Dell Technologies is acknowledging that infrastructure and storage supply still cannot keep up with the enormous resource demands of agentic AI. The company reported a record AI backlog of $95 billion in orders waiting to be filled, a sign that the infrastructure crunch across the data center industry is far from over.

Dell’s quarterly earnings also reflected a more than 50% year-over-year increase in AI demand, underscoring how quickly enterprises, cloud providers, and other organizations are shifting toward AI-driven workloads. For Dell’s management, the message is clear: suppliers are still struggling to deliver enough servers, storage systems, memory modules, networking gear, and other critical components to satisfy an accelerating market.

Dell COO Jeff Clarke acknowledged on the earnings call that supply constraints begin with servers and storage and extend across the broader technology stack. Clarke said the constraints now touch "just about every product going through a leading node," pointing to shortages in DRAM, NAND flash, CPUs, disk drives, microcontrollers, and other advanced components.

"We are doing everything we can to get more supply," Clarke said. "In today's environment, that's a very difficult task."

A glimpse of infrastructure demands ahead

Dell’s results for the financial quarter ending July 31 show that the appetite for AI infrastructure remains intense. The company posted $47 billion in revenue, a 58% year-over-year jump. More notably, revenue in Dell’s Infrastructure Solutions Group (ISG) surged 89% to a record $31.8 billion, driven by both conventional server deployments and AI-optimized systems.

Much of that growth came from servers powered by traditional central processing units (CPUs). These so-called CPU-based servers are increasingly running agentic AI workloads, and Dell said demand in this area was "exceptionally strong" with earnings up 122% year-over-year. The fact that mainstream servers are now part of the AI expansion underlines how agentic AI is being integrated into enterprise workflows instead of being limited to specialized GPU clusters.

Perhaps the most telling figure in the report is the pace of new bookings. Dell booked nearly $61 billion in AI server orders in the three months ending July 31 alone. Over the last 12 months, the company has signed more than $130 billion in AI server orders. Clarke also said Dell converted $131.7 billion of demand into orders over the past year. That figure includes broad demand from enterprise customers, neoclouds, sovereign cloud providers, and other organizations building or expanding AI-ready data centers.

The growing customer base reflects more than just interest. Dell reported that more than 6,500 customers are now using Dell AI Factory, the company’s integrated platform for building AI workflows and applications. Of those, 3,300 customers signed on during the last three quarters. By contrast, it took Dell roughly two years to sign on its first 3,200 customers after the AI Factory launched in May 2024. That pace suggests enterprise adoption of AI infrastructure is accelerating much faster than earlier waves of technology spending.

Agentic AI and inference reshape data centers

Clarke described agentic AI as a force that is reshaping the modern data center. He said inference is now "pure demand" in the industry because enterprises are deploying models that need continuous processing and decision-making. As organizations build agents capable of taking action, they need more compute capacity, more memory, more storage, and more resilient power and networking architectures to keep data moving without interruption.

Dell’s projections for the next several years illustrate the scale of the shift. The company anticipates that 3,600 quadrillion tokens will be in use by 2030, representing an 87-fold increase from today. Training demand is also expected to grow to 850 zettaflops by 2030, a fivefold jump from current levels. These numbers point to a data center market that will need continuous upgrades and expansion across every layer of infrastructure.

Clarke said enterprise agentic AI is expected to become the single largest workload by 2028 and could account for 75 percent of all data center demand by 2030. That projection helps explain why Dell and other major infrastructure providers are building backlogs measured in tens of billions of dollars. It also explains why component shortages are unlikely to disappear quickly, even as semiconductor manufacturers announce new fabrication plants and suppliers increase memory production.

Enterprises clamor for traditional servers

Dell is seeing another significant trend: customers increasingly require "meaningful CPU compute capacity" to support AI and agentic workflows. This is not limited to GPU-based systems. Enterprises are refreshing their traditional server fleets to handle the massive volumes of data that accompany AI initiatives, while also modernizing core applications that need more processing power.

The demand has been striking in recent quarters. In just its last two financial quarters, Dell generated nearly as much revenue from traditional servers and networking as it had in any prior full year in the company’s history. Much of the growth is coming from existing customers who are accelerating investments in traditional IT environments to refresh aging equipment, improve performance, increase efficiency, and harden their infrastructures against evolving threats.

Clarke said Dell anticipates "significant and durable" refresh cycles ahead. Security and resiliency requirements are also pushing customers to replace older servers and storage arrays. AI workloads demand modern, disaggregated architectures that can keep data accessible and moving across compute, storage, and networking resources. He noted that AI deployments involve much more than assembling and shipping components. Many customer engagements require engineering, design, and deployment expertise, with some individual customers asking for as many as 50 unique server or system designs as they optimize for workload performance, power, cooling, and physical data center constraints.

Enterprises want new servers with more cores, more dynamic random-access memory (DRAM), and more storage capacity. Yet the biggest bottlenecks remain memory and flash storage. Clarke summarized the situation bluntly: "DRAM, DRAM, DRAM, followed by NAND, NAND, NAND." There are also spotty shortages of CPUs and disk drives, and constraints run all the way down the supply chain, from microcontrollers to transistors to complete storage systems.

Infrastructure costs and customer responses

The shortage environment is not just limiting volume; it is also changing costs. Clarke acknowledged that modernization is driving higher core counts, more DRAM, and more storage. Those richer configurations "cost more than they did last quarter, and the quarter before, and the quarter before." Rising component prices, transportation costs, and the cost of advanced manufacturing are all flowing into system pricing.

Customers are adjusting in different ways. Some are deferring purchases because they cannot flex existing budget dollars enough to absorb higher prices. Others are placing orders much further ahead of their actual deployment time to secure access to constrained supplies. Clarke said large and sophisticated customers are acting first, often working collaboratively with Dell to project their infrastructure needs further into the future.

That kind of advanced planning is a new phenomenon, according to Clarke. In the past, enterprise buyers tended to order closer to the time of deployment. Now, with lead times extending across memory, storage, and other components, customers are tightening their partnerships with suppliers and making commitments earlier than they normally would. Clarke noted Dell is trying to help customers manage through the environment, even while the company remains limited by how much it can build in any given quarter.

Storage demand grows as data volumes mount

The infrastructure crunch is also visible in storage products. Dell reported strong growth across its PowerFlex, PowerStore, PowerProtect, and PowerVault product lines. These systems help enterprises manage, protect, and secure vast amounts of data that is increasingly crucial for AI model training, retrieval, and inference.

PowerFlex provides software-defined block storage that can scale independently of compute and is often used in demanding database and enterprise AI workloads. PowerStore is Dell’s midrange all-flash array family, designed for modern applications that need low latency and high performance. PowerProtect focuses on data protection, backup, and cyber resilience, an increasingly critical area as organizations try to recover quickly from ransomware attacks and other incidents. PowerVault spans storage for file, block, and archive use cases, giving customers flexible options for managing data growth.

Clarke said storage demand remains broad-based as enterprises modernize their environments and data growth makes it more important to keep data available and secure. The challenge, however, is that storage is also constrained. NAND flash, disk drives, and the controllers and packaging materials required to build storage systems are all in tight supply. That means Dell’s storage backlog is likely to remain high even as the company works to expand manufacturing output.

While Dell’s server business has attracted the most attention because of its blockbuster earnings growth, storage is becoming a more strategic part of large AI infrastructure deals. Agentic AI systems need not only raw compute but also persistent data repositories, metadata management, retrieval pipelines, and secure backup environments. The company’s storage customers are adding capacity in parallel with server purchases, compounding the pressure on component supply chains.

Supply constraints and the road ahead

For Dell and the wider infrastructure industry, the central question is not whether AI demand will continue, but whether supply can ever fully catch up. The company’s $95 billion backlog reflects orders that already had to be scheduled and potentially held because components could not be shipped fast enough. With AI server orders still arriving at a record pace and traditional servers also experiencing strong demand, Dell has to make difficult allocation decisions.

Clarke said Dell is planning accordingly and optimizing configurations with the components it does receive. The company is focused on getting systems out the door as quickly as possible while still designing products that meet each customer’s performance, power, and cooling requirements. Dell has increased shipping volumes, managed lead times with customers, and prioritized the systems that can be completed with available "bits and bytes."

"We'll continue to focus on trying to get more supply, and take the supply we have and optimize the output," Clarke said. That approach suggests the infrastructure crunch is not a short-term disruption. It is the new operating reality for a data center industry trying to build the compute capacity for agentic AI, massive token workloads, and a data economy that shows no signs of slowing.

The record backlog is important not just for Dell’s quarterly result, but for what it signals about the broader market. Enterprises, cloud providers, sovereign governments, and telecommunication companies are all competing for the same underlying components. Even after current expansion plans in the semiconductor industry come online, demand is expected to continue growing as AI moves deeper into everyday business processes.

Clarke’s observation that some customers are now planning future needs collaboratively with Dell is a strong indicator that the shortage era is likely to persist. The companies placing large orders now are trying to secure capacity not only for their immediate projects but also for the next waves of Open RAN, sovereign AI, edge computing, and enterprise automation. As those workloads become more embedded in daily operations, infrastructure providers will face even greater pressure to keep pace.

Dell’s results also make clear that the infrastructure crunch is no longer limited to GPU-accelerated systems. Traditional servers are now front and center in the AI buildout, as are storage arrays, networking switches, and power distribution systems. The modern data center must support both centralized training environments and distributed inference applications, with data continuously in motion across every layer.

The company expects enterprise agentic AI to become the dominant workload in the next few years. If those projections prove accurate, the industry will need massive investments in fabs, memory production, storage media, and system design. Dell’s executives say they are working with customers and suppliers to increase capacity, but the current environment makes it difficult to predict when supply will finally match demand. For now, the backlog is a clear reminder that AI infrastructure remains one of the most constrained and most important markets in technology.

Source: Network World News