Dell COO warns agent AI could cause memory and HDD shortages lasting over five years

Dell Technologies (DELL.US) Chief Operating Officer Jeff Clarke, in a recent meeting with Morgan Stanley, stated that agent AI represents a fundamental departure from past hardware cycles. If related AI workloads continue to grow, shortages of key components could persist for more than five years, with memory and HDDs being his primary concerns.

Unlike typical hardware supply tightness that lasts several quarters, this wave of demand may create a longer-term supply-demand gap.

Following the meeting, Morgan Stanley reassessed the current AI infrastructure cycle. Analysts noted that the eight hardware boom-and-bust cycles over the past 40 years have largely revolved around equipment upgrades and capital expenditure fluctuations. The key variable in this cycle is whether AI workloads can sustain their expansion. In other words, Dell's executives are not describing an ordinary hardware replenishment cycle, but rather a potential long-term infrastructure expansion driven by growing agent AI demand.

Agent AI Reshapes Infrastructure Demand Logic

According to the report, Clarke believes this AI cycle is fundamentally different from the hardware booms and busts of the past several decades. Previous hardware cycles—covering servers, storage, PCs, and other categories—were primarily driven by changes in device penetration rates and upgrade cycles. Demand rose and fell with upgrade rhythms, but the overall size of the hardware market did not expand substantially.

For example, during the pandemic, PC demand briefly increased as the human-to-device ratio rose. However, as society returned to normal, PC demand receded, and the total market size before and after the pandemic remained essentially unchanged.

Agent AI operates on a different logic. Clarke argues that agent AI decouples cognitive output from human input, allowing companies to accomplish more work without adding staff, resulting in 10x to 100x productivity improvements.

In this process, AI is no longer just a tool to assist humans; it becomes a "productivity tool" that continuously consumes computing resources. Productivity gains will drive more enterprises—regardless of size—to undergo AI transformation, thereby continuously expanding the infrastructure total addressable market (TAM).

Dell projects that by 2030, data center computing power will increase by 200 gigawatts, ZettaFLOPS computing power will grow fivefold to 830, and inference token generation will grow exponentially. Agent AI requires significantly more tokens per unit of work compared to basic chatbots. Clarke expects inference token generation to increase 87-fold by 2030.

Cloud and On-Premises Deployment Are Not a Zero-Sum Game

A core question for the market regarding this AI infrastructure cycle is whether rapidly growing token demand will ultimately flow primarily to cloud platforms rather than on-premises infrastructure.

Clarke's assessment is that agent 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.

Using Dell as an example, Clarke noted that the company deploys content-related workloads on the public cloud but will not move proprietary source code or telemetry data out of its own facilities. This means there is no single deployment path for AI workloads; on-premises infrastructure will still absorb a portion of the growing computing demand.

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

Server Shipment Declines Do Not Alter Long-Term Demand Growth Logic

Current traditional server shipments are still declining year-over-year, which appears to contradict the explosive growth in inference tokens. Clarke explains that significant increases in server performance density are reducing device shipment volumes.

Dell's 17th/18th generation servers can replace up to 13 traditional 14th generation servers. Therefore, in the short term, even if server shipments decline, the value and capacity of individual units continue to rise.

As data centers gradually complete their architectural transformation toward accelerated computing, and as agent AI adoption further drives CPU server demand, server shipments are ultimately expected to resume growth. Clarke even stated that if inference tokens truly achieve 87-fold growth within five years, traditional server shipments could also experience exponential changes.

Storage will also be continuously driven by agent AI. Every operation performed by agent AI—including memory retention, artifact generation, and others—generates ongoing data storage demand. With the large-scale application of KV Cache, storage capacity requirements will grow further.

Supply Shortages May Become a Structural Constraint for Five Years or More

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

Morgan Stanley notes that historical commodity supply cycles typically last 2 to 4 quarters, primarily driven by hardware upgrade cycle booms and busts and supply-side decisions.

However, Clarke argues that if the growth logic for gigawatts and inference tokens is accepted, then key components such as memory and HDDs are no longer in an ordinary short-term cycle.

Consequently, Dell is currently operating under the assumption that key components may face multi-year persistent shortages, with particular focus on memory and HDDs.

In other words, the core risk of this supply tightness is not short-term insufficiency, but rather that after sustained AI demand expansion, the supply side may take years to catch up.

Clarke also stated that Dell's supply management is superior to its peers, and this advantage is translating into market share gains across multiple segments, including servers, storage, and PCs.

Margin Expansion Reflects Structural Changes in Supply and Demand

Persistent tightness in key components is also changing the pricing environment in the infrastructure industry.

Morgan Stanley previously attributed part of the gross margin expansion in server and storage businesses to "profit stacking"—where manufacturers add additional profit margins on top of rising component costs. Clarke offered a different explanation.

He acknowledged that profit margins for similar server and storage products are indeed expanding, but attributed this to two structural factors:

First, when component supply is limited, companies prioritize allocating scarce resources to the highest-margin products. Second, as the infrastructure market size continues to expand, the pressure to compete for incremental customers decreases, eliminating the need to lower prices to attract new clients.

Therefore, margin improvement is not entirely an opportunistic gain from short-term supply-demand mismatches. It also reflects a change in the industry's pricing benchmark following sustained infrastructure demand expansion.

If the workload growth brought by agent AI continues to materialize, the core variable of the AI infrastructure cycle will shift from "equipment upgrades" to "new demand." This means that demand expansion for computing power, servers, and storage may persist longer, and supply constraints for key components such as memory and HDDs could transform from short-term cyclical fluctuations into multi-year structural issues.

This article is reprinted from "Wall Street CN." Edited by Chen Siyu, Zhitong Finance.