AI Servers Are Emerging as the Core Investment Theme for Global Institutional Capital
The back-to-back earnings beats from Dell and HPE are not isolated events—they point to a rapidly forming structural trend: the foundational logic of enterprise AI investment is undergoing a fundamental transformation, shifting from simply purchasing GPUs to building AI factories. At the same time, demand from cloud service providers (CSPs) and sovereign clouds remains red-hot, fueling a new global cycle of AI infrastructure expansion.
AI Server Market: A Quantitative Growth Story
To understand the current position of AI servers within the industry, it’s essential to establish several key quantitative benchmarks.
By 2025, the global server market is projected to reach a total value of $382 billion; by 2026, this figure is expected to grow to $466 billion. The primary driver behind this growth is the ongoing increase in AI-related investments by service providers. According to Gartner, AI-related expenditures will account for 76% of total global server spending by 2026.
Focusing specifically on AI-optimized servers, Gartner’s data shows that global spending on AI-optimized servers will reach approximately $280 billion in 2025, rising to $353 billion in 2026—a year-over-year increase of 26.2%. Over the next five years, the compound annual growth rate (CAGR) is expected to remain at 28.2%.
In terms of shipments, TrendForce’s February 2026 forecast predicts that global server shipments, including AI servers, will see a 12.8% year-over-year increase in 2026. Sigmaintell Consulting offers an even more aggressive outlook—anticipating global AI server shipments will reach around 3.7 million units in 2026, up 51.3% year-over-year, with double-digit growth continuing into 2027 and 2028. While data varies due to differences in statistical methodologies, there is strong consensus on the market’s rapid growth.
Structural Increase in Output Value. Notably, the growth rate of AI server output value is significantly outpacing shipment growth. TrendForce analysis indicates that in 2025, AI server output value will benefit from new Blackwell solutions and high-value integrated AI offerings such as the GB200/GB300 series, with annual growth expected to approach 48%. In 2026, as GPU vendors roll out full-rack solutions and CSPs ramp up investment in ASIC-based AI infrastructure, AI server output value could increase by more than 30% over 2025, with revenue accounting for 74% of the total server market. This means the value per unit continues to rise—full-rack and integrated solutions are reshaping the pricing structure of AI servers.
Dell and HPE: Industry Signals Revealed by the Data
Dell Technologies
Dell’s AI server business has shown a clear acceleration over the past several quarters.
In fiscal year 2025 (ending January 2025), Dell’s quarterly AI server revenue reached $9 billion in the fourth quarter. Moving into fiscal 2026, the company has repeatedly raised its outlook: in August 2025, Dell increased its FY2026 AI server revenue guidance from $15 billion to $20 billion; in November 2025, driven by demand far exceeding supply, the company raised its forecast again to $25 billion. This guidance implies that FY2026 AI server revenue is expected to grow by more than 150% year-over-year.
As of November 2025, Dell’s AI server backlog had reached $18.4 billion, with $12.3 billion in new orders in just the third quarter. In its latest earnings report in May 2026, Dell’s AI server backlog hit a record $51.3 billion—order fulfillment is now lagging well behind the pace of customer demand. Dell has also raised its full-year revenue outlook to between $111.2 billion and $112.2 billion, with non-GAAP earnings per share revised upward to $9.92.
From a customer perspective, Dell’s AI servers serve three main segments: large CSPs, second-tier cloud providers, and enterprise clients. The company revealed that it has closed deals with over 2,000 enterprise customers in the past six quarters and made significant inroads in the sovereign AI market—including clients such as Elon Musk’s xAI, Abu Dhabi’s G42, and the US Department of Energy.
Hewlett Packard Enterprise
HPE has also seen significant performance gains from the AI infrastructure cycle.
According to a Futurum Research report, HPE secured $6.8 billion in new AI system orders during fiscal 2025, with more than 60% coming from sovereign and enterprise customers. In the fourth quarter (ending October 2025), HPE’s AI system revenue hit a record $1.6 billion, up 21% quarter-over-quarter. By the end of fiscal 2025, HPE’s AI system backlog stood at $3.7 billion, up from $3.2 billion at the end of the previous quarter, though major AI deals are expected to contribute significantly to revenue only in the second half of fiscal 2026.
On the financial side, HPE’s Q4 FY2025 total revenue was $9.7 billion, up 14% year-over-year. For the full fiscal year, the server business performed strongly, with total revenue expected to reach about $34.5 billion, up roughly 14.2% year-over-year. In Q3 2025, server business revenue hit a record $4.9 billion, AI system revenue was $1.6 billion, AI orders nearly doubled quarter-over-quarter, and sovereign AI orders surged by approximately 250%.
Data from both companies highlight a common pattern: demand for AI servers is not a short-term spike, but a structural, performance-driven leap fueled by sovereign AI projects, major CSP capital expenditures, and enterprise digital transformation.
AI Factory: A New Industry Paradigm Beyond "GPU"
While the growth of AI servers represents quantitative demand expansion, the rise of the AI factory marks a qualitative shift in supply structures and business models.
The Core Positioning of the AI Factory
The AI factory concept was first introduced by NVIDIA CEO Jensen Huang in 2022 and further clarified during his 2024 GTC keynote as: "An AI factory’s goal in life is to generate revenue, generate intelligence." The core output unit is the token (inference token), not the number of tasks executed as in traditional data centers. In this paradigm, the data center shifts from being a "cost center" to becoming a "digital manufacturing plant for intelligence."
Omdia defines the AI factory as "a new type of heavy industrial infrastructure aimed at producing intelligence," structured around four layers: energy and physical infrastructure, hardware and network interconnect, scheduling and virtualization orchestration, and model-as-a-service with an AI application ecosystem.
Fundamental Differences from Traditional Data Centers
Difference 1: Paradigm Shift in Design Logic.
Traditional data centers are designed around CPU workloads, with a typical rack power capacity of 5 to 15 kilowatts. In contrast, a single NVIDIA GB200 NVL72 rack can draw as much as 140 kilowatts—ten times the power of a conventional rack. This massive difference requires a complete redesign of everything from civil engineering to power distribution and internal networking.
Difference 2: Systemic Surge in Power Demand.
US data center power demand is expected to rise from 31 GW in 2025 to 41 GW in 2026, and to double again to 66 GW by 2027. This surge is largely driven by accelerated AI infrastructure buildout. According to Goldman Sachs Research, the share of summer peak electricity usage by data centers will jump from 4.1% in 2025 to 8.5% in 2027, making power supply a core constraint—no longer just an option—for AI data center construction.
Globally, RAND research estimates that worldwide AI data center power demand could reach 68 GW by 2027, compared to a total global data center capacity of just 88 GW in 2022. In other words, in just five years, incremental AI power demand will nearly match the entire existing capacity. Goldman Sachs projects that global data center power demand will reach 84 GW by 2027, with AI workloads accounting for 27%. While figures vary due to differing methodologies, all research firms agree: AI data center power demand is set for a dramatic leap.
Difference 3: Liquid Cooling Moves from "Optional" to "Mandatory."
As rack power density jumps from the traditional 5–15 kW to over 100 kW, conventional air cooling can no longer meet thermal requirements. Liquid cooling is now becoming the baseline for AI factory construction. This shift explains why cooling vendors like Vertiv are gaining prominence in the AI server supply chain—under the AI factory model, cooling systems have evolved from "auxiliary facilities" to "core infrastructure."
The Full AI Server Industry Chain: From Chips to Infrastructure
Within the AI factory framework, the beneficiaries of the AI server industry chain extend far beyond server OEMs. Here’s a breakdown of the core players across the value chain.
Compute Chip Layer
NVIDIA holds a dominant position. In fiscal 2026 (ending January 2026), NVIDIA’s annual data center revenue reached $215.9 billion, with total annual revenue also at $215.9 billion—a 65% year-over-year increase. In Q4 alone, data center revenue hit $62.3 billion, up 22% quarter-over-quarter and 75% year-over-year. The new Blackwell platform is set to become the mainstream high-end GPU solution for 2025–2026, with B300 and GB300 expected to further drive shipments of Blackwell-based HGX and GB Rack series.
Meanwhile, major CSPs are ramping up their own ASIC development. TrendForce estimates that NVIDIA will hold about 70% of the AI chip market in 2025, but in 2026, North American CSPs and Chinese AI chipmakers are expected to accelerate ASIC adoption, with ASIC shipment growth outpacing GPUs. Google’s TPU and AWS’s Trainium series have become important alternatives for AI inference.
Server OEM Layer
Beyond Dell and HPE, Super Micro Computer derives more than 80% of its sales from AI GPU platforms. Contract manufacturers like Foxconn Industrial Internet, Lenovo Group, and Quanta are also aggressively expanding AI server capacity. Notably, shipments of full-rack AI servers are ramping up rapidly—an estimated 19,000 full-rack AI servers will ship in 2025, rising quickly to 80,000 units by 2027, with the market size expanding to $255 billion.
Network Interconnect Layer
AI training and inference place enormous demands on GPU interconnect bandwidth. Arista Networks is a key player in AI data center switches. According to 650 Group, the AI data center switch market is expected to grow at a five-year CAGR of 36%, reaching about $26 billion by 2029. As AI clusters scale from thousands to tens of thousands or even hundreds of thousands of GPUs, interconnect efficiency becomes more important than single-node compute power.
Cooling and Power Layer
Vertiv is a leader in AI data center thermal management. Eaton’s power infrastructure and Amphenol’s high-speed connectors also play critical roles in AI data center construction.
The Macro Context of AI Capital Expenditure
To understand the structural shift in AI server demand, one must consider the macro backdrop of global AI infrastructure capital expenditure.
From 2024 to 2025, the AI infrastructure spending of tech giants Amazon, Google, and Meta each surged by over 50%. The combined capital expenditures of the four major players jumped from about $256 billion to $427 billion, with all funds concentrated on data center expansion, AI chip procurement, and compute network buildout.
Looking ahead to 2026, the total planned capital expenditure for the four hyperscale cloud providers (Microsoft, Google, Amazon, Meta) has been raised to about $710 billion, a sharp increase from $416 billion in 2025. Amazon CEO Andy Jassy captured the mood on the Q1 earnings call: "We are in a period where demand far exceeds supply. There simply isn’t enough capacity to meet demand." Amazon’s 2026 capital expenditures are expected to reach $200 billion, up about 50% year-over-year.
This sustained expansion in capital spending provides long-term support for downstream AI server demand.
Structural Revaluation Directions Across the Value Chain
From an investment perspective, the value logic of the AI server industry chain is undergoing structural recalibration.
Dimension 1: Structural Diversification of Demand Sources. Compared to the initial wave of GPU procurement, today’s AI server demand is more diversified and sustainable. Sovereign AI projects—where governments treat AI infrastructure as a national strategic asset—offer stable, long-term, high-value, low-price-elasticity revenue streams. CSP capital expenditures are focused on next-generation GPU platform, rack-level products. Enterprise AI deployments represent the largest long-term incremental opportunity.
Dimension 2: Evolution of the Supply Chain Landscape. In the traditional server era, high standardization and limited differentiation led to a stable competitive landscape. In the AI server era, rapidly rising engineering complexity is redrawing competitive boundaries. Dell’s market share is expanding rapidly among second-tier cloud and service providers, thanks to its end-to-end capabilities covering design, integration, deployment, and operations. The "hardware + software + services" model is becoming the key path for AI server vendors to build competitive moats.
Dimension 3: Value Migration Upstream and to "Bottleneck Links." Within the AI factory architecture, key bottleneck segments—rack-level system integration, high-speed networking, high-power cooling solutions, and power infrastructure—are undergoing systemic value revaluation. These segments are characterized by high technical barriers, significant differentiation, strong customer lock-in, and are difficult to replace by any single chip supplier.
Conclusion: The Paradigm Shift from AI Server to AI Factory
The rise of AI servers as the main theme in AI infrastructure investment reflects a comprehensive upgrade in industry thinking. From GPUs to AI servers to AI factories, the focus has shifted from individual compute units to system integration, server engineering, and data center-scale operations—a leap from "point" to "line" to "plane."
The market’s outsized reaction to Dell and HPE’s earnings is, at its core, a delayed repricing of this new understanding. As sovereign AI projects continue to roll out, next-generation GPU platforms are deployed at scale, and enterprise AI applications move from pilot to production, AI servers—as the core physical foundation of AI factories—will see their structural demand base continue to strengthen.
