Dell Technologies has put a fresh number on the infrastructure industry’s biggest problem. The company is carrying a record $95 billion backlog of AI orders, a vivid sign that the supply of AI-ready data center hardware is still nowhere close to meeting demand.
The backlog was disclosed alongside quarterly results that showed nearly every part of Dell’s infrastructure business is now being reshaped by AI. Revenue climbed at a double-digit rate, infrastructure sales hit a record, and the order pipeline continued to expand even as customers waited longer for components. Dell’s chief operating officer, Jeff Clarke, said supply constraints begin with servers and storage and stretch across the technology stack to “just about every product going through a leading node.” He stressed that the company is doing all it can to secure more inventory and acknowledged that, in the current environment, that is a difficult task.
Quarterly numbers reveal accelerating infrastructure demand
Dell’s fiscal quarter ended July 31 with revenue of $47 billion, a 58% increase over the same period a year earlier. The Infrastructure Solutions Group, which houses the company’s servers, storage, and networking products, generated $31.8 billion in revenue, an 89% jump and a record for the division. Server-related earnings were particularly strong, rising 122% year over year, and Dell said demand for CPU-based servers is “exceptionally strong” as those systems increasingly support agentic AI workflows.
The order book shows no sign of cooling. Dell booked almost $61 billion in AI server orders in the three months through July 31, and the company recorded more than $130 billion in AI server orders over the prior 12 months. According to Clarke, Dell converted $131.7 billion of customer demand into purchase orders during that period. The sources of that demand are widening, he said, with enterprise customers, neocloud providers, and sovereign cloud operators all competing for capacity.
What agentic AI means for the data center
Agentic AI refers to systems that can reason through complex tasks, plan next steps, call external tools, and operate with less direct supervision than conventional chatbots or copilots. Every one of those steps often requires another inference pass, more data retrieval, and additional model context. That makes agentic workloads far more demanding than earlier AI applications, and it is forcing infrastructure teams to think differently about capacity planning.
Dell’s long-range projections paint a dramatic picture. Clarke said the number of tokens used could reach 3,600 quadrillion by 2030, an 87-fold increase from current levels. Training demand is expected to rise to 850 zettaflops over that period, a fivefold increase. The more consequential shift may come from inference, which Clarke described as “pure demand” in the industry. He predicted that enterprise agentic AI will become the single largest data center workload by 2028 and could account for 75% of all data center demand by 2030.
Traditional servers are back in the growth story
One of the most striking findings in Dell’s report is how much of the recent expansion is tied to traditional CPU-based servers. It has been easy to assume that AI infrastructure is mostly about GPUs and specialized accelerators, but Dell says customers also require “meaningful CPU compute capacity” to support their AI and agentic workflows. In just the last two fiscal quarters, Dell generated nearly as much revenue from traditional servers and networking as it had in any previous full year in company history, according to Clarke.
A large part of that growth is coming from existing customers accelerating their investments in conventional IT environments. Enterprises are refreshing servers to modernize, improve performance, and become more resilient. Security requirements are also adding urgency. Dell expects these refreshes to be both significant and durable, especially as businesses prepare their data centers for workloads that use more cores, more dynamic random-access memory, and more storage capacity.
Dell also emphasized that AI infrastructure is not simply a matter of assembling components and shipping a rack. Modern AI deployments require disaggregated architectures that keep data accessible and in motion across compute, storage, and networking. Clarke said some customer engagements require more than 50 unique server designs because enterprises must optimize for workload performance, power efficiency, cooling, and the physical data center environment. That engineering burden is becoming one more reason why the infrastructure market cannot respond quickly to sudden demand spikes.
The supply chain remains the limiting factor
Even with rising demand, the biggest obstacle for Dell is not customer interest. It is the global component supply chain. Clarke described the pressure in unusually direct terms, saying the constraints begin with DRAM and memory, then extend to NAND flash and a long list of other parts. He summed up the moment as “DRAM, DRAM, DRAM, followed by NAND, NAND, NAND.”
The shortage is not limited to the highest-profile AI components. Dell said there are “spotty” shortages in CPUs and disk drives, and constraints reach down through microcontrollers, drives, transistors, and the broader electronics supply base. The memory shortage is especially painful because new AI servers require more DRAM per system than typical enterprise servers. High-bandwidth memory used in accelerators has also absorbed a large share of memory industry output, making the squeeze even more visible in other parts of the market.
Those shortages limit how many complete AI systems Dell can build in any given quarter. Clarke said large enterprises and multinational corporations around the world would prefer to receive products immediately if supply allowed. “We are supply constrained in the sense of what we can build in any given quarter,” he explained. The company is working through lead times and trying to optimize the configurations it can produce with available components.
Storage is becoming another bottleneck
Agentic AI depends on continuous access to large data sets, which makes storage a strategic concern rather than an afterthought. Dell reported strong growth across its PowerFlex, PowerStore, PowerProtect, and PowerVault product families. The company sees steady demand as organizations modernize storage environments to handle growing volumes of operational data and AI training data.
Clarke said demand remains broad based and that enterprises are increasingly focused on keeping data available and secure while also moving it efficiently across compute and networking resources. AI systems need to read and write data constantly during both training and inference, so storage latency and throughput can quickly become a performance ceiling. The record order backlog suggests that many enterprises have recognized this and are trying to build AI-ready storage before their next wave of model deployments begins.
How customers are responding to higher costs and longer lead times
The shortage-driven environment is also affecting pricing. Clarke noted that modernized configurations with higher core counts, more DRAM, and more storage cost more than they did in previous quarters. Some customers are deferring purchases because they cannot stretch existing budgets enough to cover the rise. Others are placing orders much further in advance just to guarantee access to constrained components.
Large, sophisticated customers are often the first to act, according to Clarke. He said some are collaborating with Dell on longer-range planning so they can map their infrastructure needs well into the future. Clarke described that behavior as “a new phenomenon” in the market and said Dell is working to help customers manage in an environment where demand is running well ahead of supply.
In the meantime, Dell continues to make operational adjustments. The company is optimizing configurations with the components it has and focusing on getting systems out the door as quickly as possible. Clarke said Dell has been able to increase shipments despite the constraints, but the effort requires constant attention to supply allocation, engineering resources, and customer priorities. The $95 billion backlog is likely to remain a central metric for the AI infrastructure market until manufacturing capacity catches up with the orders already on the table.
Source: Network World News