Within just two weeks from late May to early June, two events occurred in the global server market that nearly defined the industry’s trajectory. On May 28, Dell released its Q1 FY2026 report, revealing a staggering 757% year-over-year surge in AI-optimized server revenue, with its stock jumping nearly 33% in a single day. Less than a week later, Hewlett Packard Enterprise followed suit, reporting a 40% year-over-year increase in revenue and a doubling of AI system orders in Q2. Its after-hours trading saw gains of up to 36%, marking the largest single-day stock movement since 2018.
Taken together, these two earnings reports are no longer isolated wins for individual companies—they mark a collective confirmation of the AI infrastructure supercycle. The market once feared that demand for generative AI compute would slow as the marginal returns on model training diminished. However, Dell and HPE’s data signal the opposite: customers are shifting from single-model training to large-scale inference deployments, and enterprise infrastructure upgrades are becoming the new engine for order growth.
The core question has shifted from "Will AI server demand grow?" to "How long will this supercycle last, and who will capture the largest share of profits?"
Both Earnings Reports Point to the Same Signal: AI Infrastructure Spending Enters Historic Expansion
Dell and HPE differ significantly in business scale and customer structure, but their Q2 data reveal the same trend across multiple dimensions.
In Q1 FY2026 (ending early May 2026), Dell achieved total revenue of $43.84 billion, up 88% year-over-year. AI-optimized server revenue reached $16.13 billion, up 757% year-over-year, with new AI orders totaling $24.4 billion for the quarter. Management raised its full-year FY2026 guidance for AI server revenue to around $60 billion. Notably, Dell’s traditional server orders also saw rapid growth. Vice Chairman and COO Jeff Clarke stated clearly that semiconductor companies and large tech firms are leveraging servers to support inference and agent workloads.
HPE’s Q2 FY2026 (ending late April 2026) delivered revenue of $10.7 billion, far exceeding analyst expectations of $9.8 billion. Adjusted EPS came in at $0.79, doubling from $0.38 a year earlier. In AI systems, Q2 brought in $1.8 billion in new orders, with cumulative AI system orders reaching $16.4 billion and backlog at quarter’s end totaling $5.9 billion. CEO Antonio Neri noted that customers are accelerating infrastructure modernization to scale AI workloads, and the company’s current operational progress is about two years ahead of its long-term financial plan.
These companies share two key signals. First, AI-related server demand is no longer limited to a handful of hyperscale CSPs (cloud service providers); it’s spreading to traditional enterprises, semiconductor companies, and national labs. Second, inference-side demand is ramping up faster than expected, directly reflected in the doubling of traditional server orders. Together, these signals confirm that the AI server market is transitioning from a "training arms race" to "inference infrastructure."
From Training to Inference: Structural Shift in Server Demand
To understand the sustainability of the AI server supercycle, it’s essential to distinguish between two workloads: training and inference. Training requires large-scale, high-density GPU clusters, typically deployed by CSPs. Inference, however, is more sensitive to latency, cost, and deployment location, resulting in more diverse hardware choices.
TrendForce’s forecasts clearly illustrate this structural shift: in 2026, high-end AI training machines will account for about 55% of total AI server shipments, but over the medium and long term, growth leadership will shift to inference machines. Based on rack-scale deployments by North America’s top five CSPs, total AI training compute is expected to grow by over 56% annually in 2026, while total AI inference compute is projected to surge by 122% year-over-year.
| Dimension | AI Training Servers | AI Inference Servers |
|---|---|---|
| Main Customers | Hyperscale CSPs | Enterprises, industry clients, CSPs |
| Deployment Models | Large-scale clusters, rack-scale solutions | Hybrid deployments, edge nodes, private cloud |
| Hardware Features | High-density GPUs (e.g., NVIDIA Rubin) | Diverse combinations (GPU, ASIC, traditional CPU) |
| Growth Characteristics | Current main driver (approx. 55% share) | Medium/long-term leader (2026 inference compute growth projected at 122%) |
The explosion in inference demand is driven by two fundamental factors. First, after the large-model arms race of 2024–2025, the market has accumulated enough foundational models, and industries are now exploring real-world applications. From code generation to customer service automation, industrial design to drug discovery, each scenario requires dedicated inference resources. Second, enterprises are reluctant to send sensitive data to public clouds for inference, fueling demand for private deployments and edge nodes—creating an almost entirely new server market.
The trend of CSPs developing their own ASICs is also diversifying inference server types. According to TrendForce, the share of AI servers based on ASICs will rise from about 27.8% in 2026 to nearly 40% by 2030. What does this mean for Dell and HPE? On one hand, they must offer server platforms supporting multiple accelerator architectures. On the other, ASIC solutions typically carry lower unit prices, which may structurally constrain revenue growth.
The Trillion-Dollar Market Battle: Divergent Forecasts and Their Underlying Logic
Forecasts for the total size of the AI server market vary widely among institutions. This divergence isn’t a technical issue—it stems from differing fundamental views on whether AI applications can achieve large-scale commercialization.
Gartner projects global server spending will reach $466 billion in 2026, with AI-related spending accounting for about 76%. Goldman Sachs takes a more aggressive stance, raising its 2026 global server revenue forecast to $652.074 billion, up 45% year-over-year, with AI training server revenue expected to jump 71% to $404.155 billion. AMD offers a longer-term outlook—predicting AI accelerator sales will hit $500 billion by 2028, suggesting the overall AI server market could approach $1 trillion.
| Forecast Dimension | Specific Value | Source |
|---|---|---|
| 2026 Global Server Spending | $466 billion | Gartner |
| 2026 AI Share of Server Spending | 76% | Gartner |
| 2026 Global Server Revenue Forecast | $652.074 billion | Goldman Sachs |
| 2026 AI Server Revenue Growth Rate | 71% | Goldman Sachs |
| 2027 AI Server TAM | $232 billion | Industry Report |
| Top Five CSPs 2026 Combined CapEx | $658 billion | Omdia / Futurum |
The main divergence centers on post-2027 projections. Macquarie analysts believe the risk of an AI infrastructure bubble bursting in 2026 or 2027 is low. However, Goldman Sachs highlights a key risk: many current AI applications are free, and it’s unclear whether users will shift to paid subscriptions. If enterprise AI application ROI falls short of expectations, CSPs may slow capital expenditure growth in 2027–2028.
Another critical data point is the top five CSPs’ combined 2026 CapEx forecast—$658 billion. This figure already exceeds Gartner’s global server spending projection, indicating CSPs are investing heavily in proprietary chip development, data center construction, and network equipment. Servers are just one part of this. If CSP CapEx growth slows, server OEMs will feel the pressure first.
Competitive Landscape Reshuffles: Diverging Paths for Dell, HPE, and SMCI
A major structural event occurred in the AI server market in the first half of 2026—Super Micro Computer, facing governance and financial controversies since 2025, opened a window for market share redistribution. Dell and HPE are seizing this opportunity with different strategies.
Dell’s approach prioritizes scale. Management expects FY2026 AI server orders to reach about $64 billion, with roughly $25 billion already shipped and $43 billion in backlog. Dell’s core strength lies in its global enterprise sales channels and synergy with client devices (PCs, monitors). For mid-sized businesses, purchasing both AI servers and employee workstations from Dell is a natural procurement choice.
HPE, meanwhile, has opted for differentiation. While its AI server shipment volume is lower than Dell’s, HPE possesses two assets rivals can’t easily replicate: the GreenLake consumption-based cloud platform and Cray supercomputing technology. GreenLake allows customers to pay based on actual usage, appealing to budget-sensitive enterprises. Cray technology gives HPE an edge in national lab and supercomputing center bids. CEO Antonio Neri emphasizes that working with enterprise and national lab clients delivers higher margins than competing with "new cloud rivals."
In 2026, the three main suppliers show diverging performance in growth and profitability. Dell’s Q1 FY2026 total revenue rose 88% year-over-year, with AI server growth at 757%. HPE’s Q2 FY2026 revenue increased 40%, and gross margin climbed from 28.4% to 36.5%. SMCI maintained 123% year-over-year revenue growth in Q3 FY2026, but its gross margin faced compression. Stock performance year-to-date also diverged, with Dell leading and SMCI lagging.
A notable trend is that pricing power in the server market is shifting from pure hardware assembly to system integration and value-added services. Both Dell and HPE are strengthening their rack-scale solution delivery, which offers higher added value and customer stickiness than simply selling server nodes.
Supply Chain Constraints Are Becoming an Invisible Lever for Pricing Power
In the short term, the ceiling for AI server market growth is not demand, but supply. Three constraints are simultaneously at play: semiconductor production capacity, memory supply, and data center power infrastructure.
HPE CEO Antonio Neri issued a clear warning: component costs are expected to remain elevated through 2027. The global memory shortage is unlikely to ease soon, and Dell also notes that memory, CPUs, and hard drives will remain in short supply in the second half of the year. This means server OEMs can’t endlessly expand shipments, but it also gives them stronger pricing power—HPE’s gross margin jumped from 28.4% to 36.5%, largely due to tight supply and demand.
Data center power supply is becoming an even more rigid constraint than chips. McKinsey forecasts that by 2027, data center capacity required for AI workloads will exceed available supply, with power limitations worsening further. Approval cycles for data center construction in the US, Europe, and Singapore have noticeably lengthened. For server vendors, this means that even with sufficient production capacity, customers may delay purchases due to rack and power shortages.
Supply constraints impact the industry in two ways. In the short term, they support gross margins and pricing power. In the medium term, if power bottlenecks intensify, they could trigger downward revisions in demand forecasts. Dell and HPE management have mentioned this risk in earnings calls, but have not yet incorporated it into official guidance.
Conclusion
The AI server supercycle has entered its second phase. The first phase (2024–2025) was driven by large-model training demand, with concentrated purchases by a few CSPs and market characteristics of explosive volume but high customer concentration. The second phase (starting in 2026) is driven by inference demand and enterprise deployments, with a broader customer base, diverse hardware solutions, and a shift in competition from shipment volume to system integration and margin management.
Looking ahead, 2026–2027 will remain a period of rapid growth for the AI server market. Dell and HPE’s backlog, CSP capital expenditure plans, and the projected 122% annual growth in inference compute collectively underpin this outlook. After 2028, market growth will gradually converge to the 10–15% range, with clear divergence in profit distribution within the industry—vendors with value-added service capabilities and diversified customer structures will maintain higher valuations, while pure hardware assemblers may face margin pressure.
For investors focused on the AI infrastructure sector, it’s crucial to track three core variables: quarterly CapEx data from the top five CSPs, changes in inference chip and ASIC shipment shares, and progress in data center power approval. These variables will collectively determine the slope and endpoint of the supercycle.
FAQ
What is Dell’s FY2026 AI server revenue target?
Dell management raised its FY2026 AI server revenue guidance to approximately $60 billion.
What was HPE’s AI system backlog at the end of Q2 FY2026?
HPE’s AI system backlog at the end of the quarter was $5.9 billion, with cumulative orders totaling $16.4 billion.
What is the projected growth rate for inference server demand in 2026?
Total AI inference compute for North America’s top five CSPs is expected to grow by about 122% in 2026.
Why are server component costs expected to remain high through 2027?
Global memory shortages and semiconductor production constraints persist, with both HPE and Dell confirming ongoing supply tightness.
When might data center power bottlenecks begin to impact AI server demand?
McKinsey forecasts that by 2027, data center capacity required for AI will exceed available supply, with power constraints worsening further.
What is the projected share of ASIC-based AI server shipments by 2030?
TrendForce projects the share of ASIC-based AI server shipments will rise from about 27.8% in 2026 to nearly 40% by 2030.
What is the combined CapEx forecast for the top five CSPs in 2026?
Omdia / Futurum forecasts combined CapEx for the top five CSPs will reach $658 billion in 2026.
What is HPE’s FY2027 revenue growth guidance?
HPE expects FY2027 revenue growth of 8%–12%.
