also face disruption risk, as AI chip startups have secured a total of $7.6 billion in global venture capital funding during the second through fourth quarters of 2024 alone $^{18}$ —a relatively small but growing figure. Hyperscalers currently look appealing, as cloud infrastructure growth continues to accelerate while valuations are below prior peaks. The proportion of enterprise workloads in the public cloud has increased more than $60\%$ since 2018 while their penetration remains just over $50\%$ .[19] The overall market opportunity for cloud infrastructure could increase tenfold by 2040.[20] Investing across the AI ecosystem Source: Wilmington Trust. Data centers, energy, and utilities are all related to the servicing of cloud and AI computing demand. Data centers have a development pipeline of 50 million square feet in the U.S. alone,[21] and leasing demand from hyperscalers is reaccelerating. Current vacancy rates of $2.3\%$ represent a record low.[22] At the same time, valuations are becoming stretched, over-investment risk is high, and energy is a major constraint. Many hyperscalers are building out their own data center capacity, leaving the investment opportunity more limited within real estate investment trusts and private real estate. On the energy side, the U.S. Department of Energy estimates that data centers consumed $4.4\%$ of U.S. electricity generation in 2023, a figure set to increase to $7\% -12\%$ by 2028.[23] Public and private liquid cooling solutions look compelling, as servers for AI model training expend 20x more heat per server than standard computer cloud servers.[24] Intelligence. AI model providers exist in the private and public markets, and valuations of private companies appear stretched. In the public market, hyperscalers like Alphabet, Meta, and Microsoft also are investing in model development and offer more reasonable valuations with diversified revenue streams. Still, these companies have become quite large, and aggressive growth projections could be harder to meet as they get larger. Applications. Software applications will augment the productivity benefits of inference models, particularly through agentic AI. As described earlier, agentic AI is more proactive, autonomous, and adaptable than LLMs. It presents the added value of not only answering a question or generating content, but also of performing a task to completion. With estimates for active users in the low-single-digit billions (Figure 5), monetization (specifically through enterprise pricing) will be critical to margin expansion. We see the investment opportunity within software applications as in its early innings—in fact, almost too early. While some large software companies are implementing agentic AI, disruption risk is very high and it is generally too early to assess which companies will emerge as winners. Adopters. Perhaps the most nascent-yet-exciting investment opportunities related to AI are the adopters of AI, that is, the businesses deploying the technology to realize new levels of efficiency. As we mentioned earlier, a J-curve trajectory is common for realization of productivity benefits associated with AI adoption, as early investments require retooling of work processes to fully realize the efficiency of the new technology. Figure 5 Growing user base and rapid adoption for AI applications Monthly active users (billions) and months to 100 million users for popular apps Sources: PricewaterhouseCoopers (company website and statements), Wilmington Trust Investment Advisors. Data as of August 2025. * OpenAI does not publish monthly active users (MAU) for ChatGPT, but it does report weekly active users (WAU) (of 700 million). The figure above assumes that WAU represents $80\%$ of MAU. **Myspace no longer tracks monthly active users. At its peak, the website had 115 million monthly active users and that figure is estimated to be less than 1 million today. Encouragingly, evidence does show that small businesses thus far have been the earlier and more agile adopters of AI,[25] suggesting greater productivity gains for the businesses that arguably need it most. Ultimately, we think the AI investment opportunity still has long legs. As with all new technologies, investors should anticipate volatility and pullbacks along the way. For example, the unveiling of the DeepSeek LLM almost a year ago triggered a $15\%$ selloff for the tech sector,[26] which has more than recovered. We expect such pockets of volatility to persist as AI technology matures and new winners or threats emerge. In fact, the prior decade's winners rarely dominate the next decade, and we cannot rule out pain ahead for some of the largest and best-performing stocks of this cycle. As such, when economies realize transformative technological experimentation, fundamental analysis, active management, and diversification all assume greater import. THEME 03 # The Debt Experiment # Investing in a Debt Cycle 03 Debt Figure 1 U.S. deficits projected to remain large Deficits (% share of GDP) and IMF projections # The Debt Experiment # Investing in a Debt Cycle The third major experiment should surprise no one: a global sovereign debt burden fueled by ever-growing deficits with no end in sight. The U.S. is projected to continue running unprecedented budgetary shortfalls in an economic expansion, and many other large economies are following suit. This creates the risk of an unsustainable debt trajectory, and observation thus far shows investors demanding higher interest rates to take on higher risk. That said, we see two possible mitigants: stronger-than-anticipated productivity growth (especially in the U.S.) and the burgeoning stablecoin industry. The latter may prove an important new source of demand for U.S. Treasuries that could help keep a lid on rates. Higher rates in the U.S. also are, for the moment, encouraging a carry trade where international investors buy Treasuries to take advantage of their higher yields, increasing demand. Overall, a worsening debt picture affects the path of interest rates, the U.S. dollar, precious metals, and cryptocurrency. # A widening U.S. budget deficit Government budgets are expected to deteriorate for most countries over the next several years. The U.S. finished fiscal year 2025 with a \(1.8 trillion budget gap, roughly equal to \(6\%\) of GDP (Figure 1). That is down from about \(6.3\%\) of GDP in fiscal year 2024, but is expected to expand to \(7\%-8\%\) over the next five years.\(^{1}\) United States France Germany Italy Japan United Kingdom Sources: International Monetary Fund, Wilmington Trust. Data as of December 31, 2024. We think current fiscal outlooks for most countries have been priced into markets, but could be susceptible to adjustments in 2026. In the U.S., a change in tariff policy could reduce expected tariff revenue and drive deficits—and, therefore, interest rates—higher. Continuing the trend from the past 20 years, the U.S. is projected to realize wider deficits than nearly all countries we consider in our investment process. The sum of projected U.S. deficits from 2026 to 2030 totals $39.2\%$ of GDP, greater than all countries in developed and emerging market equity indices except for China. # Rising interest rates The projected path for deficits and debt, all else equal, is likely to drive interest rates higher. There is considerable academic debate about whether annual deficits or total accumulated debt is more important. Based on total accumulated debt, estimates across a range of studies find that each $1\%$ increase in the debt-to-GDP ratio pushes long-term interest rates higher by two-to-three basis points (bps). The International Monetary Fund projects that the U.S. ratio will rise by $18\%$ from 2025 to 2030, which would correspond with long-term rates moving up by 36-54 bps. Research similarly finds that annual deficits drive interest rates higher, but the effects can be larger and more volatile. This makes sense, as the path of deficits can shift fairly quickly with a turn in the economy or a change in policy. Accumulated debt moves more slowly. We think current fiscal outlooks for most countries have been priced into markets, but could be susceptible to adjustments in 2026. In the U.S., a change in tariff policy (either from a Supreme Court decision or trade deals) could reduce expected tariff revenue and drive deficits—and, therefore, rates—higher. A slower economy also would raise deficits and rates, while a stronger economy would help bring both down. # Productivity to the rescue? The prospect of stronger productivity growth stemming from AI discussed in Theme 1 could have a profound impact on the U.S. debt outlook. The CBO currently projects debt to inexorably rise from $100\%$ of GDP in 2025 to $109\%$ of GDP in 2030 and $127\%$ in $2040^4$ (Figure 2). Figure 2 Debt projections vary widely with productivity growth Source: Congressional Budget Office. Data as of May 31, 2025. Figure 3 More than $70\%$ of existing mortgages have an interest rate below $5 \%$ Share of outstanding existing mortgage loans by mortgage rate Sources: "17% of Homeowners With Mortgages Have an Interest Rate of at Least 6%, the Highest Share in Nearly a Decade," Redfin, February 6, 2025; Federal Housing Finance Agency, National Mortgage Database. Data as of September 30, 2024. But current productivity assumptions underlying the extended baseline forecast average only $+0.9\%$ per year from 2026 to 2030 and $1.1\%$ thereafter. If productivity growth is just $+0.5\%$ stronger, the faster-growth scenario results in a downward bend of the debt-to-GDP curve (the "higher productivity" scenario in Figure 2), all things equal. To stabilize debt, all that is needed is for nominal GDP growth to exceed the average interest rate on the debt.[5] That is achievable, and already the case, with the average interest on outstanding debt at $3.4\%$ as of October 2025.[6] The CBO's baseline projection has nominal GDP growth falling from about $4.5\%$ in 2025 to $3.5\%$ over the forecast, but if AI is able to achieve the $1\%-1.5\%$ productivity boost that we expect and the primary deficit—the deficit before including interest payments—is small, then the government could run surpluses and pay down debt in coming years. Should this come to pass, it will essentially mirror the late 1990s, when the personal computing and internet boom boosted productivity. In 1996, the CBO projected debt-to-GDP to rise from $50\%$ to $52\%$ over the next five years. But annual productivity surged to $3.5\% -4\%$ , driving strong economic growth and leading to budget surpluses. Debt-to-GDP fell to $31.5\%$ by 2001.[7] # Rates and the housing market One of the most important implications of the deficit and interest rates is their effect on the housing market. A housing market cycle starts with low interest rates, encourages purchases of new and existing homes, and helps consumers build wealth. It also leads to second-order purchases that support areas of the retail market (think home improvement) and the services economy. Mortgage rates, which correlate with the movement of the 10-year Treasury note yield, declined in 2025 but are unlikely to fall below $4\% - 5\%$ in the near term, absent a recession. (We cover more of our outlook for rates further along in this theme.) While lower interest rates and increased housing supply could improve affordability at the margin, more than $70\%$ of existing mortgages have interest rates below $5\%$ (Figure 3). Figure 4 Stablecoin issuers a growing source of demand for U.S. Treasuries Top 10 stablecoins by market capitalization (USD billions) <table><tr><td>Stablecoin</td><td>Company</td><td>Country of incorporation</td><td>Market capitalization</td></tr><tr><td>USDT (Tether)</td><td>Tether Limited Inc.</td><td>El Salvador</td><td>$167.0</td></tr><tr><td>USDC</td><td>Circle Internet Financial LLC</td><td>United States</td><td>$67.6</td></tr