Dell executive: This time, "proxy AI" is different; shortages of key components (mainly memory and HDDs) may last for more than 5 years.

Agent AI is changing the underlying logic of the traditional hardware cycle and may also cause infrastructure supply shortages to evolve from short-term fluctuations into structural constraints that last for years.

In a recent meeting with Morgan Stanley, Dell Technologies Chief Operating Officer Jeff Clarke stated that the AI cycle is significantly different from past hardware cycles. If AI workloads continue to grow, shortages of key components could persist for more than five years, with memory and HDDs being his biggest concerns. Compared to past hardware supply shortages that typically lasted several quarters, this round of demand growth could lead to a longer-term supply-demand gap.

Morgan Stanley reassessed the current AI infrastructure cycle following the meeting. Analysts pointed out that the eight hardware boom-bust cycles of the past 40 years have largely revolved around equipment upgrades and fluctuations in capital expenditure, while the key variable in this cycle lies in whether AI workloads can continue to expand. In other words, what Dell executives emphasized was not a typical hardware restocking cycle, but rather the long-term infrastructure expansion that may result from the growth in AI demand.

AI Reshapes Infrastructure Demand Logic

The report points out that Clarke believes this AI cycle is fundamentally different from the hardware booms and busts of the past few decades. The previous eight hardware cycles—covering categories such as servers, storage, and PCs—were mainly driven by changes in device penetration and upgrade cycles. Demand rose and fell with the pace of upgrades, but the hardware market itself did not experience substantial expansion.

Taking PC demand during the pandemic as an example, the human-machine ratio briefly increased, but as society reopened and returned to normal, PC demand declined, and the overall market size before and after the pandemic did not fundamentally change. The logic of proxy AI is different. Clarke believes that proxy AI is decoupling cognitive output from human input, allowing companies to complete more work without increasing manpower, thereby bringing about a 10-fold or even 100-fold increase in productivity.

In this process, AI is no longer just an assistant to humans in completing tasks, but has become a "productivity tool" that continuously consumes computing resources. Increased productivity will drive more companies, regardless of size, to begin AI transformation, thereby continuously expanding their infrastructure (TAM).

Dell projects that by 2030, data center computing power will increase by 200 gigawatts, ZettaFLOPS will grow fivefold to 830, and inference tokens will grow exponentially. The number of tokens required for agent AI to complete each unit of work is far greater than that of a basic chatbot; Clarke predicts that inference token generation will increase 87-fold by 2030.

Cloud and on-premises deployment are not a zero-sum game.

A key question in the market regarding this round of AI infrastructure cycle is whether the rapidly growing demand for tokens will ultimately flow primarily to cloud computing platforms rather than on-premises infrastructure.

Clarke believes that proxy AI workloads will adopt a hybrid deployment model. Enterprises will continuously adjust the location of inference workloads between public cloud and on-premises environments based on security and cost-effectiveness.

Clarke used Dell as an example: the company deploys content-related workloads in the public cloud, but does not move proprietary source code or telemetry data off its own facilities. This means that there is no single deployment path for AI workloads, and on-premises infrastructure will still handle a portion of the continuously growing computing demand.

Within this framework, the exponential growth in token consumption does not necessarily imply a trade-off between cloud and on-premises infrastructure. Clarke believes that the widespread adoption of open-source weighted models will further drive investment in on-premises infrastructure, as enterprises can optimize model outputs for cost in their local environments.

Server shipments decline does not change the long-term demand growth logic.

The current shipment volume of traditional servers is still declining year-on-year, which seems to contradict the explosive growth of inference tokens. Clarke explains that the significant increase in server performance density is suppressing the number of devices shipped.

Dell's 17th/18th generation servers can replace up to 13 legacy 14th generation servers, so even if server shipments decline in the short term, the value and capacity of a single device are still increasing.

As data centers gradually complete their architectural restructuring centered on accelerated computing, and the widespread adoption of proxy AI further drives demand for CPU servers, server shipments are expected to eventually rebound. Clarke even stated that if inference tokens truly achieve an 87-fold increase within five years, traditional server shipments could also experience exponential growth.

Storage will also be continuously driven by agent AI. Every operation performed by agent AI, including memory retention and artifact generation, will generate continuous data storage demands; with the large-scale application of KV Cache, storage capacity requirements will further increase.

Supply shortages may become a structural constraint for more than five years.

Compared to macroeconomic factors, geopolitical conflicts, data center overcapacity, and power shortages, Clarke listed supply issues as the biggest concern at present, describing the current supply environment as "We're in neverland."

Morgan Stanley points out that historically, commodity supply cycles typically last two to four quarters, primarily influenced by hardware upgrade cycles and supply-side decisions. However, Clarke argues that if the growth logic of gigawatts and inference tokens is accepted, then key components such as memory and HDDs are no longer simply in a typical short-term cycle.

Therefore, Dell is currently operating under the assumption that there may be a multi-year shortage of key components, with a particular focus on memory and HDDs. In other words, the core risk of this supply shortage is not a short-term lack of supply, but rather that the supply side may need several years to catch up with the continued expansion of AI demand.

Clarke also stated that Dell excels in supply management compared to its competitors, and this advantage is translating into increased market share across multiple segments, including servers, storage, and PCs.

The expansion of profit margins reflects changes in the supply and demand structure.

The continued shortage of key components is also changing the pricing environment in the infrastructure industry.

Morgan Stanley previously attributed the expansion of gross margins in its server and storage businesses to "profit stacking," meaning that as the cost of components such as memory rose, manufacturers further increased their profit margins. Clarke, however, offered a different explanation.

He acknowledged that profit margins for similar products, such as servers and storage, are indeed expanding, but believes there are two structural factors behind this: First, with limited component supply, companies prioritize allocating scarce resources to the most profitable products; second, as the infrastructure market continues to expand, the pressure on companies to compete for incremental customers has decreased, eliminating the need to lower prices to acquire new customers.

Therefore, the increase in profit margins is not entirely due to opportunistic gains from short-term supply-demand mismatch, but also reflects changes in industry pricing benchmarks following the continued expansion of infrastructure demand.

If the workload growth driven by proxy AI continues to materialize, the core variable in the AI infrastructure cycle will shift from "equipment upgrades" to "new demand." This means that the expansion in demand for computing power, servers, and storage may continue for longer, while supply constraints on key components such as memory and HDDs may shift from short-term cyclical fluctuations to multi-year structural problems.

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.