Goldman Sachs is telling clients that the most dramatic phase of the artificial intelligence trade may finally be shifting. The bank argues that the market has spent years rewarding companies building the most advanced computing infrastructure, and now it expects that capital spending growth to cool by 2027 for the first time since the AI buildout began. That forecast matters because it suggests the enormous capital expenditure cycle could be approaching a peak. If infrastructure spending slows, the enormous profits that many semiconductor and data center suppliers captured may no longer grow at the same frantic pace. Goldman instead points toward the next winners: businesses that adopt AI to improve operations, sell new products, or charge higher prices for better services. The investment community often celebrates the tool makers before it understands how the tool changes the underlying business. Goldman’s latest outlook asks investors to look one step further and examine the customers who put AI to work. That is a subtle but powerful reallocation story, one that could reshape portfolios in the years ahead. The move is not merely thematic; it reflects a deeper belief about where durable value will originate. Investors should not dismiss the shift. Timing will matter. Patience will reward them.
Goldman Sachs expects artificial intelligence related capital expenditure growth to begin slowing in 2027, a milestone that would mark the first such slowdown since the modern AI infrastructure buildout kicked into high gear. The bank shared this outlook in a note to clients, and it carries practical weight for investors who have crowded into graphics processing units, custom chips, data centers, and cloud capacity. A slowdown in spending does not mean the technology is failing. Rather, it suggests that the easy part of the investment cycle may be ending. Companies that need enormous computing power to train models are likely nearing the limits of what they will fund in the near term, or they are finding that efficiency gains reduce the need for ever more capacity. Goldman recommends that investors start moving away from some of the pure AI infrastructure names and begin focusing on adopters. The reasoning is simple. When builders stop buying so much compute, the profits from the buildout may fade, while the profits from using AI to improve margins, automate service, or create new offerings may become more important. Chief US equity strategist Ben Snider framed the shift as a continuation of the AI trade, moving from infrastructure beneficiaries toward companies that benefit from monetization and adoption, including the hyperscalers. This is not a call to abandon technology altogether. It is a call to distinguish between the tool and the user, between capacity and capability, between speculation and execution.
The next phase of the artificial intelligence trade is arriving, and it is more nuanced than the first. During the initial frenzy, investors placed big bets on companies whose products depended on scarce computing resources. Now Goldman expects the industry to move toward a phase where adoption and monetization determine winners. The bank points to hyperscalers as beneficiaries of this transition, but the broader message is that many ordinary firms could find AI useful if they have the right data, distribution, or physical presence. A software company with clean data may see quick productivity gains, while a manufacturer with messy operational records may struggle. That is why Goldman favors firms with physically intensive business models, meaning companies that interact with customers through tangible assets such as gas stations, vehicles, snacks, or real estate. These businesses are not immune to automation, but their disruption risk may look different. It is harder for artificial intelligence to replace a candy bar or a ski resort than it is to replace software or an insurance underwriter. The physical world imposes frictions, logistics, taste preferences, safety rules, and local knowledge that can slow purely digital substitution. For investors, this creates a potential bridge between the high growth of AI and the defensive characteristics of tangible businesses. The question is whether the market has already recognized these adopters or whether it remains too focused on infrastructure.
Goldman’s argument about physical assets is rooted in a simple but often overlooked truth: not every industry can be disintermediated by software at the same speed. Financial services, legal research, and administrative tasks have obvious digital substitutes. But industries built around physical delivery, food production, transportation, energy infrastructure, and retail often have operational realities that are harder to replicate online. A gas station cannot be fully replaced by a mobile app without rethinking supply chains, fuel logistics, convenience store inventory, and local customer behavior. A snack company cannot simply become a digital brand without factories, distributors, and retail relationships. A ski resort cannot become a virtual experience without losing its actual terrain, lodging, and weather risk. These differences matter because AI disruption risk should not be measured only by whether a task can be described in code. It should also be measured by whether the task depends on physical presence, proprietary equipment, regulatory approval, or complex human judgment in context. Goldman sees this as a way to reduce disruption risk in the AI trade. By concentrating on firms with hard assets and operating leases, the bank may be identifying businesses that can withstand technological change better than pure software platforms. For investors, the implication is to examine revenue durability, asset intensity, lease obligations, and the practical limits of automation before assuming that any AI beneficiary will keep soaring.
Goldman compiled a list of stocks using three main screens, and the methodology offers clues about how the bank thinks about AI investing. First, the firm looked for companies with a high proportion of hard assets and operating leases as revenue. Hard assets might include factories, vehicles, equipment, or property, while operating leases can provide flexible exposure to assets without immediate cash outflow. Second, it favored industries with high physical proximity, such as telecommunications, energy, transportation, and related sectors. Third, it included companies that had discussed how artificial intelligence could improve productivity on earnings calls. The final list reportedly included 25 stocks with at least thirty percent of their wages exposed to AI automation, according to Goldman Sachs. The wage figure is a rough proxy for labor exposure, but it is not the whole story. A business with physical assets may still have highly skilled labor, and a low automation score does not guarantee that AI will not change the business over time. Still, the screens capture a particular niche: firms where tangible assets and real-world operations are part of the economic model. For investors, the practical takeaway is to look beyond generic AI themes and ask whether a company has asset-heavy revenue, physical operations, and credible productivity benefits. That combination may produce a more resilient position than buying pure AI hype.
The 25 stocks that Goldman highlighted are not merely a thematic basket; they are a bet on where the next dollar of artificial intelligence profit might appear. The bank did not simply recommend that investors buy artificial intelligence. It recommended a specific type of company, one that combines digital capability with physical presence and measurable productivity gains. That distinction could be crucial as the market matures. Many AI stocks have delivered spectacular returns, but they have also been criticized for reliance on scarce compute, lofty valuations, and uncertain revenue timelines. By contrast, physical asset adopters may offer a more tangible link between technology spending and cost savings. For example, an airline using AI to optimize fuel routes, a retailer using demand forecasting, or a logistics firm using route planning could all see margin benefits. Goldman likely included companies that mentioned artificial intelligence on earnings calls, which suggests management believes the technology is moving beyond experimentation. The physical proximity criterion may also help investors avoid pure software plays that are vulnerable to rapid disruption. In a market where enthusiasm can fade quickly, a focus on assets that must be built, operated, and maintained may provide a more durable anchor. The risk is that the list is backward looking or incomplete, but the conceptual framework remains useful.
The reference to at least thirty percent of wages exposed to AI automation adds another layer of interest to Goldman’s list. This figure is not just about headcount; it speaks to how much of a company’s labor cost base could be affected by tools such as large language models, computer vision, robotics, or process automation. In physically intensive industries, automation may augment workers rather than fully replace them, but the potential for task-level substitution is still meaningful. A warehouse using AI to sort inventory, a factory using computer vision to inspect defects, or a call center using generative AI to draft responses could see significant cost pressure or optimization. Goldman is not necessarily saying that thirty percent of wages will disappear soon. It may be estimating exposure based on the nature of the business and the likely pace of adoption. Still, investors should take the number seriously because it highlights where labor economics and AI intersect. Companies with high physical proximity may have lower automation exposure than software firms, but they are not immune. The difference is that physical systems often require human oversight, maintenance, and local adaptation. That can make AI a complement rather than a pure substitute. For investors, the lesson is to map AI exposure at the task level, not just the company level, and to distinguish between automation potential and actual deployment.
Market context matters because the shift Goldman describes is not happening in a vacuum. The artificial intelligence trade has been extraordinarily powerful, but it has also produced concentration in a handful of large technology names. Investors have poured money into data centers, cloud regions, and specialized chips, often with the expectation that demand would keep rising. Now the bank expects capex growth to slow in 2027, which could change the marginal buyer for infrastructure. If the largest customers reduce spending, suppliers may face a tougher financing environment and a more competitive second half. That does not mean the AI era is ending. It means the investment narrative may need to evolve. The next phase could be about adoption, efficiency, and monetization. Goldman highlights adopters because they may be able to show cost savings, revenue uplift, or improved customer experiences faster than infrastructure providers. The hyperscalers are included because they can offer AI capacity while also building AI products. But the broader basket of physical asset companies could appeal to investors seeking a less abstract exposure. For portfolio managers, this is a reminder that thematic investing requires timing. The same AI trade that rewarded infrastructure can eventually reward the businesses that use AI to become more profitable. Understanding that transition may help investors avoid chasing already extended stocks or missing a potentially durable theme.
No investment thesis is complete without acknowledging the risks, and Goldman’s physical asset list is no exception. The expectation that AI capital expenditure growth will slow is a forecast, not a certainty, and actual spending could remain resilient for longer than Wall Street assumes. If new models create new demand for compute, the infrastructure trade might not fade as quickly as expected. Similarly, physical asset companies may still face margin pressure from labor, materials, energy costs, and competition. A candy bar company or ski resort might be less exposed to pure software disruption, but it is not automatically a low risk or high growth machine. The AI adoption story also depends on execution, data quality, and organizational change. Many companies may buy AI tools without redesigning workflows or generating meaningful returns. Goldman’s screens rely on management commentary and revenue characteristics, which are useful but imperfect signals. Investors should therefore treat the 25 stock list as a starting point rather than a final recommendation. The real opportunity may lie in identifying which adopters can convert AI into durable advantages. Without that discipline, a physical asset screen could become a value trap or a vague defensive trade. Risk management should include position sizing, liquidity checks, and an understanding of how AI exposure is changing.
For investors looking to position for the next phase of the AI trade, the Goldman note suggests a framework worth examining. Start with companies that have tangible assets or long-term operating leases tied to revenue. Then filter for industries where physical proximity matters, such as energy, transportation, telecommunications, or consumer goods. Next, look for management teams that have publicly linked artificial intelligence to productivity. From there, assess the degree of AI automation exposure, even if the final screen focused on at least thirty percent of wages. A portfolio could combine such names with broader AI infrastructure leaders to manage risk, or it could concentrate on a smaller set of high conviction adopters. The key is to avoid treating all AI spending as identical. Capital expenditures may slow, but adoption may accelerate. That means the marginal value of AI shifts from building models to integrating them into operations. Companies with physical assets may find it easier to defend margins because they cannot be easily virtualized. However, they still need to invest in software, sensors, and analytics. The smart approach is to look for firms that are using AI to improve asset utilization, maintenance, logistics, or customer experience. Those are areas where physical constraints can become durable competitive advantages.
The market may be entering a period where artificial intelligence enthusiasm is maturing into something more selective. Goldman’s outlook implies that the biggest driver of recent equity gains may be losing momentum, and that investors should prepare for a rotation. This does not mean that artificial intelligence is overrated. It means that the financial markets are moving from a phase defined by infrastructure spending to one defined by adoption and productivity. That shift can be harder to trade because it depends on corporate execution. Adopters must actually integrate tools, change processes, train staff, and measure outcomes. The physical asset screen may help investors find companies with real operational leverage, but it can also lead to value traps if the AI component is marginal. Investors should therefore look at free cash flow conversion, return on invested capital, lease coverage, and asset utilization. They should also watch whether hyperscalers are passing AI benefits down to customers. If cloud providers can offer better models and cheaper compute, they may become the true winners of the adoption phase. Meanwhile, traditional industrials, energy firms, and consumer staples could be overlooked until they are recognized as AI beneficiaries. The next few quarters may bring more clarity, but the groundwork for a new trade is being laid now.
So what should investors do with this Goldman Sachs note? First, watch whether AI capital expenditure growth actually slows in 2027. If it does, the market may reward adopters rather than pure infrastructure names. Second, screen for physical asset companies with hard assets or operating leases, physical proximity industries, and credible AI productivity commentary. Third, examine the automation exposure of at least thirty percent of wages, but interpret it alongside cash flows and competitive positioning. Fourth, consider building a basket of AI adopters across sectors such as energy, transportation, telecommunications, consumer goods, or industrials, rather than relying on a single technology index. Fifth, use risk controls such as position sizing and diversification because the AI trade may remain volatile. The actionable takeaway is not to abandon artificial intelligence investing entirely. It is to broaden the definition of AI winners beyond the obvious infrastructure names. Investors who want exposure to the next phase may rotate toward companies that combine digital efficiency with physical operations. They should confirm the thesis with earnings calls, order books, and productivity metrics before making large bets. In a market that rewards both innovation and resilience, the Goldman Sachs approach offers a roadmap, but execution still belongs to the investor.