Beyond Nvidia: Where the AI Infrastructure Opportunity Is Expanding

Memory is becoming central to the economics of AI

The growing importance of memory represents one of the most consequential changes in the semiconductor investment landscape. Traditional computing systems have generally been evaluated through the performance of their processors, but advanced AI workloads place substantial demands on the speed and volume of data moving between processing units and memory. A processor capable of performing enormous numbers of calculations offers limited practical advantage if it spends too much time waiting for the information required to complete them.

High-bandwidth memory addresses part of this problem by placing multiple layers of memory in close proximity to advanced processors, allowing data to move at substantially higher rates than would be practical through conventional memory arrangements. Its importance is particularly pronounced in systems designed to train and operate large AI models, where memory bandwidth, capacity and power efficiency can materially influence overall performance. As a result, the commercial value of an AI accelerator increasingly depends on the wider memory architecture surrounding it rather than its computational capability in isolation.

The financial consequences are becoming evident in the results of major memory manufacturers. In its fiscal fourth-quarter and full-year 2026 results, published on 30 September, Micron reported quarterly revenue of $54.23 billion, compared with $11.32 billion a year earlier. Full-year revenue increased from $37.38 billion to $133.19 billion, while the company's reported annual gross margin rose from 39.8% to 80.7%.

Those figures illustrate the exceptional strength of the current memory market, although they should not be interpreted as evidence that such profitability has become a permanent industry characteristic. Memory manufacturing has historically been highly cyclical, with periods of constrained supply and strong pricing encouraging investment that can eventually create excess capacity. The transition towards increasingly sophisticated memory products may improve the competitive position of manufacturers capable of meeting demanding technical requirements, but it does not eliminate the industry's sensitivity to supply, inventories and customer purchasing behaviour.

The distinction matters because not all memory demand is driven by the same economic forces. High-bandwidth memory requires advanced manufacturing processes and complex integration with processors, while conventional DRAM and NAND remain exposed to demand across personal computers, smartphones, enterprise systems and other applications. Capacity decisions made in one part of the market can influence supply conditions elsewhere, particularly when manufacturers redirect investment towards products offering higher margins.

Micron's financial results demonstrate that the present cycle is producing substantial profitability, but the sustainability of those returns will depend on how capacity evolves relative to demand. Its reported fiscal 2026 capital expenditure of approximately $27.37 billion also illustrates the financial resources required to participate in this expansion. Investors must therefore distinguish between the structural increase in memory requirements associated with AI and the cyclical profitability that may arise when demand temporarily exceeds available supply.

The longer-term opportunity appears substantial because improvements in processing capability do not remove the need to store and move information efficiently. Nevertheless, the investment case for memory manufacturers depends on more than the assumption that AI demand will continue growing. Product differentiation, manufacturing yields, customer agreements and the discipline with which the industry adds capacity will help determine how much of that growth ultimately becomes durable economic value.

Advanced packaging is changing how semiconductor performance is created

The increasing complexity of AI processors is also changing the importance of semiconductor packaging. Historically, much of the industry's technological progress was associated with manufacturing smaller transistors and placing more of them onto individual chips. That remains essential, but advanced computing systems increasingly rely on combining multiple semiconductor components within a single package, creating opportunities for technologies that enable those components to communicate efficiently.

This approach allows manufacturers to integrate processing units, memory and other specialised components without requiring every function to be produced on a single piece of silicon. It introduces additional engineering challenges, however, because the individual components must be connected with exceptional precision while managing heat, electrical performance and manufacturing reliability. Advanced packaging is consequently becoming an important part of semiconductor performance rather than merely the final stage of assembly.

TSMC occupies a particularly influential position within this development. Its chip-on-wafer-on-substrate technology, commonly known as CoWoS, enables the integration of advanced processors and high-bandwidth memory within sophisticated computing packages. The importance of this technology reflects the growing need to move data efficiently between different components, making packaging capacity and manufacturing yields increasingly relevant to the availability of complete AI systems.

The breadth of the opportunity became clearer on 8 October, when GlobalFoundries announced a multiyear $2 billion manufacturing agreement with TSMC. Under the arrangement, GlobalFoundries will manufacture silicon interposers at its Malta, New York facility for use within TSMC's advanced packaging ecosystem. These components provide high-density connections between semiconductor dies, illustrating how the expanding complexity of AI systems is creating opportunities for manufacturers beyond those producing the most advanced processors.

The agreement is significant because it demonstrates that semiconductor manufacturing value is not confined to the leading edge of transistor technology. GlobalFoundries does not compete with TSMC in manufacturing the most advanced logic processes, but its ability to produce specialised components can nevertheless become valuable when those components address a constraint within the broader manufacturing system. The commercial opportunity arises from providing an essential capability within an increasingly complicated supply chain.

For investors, however, identifying an important technological function is only the beginning of the analysis. The profitability of advanced packaging depends on manufacturing yields, equipment utilisation, customer concentration and the capital required to expand production. Suppliers may benefit from constrained capacity in the near term, but increasing investment and technological change can alter the balance between supply and demand over time.

There is also a distinction between companies supplying advanced packaging equipment, manufacturers producing specialised components and businesses undertaking final assembly and testing. Each participates in the same structural opportunity but faces different competitive conditions and capital requirements. Understanding those differences is essential before assuming that growth in advanced packaging demand will translate into equivalent financial returns across the industry.

Manufacturing capacity remains a critical constraint

The expansion of AI infrastructure ultimately depends on the ability of semiconductor manufacturers to produce increasingly sophisticated chips in sufficient quantities. Advanced manufacturing requires extraordinary technical precision, substantial investment and production processes that can take years to develop and optimise. These characteristics create considerable barriers to entry, but they also make the industry highly sensitive to decisions about capacity expansion and the timing of customer demand.

TSMC's financial performance provides evidence of the strength of the current manufacturing cycle. According to its September 2026 revenue announcement, published on 8 October, monthly revenue reached approximately NT$511.86 billion, representing growth of 54.6% compared with September 2025. Its monthly revenue disclosures indicate that third-quarter revenue totalled approximately NT$1.49 trillion, although these figures remain unaudited and do not establish the quarter's profitability.

The company's previously published second-quarter results offer additional perspective on its manufacturing position. TSMC reported quarterly revenue of NT$1.27 trillion and a gross margin of 67.7%, while technologies at seven nanometres and below accounted for 77% of wafer revenue. The contribution from advanced manufacturing processes illustrates how demand for increasingly sophisticated semiconductors is influencing the company's product mix and profitability.

These results are important because manufacturing leadership can provide advantages extending beyond immediate production capacity. Customers developing advanced processors must consider the reliability of manufacturing processes, expected yields, available capacity and the ability to move new designs into volume production. A manufacturer capable of delivering consistently across those requirements may possess considerable commercial leverage, particularly when alternatives are limited.

The equipment suppliers enabling that manufacturing process occupy another strategically important position. ASML's extreme ultraviolet lithography systems are essential to producing many of the most advanced semiconductor designs, giving the company exposure to investment decisions made across leading-edge manufacturing. Its business is therefore influenced not only by current semiconductor demand, but also by the capacity that manufacturers expect to require several years into the future.

In its third-quarter 2026 results, published on 14 October, ASML reported net sales of €11.66 billion, compared with €9.33 billion in the preceding quarter, while its gross margin improved from 54.0% to 57.5%. Net income reached €4.96 billion, compared with €2.92 billion in the second quarter, reflecting the strength of the company's financial performance as semiconductor manufacturers accelerate their investment plans. Management also raised its full-year revenue expectations to between €44 billion and €45 billion, with a gross margin of approximately 56%, providing further evidence of sustained demand for advanced manufacturing equipment.

The most significant development, however, may be the company's assessment of future equipment requirements. ASML disclosed that customers are continuing to increase their short- and medium-term demand forecasts, with some expressing interest in placing orders extending into 2029 and 2030. Management now expects its currently planned production capacity for both 2027 and 2028 to fall short of customer requirements, despite previously announced expansion plans, and is investigating additional increases in manufacturing capacity. These statements suggest that the investment cycle supporting advanced semiconductor production may extend considerably further than the current period of exceptional demand, although customer forecasts and expressions of interest should not be treated as binding commitments or guaranteed future revenue.

This development is particularly relevant to the broader AI infrastructure investment case because it demonstrates how demand for computing capacity is translating into increasingly long-term commitments across the semiconductor manufacturing supply chain. The scarcity of sophisticated lithography equipment may support ASML's competitive position and pricing power, but the company's ability to convert that demand into revenue will depend on production capacity, supplier availability and the timing of customer investment decisions. The distinction between strong demand and realised financial returns remains important, particularly when valuations already incorporate expectations of prolonged industry expansion.

TSMC's third-quarter earnings conference remains scheduled for Thursday 15 October, as confirmed by its investor relations calendar. Its revenue disclosures already establish substantial growth, but investors will need the forthcoming results to assess margins, capital expenditure and management's latest expectations. Distinguishing between known revenue performance and financial information that has yet to be released is particularly important when valuations reflect demanding assumptions about future profitability.

Connectivity is becoming as important as computing power

As AI systems expand, the movement of information between processors and servers becomes increasingly important to their overall performance. A modern AI installation may contain thousands of interconnected computing components, each requiring access to data and the ability to communicate with other parts of the system. The performance of the installation therefore depends not simply on the speed of individual processors, but also on the efficiency with which the wider network coordinates their activity.

This creates demand for advanced networking equipment, high-speed switching, optical connections and specialised semiconductor components designed to move information with minimal delay and power consumption. The technical requirements become more demanding as systems expand, particularly when processors must exchange large volumes of data during training or support increasingly complex inference workloads. Improvements in connectivity can consequently increase the productive capacity of an existing computing installation without necessarily requiring an equivalent increase in processing hardware.

Broadcom provides evidence of how quickly this opportunity is developing. In its third-quarter fiscal 2026 results, published on 2 September, the company reported AI semiconductor revenue of $16.7 billion, an increase of 221% from the corresponding period a year earlier. Management attributed the performance to strong demand for custom AI accelerators and networking products, while forecasting AI semiconductor revenue of approximately $21.7 billion for the following quarter.

The distinction between reported revenue and management guidance is important, but both figures demonstrate the scale of demand for technologies supporting AI infrastructure. Broadcom's opportunity extends beyond supplying components for general-purpose computing systems, reflecting the growing interest among major technology companies in designing specialised processors and communications architectures for their own workloads. Such arrangements can create substantial commercial relationships, although they may also increase dependence on a relatively small number of customers.

Marvell occupies another position within this expanding market. At its investor day on 6 October, the company emphasised the increasing importance of connectivity across AI infrastructure, including optical interconnects, switching and custom semiconductor designs. Its strategic argument is that the continued expansion of computing capacity creates corresponding requirements for moving data across increasingly large and sophisticated systems.

That proposition has considerable economic logic, but the investment implications require careful assessment. Connectivity suppliers may benefit from increasing technical complexity and the need for greater bandwidth, yet they also face competition, evolving industry standards and the possibility that customers will develop more of their own technology. Revenue growth associated with a rapidly expanding market does not automatically establish that every supplier will retain attractive pricing power.

There is also an important relationship between connectivity and energy efficiency. Moving information consumes power, and the cost of doing so becomes more significant as systems expand. Technologies capable of reducing communication delays and electricity consumption may therefore become increasingly valuable, particularly where data centre operators are constrained by the amount of power available to their facilities.

The opportunity increasingly extends beyond semiconductors

The physical infrastructure supporting AI computing introduces another dimension to the investment cycle. Advanced processors, memory and networking equipment generate substantial heat and consume considerable electricity, requiring sophisticated cooling systems, power distribution equipment and reliable access to energy. As computing installations become more powerful, the supporting infrastructure must evolve alongside them.

The scale of the potential electricity requirement is substantial. In its Energy and AI report, published in April 2025, the International Energy Agency projected that global electricity consumption by data centres could more than double to approximately 945 terawatt-hours by 2030. This is a forecast rather than an established outcome, but it illustrates the extent to which the expansion of digital infrastructure could influence electricity networks and investment in energy systems.

The implications extend beyond the generation of electricity. Data centres require equipment capable of distributing power reliably, managing fluctuations in demand and maintaining operating conditions suitable for expensive computing hardware. Cooling systems are similarly important, particularly as the concentration of processing power within individual installations increases.

Companies supplying these technologies are already reporting substantial growth. In its second-quarter 2026 results, Vertiv reported revenue of approximately $3.27 billion, an increase of 24% from the previous year, alongside improvements in operating profitability. The company attributed its performance partly to continued demand for power and thermal management infrastructure associated with AI and conventional computing.

Eaton's second-quarter financial results provide a complementary perspective. The company reported that the twelve-month rolling average of orders in its Electrical Americas business increased by 41% organically, while the electrical businesses continued to experience substantial backlog growth. These figures suggest that demand for supporting electrical infrastructure is extending beyond individual semiconductor suppliers into businesses providing essential components for the wider investment programme.

The importance of these developments lies in the relationship between technological ambition and physical capacity. A data centre cannot operate simply because the required processors have been purchased; it must also secure an appropriate location, electrical connections, cooling equipment and sufficient power. Delays in any of these areas can postpone the deployment of computing capacity and alter the timing of revenue for suppliers throughout the infrastructure chain.

This creates opportunities for companies capable of addressing practical constraints, but it also introduces risks. Large infrastructure projects require substantial capital, and their economics can change when financing costs rise, electricity prices increase or expected demand fails to materialise on schedule. Suppliers may report strong orders and expanding backlogs while remaining exposed to project delays, changes in customer investment plans and the possibility that anticipated expenditure is deferred.

The most attractive businesses will not necessarily be those experiencing the fastest increase in orders. Their longer-term performance will depend on whether they can convert demand into profitable deliveries, maintain pricing discipline and expand capacity without undermining returns on invested capital. The same financial principles that apply to semiconductor manufacturers therefore remain relevant to the businesses supplying the electricity and cooling systems surrounding them.

Structural growth does not eliminate cyclical risk

The breadth of the AI infrastructure opportunity is becoming increasingly evident, but the existence of structural demand should not be confused with immunity from the semiconductor industry's traditional cycles. Investment in manufacturing capacity, inventories and technological development has historically produced periods of considerable volatility, particularly when expectations of future demand encourage suppliers to expand simultaneously. AI may alter the scale and composition of demand without removing the economic consequences of excessive investment.

Memory provides a particularly clear example. Strong demand and constrained supply can support exceptional pricing and profitability, encouraging manufacturers to commit substantial capital to new facilities and production technologies. Those investments may be economically justified, but the capacity they create can eventually change market conditions, particularly if customers reduce purchases or improvements in manufacturing productivity increase available supply.

Advanced manufacturing faces different constraints, although the underlying investment problem is similar. New facilities require substantial expenditure and long development periods, making it difficult to adjust capacity quickly when demand changes. Manufacturers with strong competitive positions may be better placed to manage these cycles, but they remain exposed to utilisation rates, customer concentration and the possibility that technological progress changes the economics of existing assets.

The same considerations apply to networking and infrastructure suppliers. Rapid growth can attract competitors, encourage customers to negotiate more aggressively and create pressure to expand manufacturing or service capacity. Businesses that appear exceptionally profitable during a period of constrained supply may experience less attractive economics once capacity catches up with demand.

Valuation adds another layer of uncertainty. A company can participate successfully in a major technological transition while delivering disappointing shareholder returns if its initial share price already assumes exceptional growth and profitability. Investors must therefore distinguish between the attractiveness of the underlying industry and the returns available from purchasing a particular business at its prevailing valuation.

This distinction is especially important where forecasts extend many years into the future. Small changes in assumptions about revenue growth, operating margins, capital expenditure or the cost of capital can materially influence estimates of intrinsic value. The further those expectations extend beyond observable financial performance, the more sensitive an investment becomes to changes in market confidence.

The appropriate response is not to dismiss the structural opportunity, but to examine its economics with greater precision. Businesses benefiting from scarce manufacturing capabilities, differentiated technology or difficult-to-replicate customer relationships may possess advantages capable of supporting durable returns. Nevertheless, the strength of those advantages must be assessed alongside the capital required to maintain them and the valuation investors are being asked to accept.

Where the next phase of value creation may emerge

The expansion of AI infrastructure is creating a more interconnected investment landscape than the familiar focus on leading processor designers might suggest. Memory bandwidth, advanced packaging, manufacturing capacity, networking performance and electricity availability are increasingly linked, with constraints in one area influencing the commercial opportunities available elsewhere. As a result, the next stage of industry development may be shaped as much by improvements in the efficiency of complete computing systems as by advances in individual processors.

This creates several possible sources of economic value. Memory manufacturers may benefit from increasing requirements for capacity and bandwidth, while advanced packaging specialists can address the technical challenges associated with integrating increasingly sophisticated components. Leading foundries and equipment suppliers remain essential to expanding production, and networking businesses may benefit as computing installations require faster and more efficient communication.

The opportunity also extends into electrical infrastructure, cooling and energy systems, where physical constraints can influence the pace at which computing capacity becomes operational. These businesses may offer exposure to the expansion of AI infrastructure without depending directly on the commercial success of a particular processor architecture. Their financial performance will nevertheless remain connected to the willingness and ability of customers to continue investing in large computing installations.

For investors, the central challenge is determining which companies can convert this expanding demand into sustainable economic returns. Revenue growth provides evidence that the opportunity is developing, but profitability, cash generation, capital requirements and competitive positioning determine how much value ultimately reaches shareholders. The quality of that value creation must then be considered against the expectations already reflected in market valuations.

ASML's newly published third-quarter results provide further evidence of the strength and potential duration of semiconductor manufacturing investment, particularly given management's assessment that currently planned equipment capacity may be insufficient to meet customer demand through 2028. TSMC's forthcoming earnings announcement will offer an additional perspective on manufacturing profitability and capital expenditure, while Micron's recent financial performance, together with disclosures from Broadcom and infrastructure suppliers, already demonstrates that the benefits of AI investment are extending across a much wider industrial ecosystem. The question is whether those benefits can remain sufficiently profitable as capacity expands, competition evolves and customers become more demanding about the returns on their own expenditure.

Nvidia's success has helped establish the scale of the opportunity, but the longer-term development of AI infrastructure will depend on many businesses operating beyond the most visible parts of the semiconductor market. Some will benefit from technological leadership, others from manufacturing capabilities or their position within essential supply chains. The most compelling investment opportunities will ultimately be those where these advantages support attractive returns on capital at valuations that leave room for uncertainty.

The AI infrastructure cycle is therefore best understood not as a single investment theme centred on one company, but as a broad industrial transformation involving numerous interdependent technologies. Its potential is considerable, yet the distribution of economic value across that ecosystem remains uncertain. Understanding where the constraints are emerging, which companies are capable of resolving them and how much capital they must commit in doing so will be central to identifying where durable shareholder value is most likely to develop.