The Forgotten AI Trade: Why Dell, HPE, IBM & Lenovo Could Win the AI Server Boom

Servers. Boring, old, enterprise servers. Except they are not so boring anymore and they are certainly not old. The global server market just hit a record $444 billion in revenue in 2025, according to IDC. The Q4 of 2025 alone delivered $125.3 billion marking a 52.4% year-over-year surge. And Gartner is projecting server spending to grow another 36.9% in 2026.

Let's break down why AI is not replacing traditional servers but multiplying the need for them and how companies like Dell Technologies, Hewlett Packard Enterprise, IBM, and Lenovo are positioning themselves to capture what could be a durable infrastructure upcycle.

Why AI Is Driving Massive Demand for Enterprise Servers

Here is the misconception doing the rounds: AI will move everything to the cloud, reducing the need for physical enterprise servers. The reality is the opposite.

Think of it this way. When a new highway is built between two cities, it does not just generate traffic between those two points. It increases activity everywhere along the way with petrol pumps, dhabas, motels, logistics hubs, toll gates. AI is that highway. And servers are the entire ecosystem of activity it creates.

There are at least five distinct server demand layers that AI is simultaneously inflating:

1. Training servers: GPU-heavy, Nvidia-centric, and the ones everyone is already talking about. These are used by hyperscalers and AI labs to build models. Large, expensive, loud. Very headline-grabbing.
2. Inference servers: Where models actually do real work for users. Every time you ask a chatbot a question, an inference server responds. This market is growing faster than training. Futurum estimates the global AI inference infrastructure market will grow from $5 billion in 2024 to $48.8 billion by 2030, marking a 46.3% CAGR.
3. Database and storage servers: AI systems need to store, retrieve, and process massive amounts of training data, enterprise data, and output logs. These are largely traditional servers, and AI is massively increasing demand for them.
4. Networking and management servers: Orchestrating AI clusters, managing hybrid cloud infrastructure, running security and compliance tools around AI deployments.
5. Legacy refresh servers: The biggest and most underappreciated layer. Dell's COO Jeff Clarke put it best in a September 2025 earnings call, noting that 70% of the currently installed base of enterprise servers is running on legacy hardware. Those machines are too old and energy-inefficient to support modern AI-adjacent workloads. They need to be replaced.

The investment thesis is not just about GPU farms for LLMs. It is about the full infrastructure rebuild that enterprise AI requires.

The AI Infrastructure Pyramid Explained

Most coverage of AI hardware focuses on the very top of the stack. Here is a way to visualize where the real volume of server demand actually lives:

The top layer gets 90% of the headlines. The bottom four layers make up the bulk of unit volumes and, crucially, are the playing field for Dell, HPE, IBM, and Lenovo. These companies are not just building AI servers, they are building the entire pyramid.

AI Server Market Size and Growth Forecast

A note on the numbers: market size estimates vary across research firms due to different definitions of what counts as a "server." The IDC figures cover the broadest definition and are widely cited in earnings calls by company management, making them the most investor-relevant benchmark. Verify specific figures from primary sources before making investment decisions.

The broader point is directional consensus: every major research house is forecasting sustained growth in server infrastructure spending through at least 2030. 

Best AI Server Infrastructure Companies to Watch

Dell Technologies (NYSE: DELL)

Dell has arguably made the cleanest pivot to the AI server cycle of any traditional hardware company. The numbers are hard to argue with.

In full-year FY26 (ending February 2026), Dell shipped $25.2 billion in AI servers, up over 150% year-over-year, and closed $64.1 billion in total AI orders for the year. Revenue hit $113.5 billion, up 19%, with record earnings per share of $10.30, up 27%.

Most importantly, Dell entered FY27 with a $43 billion AI server backlog.

"AI momentum is accelerating in the second half of the year, leading to record AI server orders of $12.3 billion and an unprecedented $30 billion in orders year to date," Jeff Clarke, Dell's COO, said in November 2025 earnings remarks. For FY27, Dell is guiding AI revenue to roughly double again.

Dell's competitive moat is not just about chips. It is about bespoke engineering, large-cluster deployment at scale, lifecycle support, and its Dell Financial Services (DFS) financing arm, which removes the upfront capex barrier for enterprise buyers. With over 4,000 customers across neoclouds, sovereign governments, and enterprises, and 20% estimated market share in AI-optimized servers, Dell is currently the benchmark name in this space.

The traditional server base is also refreshing. Clarke's observation that 70% of the installed base is still running legacy hardware is not a problem statement, it is a revenue roadmap.

Hewlett Packard Enterprise (NYSE: HPE)

HPE's story in 2025-2026 is one of genuine transformation and honest growing pains.

The headline number: Q3 FY25 revenue grew 18% year-over-year to $9.1 billion, driven by AI server demand. HPE's server segment hit $4.9 billion in that quarter, a 16% jump year-over-year. AI systems alone generated $1.6 billion in a single quarter.

CEO Antonio Neri captured the underlying dynamic well: "Customers are refreshing aged infrastructure with more richly configured servers."

But HPE's story is also a cautionary example of what happens when enterprise AI adoption hits real-world friction. In late 2025, HPE missed revenue targets because some of its larger AI customers had delays in data center readiness as their buildings were not ready to receive the servers they had ordered. CFO Marie Myers acknowledged the lumpiness: "You'll see a lot more of that with AI. It tends to be a lot more lumpy." HPE's AI backlog exceeded $5 billion in Q1 FY26, with enterprise and sovereign customers representing 64% of orders representing a quality mix that should drive higher margins over time.

The company's strategic bet is distinct from Dell's: rather than competing on volume, HPE has leaned into higher-margin networking through the Juniper Networks acquisition, positioning itself at the intersection of AI, cloud, and networking. The goal, as Neri stated at HPE's Securities Analyst Meeting, is to gain share in "networking, cloud, and AI", the three fastest-growing segments in enterprise IT.

International Business Machines (NYSE: IBM)

IBM occupies a different part of the pyramid. It is not an AI server vendor in the Dell or HPE sense, it is the software-and-services layer wrapped around the hardware.

IBM's infrastructure revenue grew 11% in its most recently reported quarter, with Hybrid Infrastructure up 19%. More relevantly, IBM's AI book of business has crossed $5 billion across Consulting and Software. Red Hat, IBM's open-source division, grew 14%, with OpenShift (the enterprise Kubernetes platform that runs AI workloads) hitting $1.7 billion in ARR and growing over 20%.

IBM is playing the enterprise AI stack from the top down: its watsonx platform generates a roughly 5-6x software and consulting multiplier for every dollar spent on it. Basically, for every $1 of watsonx sold, IBM captures $5-6 in additional services. In an environment where enterprises are drowning in AI infrastructure complexity, that consulting and middleware layer could be valuable.

The z17 mainframe cycle also began in 2025, contributing to infrastructure growth and giving IBM a multi-year recurring revenue bump.

Lenovo (HKSE: 992 / ADR: LNVGY)

Lenovo is the dark horse of this group and one of the most interesting setups for investors willing to do the work.

For full-year FY26 (ending March 2026), Lenovo delivered record revenue of $83.1 billion, with adjusted net income growing 42%, twice the rate of revenue. Its Infrastructure Solutions Group (ISG) hit $19.2 billion in annual revenue, turning profitable for a sustained period after a difficult stretch.

In Q4 FY26, ISG revenue grew 37% year-over-year. AI-related revenue reached 38% of total company revenue, up 84% year-over-year. CEO Yuanqing Yang stated plainly: "Through firm execution of our Hybrid AI strategy, we are uniquely positioned to lead in the new wave of AI inferencing and democratization."

Lenovo's structural differentiation is on-premise inference economics. A Lenovo whitepaper from 2026 demonstrates that for high-utilization AI inference workloads, on-premise server infrastructure reaches a breakeven versus cloud costs in under four months, significantly compressed from the 12-18 month cycles of prior hardware generations. As open-source models like Llama and DeepSeek reduce dependence on expensive cloud API calls, the economics of on-premise inference improve further, and Lenovo is positioning itself as the preferred hardware partner for that transition.

Lenovo is also forecasting the AI infrastructure market opportunity to triple by 2028, driven by the shift from training buildouts to inference at scale.

The Picks-and-Shovels Playbook

The "picks and shovels" investment analogy from the gold rush is well-worn but worth revisiting with precision here.

In the 1990s internet boom, Cisco Systems sold the routers and switches that powered the internet. Its stock went from roughly $1 billion in market cap to over $500 billion. Sun Microsystems sold the servers. Investors in both made extraordinary returns, for a while. Then Cisco survived. Sun didn't. The difference: software (Linux, commodity hardware) ate Sun's proprietary server model. Cisco's networking became too embedded to dislodge.

What does that tell us about Dell, HPE, IBM, and Lenovo today?

The bull case is that these companies are not selling proprietary hardware that can be software-ised away. They are selling deployment services, liquid cooling expertise, bespoke cluster engineering, global support infrastructure, and financing, all of which are hard to replicate and increasingly valued by enterprises that lack the engineering depth of hyperscalers.

The bear case, and it is worth taking seriously, is addressed below.

Is The AI Server Boom A Durable Refresh Cycle or A One-Off Spike?

This is the central investor question. The honest answer is: probably durable, but with speed bumps. Here is the structural case for durability:

• The installed base is genuinely old: Dell has publicly stated 70% of enterprise servers are on legacy hardware. That refresh cycle has barely begun in non-hyperscaler enterprise accounts.
• Inference is just getting started: IDC noted in April 2026 that "as one major buyer digests, others ramp up." There is no visible digestion period at the aggregate level.
• On-premise inference is growing: Gartner projected that by 2026, over 50% of enterprise AI inference workloads would run on-premise or at the edge, up from under 10% in 2023. This directly benefits server OEMs.
• Sovereign AI is a new demand vector: Governments across the world are building national AI compute clusters, independent of US hyperscalers. Dell, HPE, and Lenovo all cite sovereign customers as a growing share of their order books.

Key Risks To The AI Server Infrastructure Investment Thesis

Every investment thesis deserves an honest audit. Here are the four scenarios that could derail the server refresh story:

1. Hyperscaler self-sufficiency accelerates. Amazon (Trainium), Google (TPU), Microsoft (Maia), and Meta (MTIA) are all building custom AI silicon in-house. If the world's largest AI compute buyers stop procuring from Dell and HPE in favour of fully internal solutions, the addressable market for OEM server companies shrinks materially. Watch capex disclosures from AWS, Azure, and GCP for signals of this.

2. Model efficiency outpaces hardware demand. The emergence of efficient models, exemplified by DeepSeek's January 2025 release, which delivered near-frontier performance at a fraction of the compute cost, raised genuine questions about whether AI inference is as hardware-hungry as assumed. If model compression and distillation continue to advance rapidly, the inference server buildout may be smaller than projected.

3. Enterprise AI ROI fails to materialise. Enterprise AI adoption is still largely experimental. If a significant share of enterprises fail to achieve measurable productivity gains from AI deployments over the next 12-24 months, capital expenditure on AI infrastructure will be cut. This is the most plausible near-term risk.

4. Margin compression on AI servers. AI servers, with their NVIDIA GPU content, carry lower gross margins than traditional enterprise servers. Dell has acknowledged this mix shift headwind. If AI revenue continues to grow as a share of the total, overall profitability could be diluted even as top-line revenue grows. Watch gross margin trends quarter-over-quarter as a leading indicator of this dynamic.

Putting It Together: A Framework for Investors

What to watch in upcoming quarters:

• Dell's Q1 FY27 AI order momentum and backlog trajectory (guidance was for AI revenue to roughly double in FY27)
• HPE's conversion of its $5B+ AI backlog into recognised revenue; the "lumpiness" metric
• Lenovo ISG margin sustainability as AI revenue scales
• IBM Red Hat / OpenShift ARR growth as the enterprise AI middleware layer thickens

The Bottom Line

Nvidia is the obvious AI hardware trade. It was obvious in 2023, it was obvious in 2024, and its valuation now reflects years of future expectations. The less obvious trade and the one that may have more room to run on a risk-adjusted basis, sits further down the hardware stack. 

Dell, HPE, IBM, and Lenovo are not AI companies in the popular imagination. But they are building and selling the physical infrastructure that every enterprise AI deployment requires, at a scale that no hyperscaler can replicate for corporate clients.