At the end of May and beginning of June 2026, three leading AI server providers—HPE, Dell, and Supermicro—released their latest earnings reports in rapid succession, revealing stark performance divergences. HPE’s quarterly revenue surged 40% year-over-year to $10.7 billion, with its stock price jumping nearly 30% in after-hours trading. Dell’s AI-optimized server revenue soared 757% year-over-year to $16.1 billion, and total AI orders for the quarter reached $24.4 billion. Supermicro posted $10.24 billion in quarterly revenue, but market attention focused more on the ongoing legal investigation facing the company. Viewed together, these reports highlight not only differences in operational capabilities, but also signal a structural shift in AI infrastructure investment logic—from the supply-driven phase of "who can secure GPUs" to a demand and operations-driven stage centered on "who can efficiently deliver and cover inference scenarios." For investors, understanding each company’s positioning as AI transitions from training to inference is far more critical than simply comparing current revenue growth rates.
Three Growth Structures Behind the Earnings Numbers
Dell delivered a breakout performance in the first quarter of fiscal year 2027 (ending early May 2026). AI-optimized server revenue hit $16.1 billion, up 757% year-over-year and 80% quarter-over-quarter, already surpassing its total AI shipments for all of fiscal 2025 ($9.8 billion). Even more notable are the order figures: total AI-related orders for the quarter reached $24.4 billion, with the customer base exceeding 5,000—up more than 50% from six months prior—spanning cloud service providers, sovereign entities, and traditional enterprises. By quarter’s end, Dell’s AI server backlog reached a record $51.3 billion, and the company disclosed that its pipeline projects are worth several times that amount. This backlog size provides Dell with exceptional revenue visibility over the next four to six quarters, but it also raises a hidden concern: whether Dell’s delivery capacity can keep pace with order growth will directly determine the speed of revenue recognition.
HPE’s financial performance follows a different logic. In the second quarter of fiscal 2026 (ending late April 2026), HPE’s total revenue reached $10.7 billion, up 40% year-over-year, with net income swinging from a $1.05 billion loss in the prior year to a $624 million profit. Server business revenue was $5.45 billion, up 32.7% year-over-year and well above analyst expectations of $4.66 billion. Unlike Dell, HPE’s AI system backlog stood at $5.9 billion, with 61% coming from government and large enterprise customers. This customer mix means HPE’s order profit margins are typically higher than bulk orders for cloud service providers, but order growth is more gradual. HPE raised its full-year AI networking business growth forecast from 68–73% to 72–75%, with integration benefits from the Juniper Networks acquisition beginning to show.
Supermicro, in the third quarter of fiscal 2026 (ending late March 2026), reported $10.24 billion in revenue and adjusted earnings per share of $0.84. The company expects fourth-quarter revenue between $11 billion and $12.5 billion. Its core competitive advantage lies in direct liquid cooling technology—as single-rack power density in data centers jumps to over 240kW, liquid cooling is shifting from an "alternative" to a "must-have," and Supermicro controls about 70% of this niche market. However, the legal investigation remains unresolved, prompting institutional investors to approach Supermicro with caution and resulting in a clear valuation discount compared to Dell and HPE.
Comparing the three companies reveals distinct growth engines: Dell wins through scale and deep ties with CSP customers; HPE builds a moat with high retention among enterprise and government clients; Supermicro achieves localized dominance through differentiated liquid cooling technology. There’s no absolute winner among these models, but as AI inference demand is set to explode, their flexibility and risk exposure will diverge sharply.
Three Key Controversies in the Market: Training Plateau, Liquid Cooling Premium, and Backlog Volatility
Within the AI server sector, three unresolved controversies directly shape how investors price these companies.
The first controversy is whether AI training demand is nearing its peak. A common concern is that as large models shift from "stacking compute" to "optimizing efficiency," growth in training servers will slow significantly. TrendForce research points in the opposite direction: AI server shipments in 2026 are expected to grow over 28% annually, with high-end AI training models still accounting for about 55% of shipments. Over the medium and long term, growth will gradually shift to AI inference models, characterized by more diverse application scenarios, lower per-instance compute needs but higher concurrency, and a customer base expanding from large CSPs to small and medium enterprises and industry clients. This shift demands new capabilities from server OEMs—not just stuffing GPUs into chassis, but optimizing system architecture, power management, and cost structure for different inference scenarios. Dell’s base of over 5,000 customers provides a natural advantage in this transition, HPE’s enterprise channels align well with traditional industry inference deployments, and Supermicro’s liquid cooling solutions will add value in high-density inference clusters.
The second controversy is how to price Supermicro’s "governance discount." Supermicro’s valuation is clearly discounted relative to Dell and HPE, justified by legal uncertainty, deep dependency on CSP customers, and liquid cooling’s unproven substitutability. Opponents of the discount argue that liquid cooling is an irreplaceable requirement amid the densification of AI data centers, and once legal risks are resolved, the upside for valuation is substantial. This is a classic risk-event pricing issue—differences in investor judgment about the probability and impact of the legal probe drive divergent portfolio strategies. Most hedge funds express caution by reducing positions or hedging with options.
The third controversy is whether HPE’s quarter-over-quarter decline in backlog signals danger. HPE’s AI system backlog at the end of Q2 FY2026 was $5.9 billion, down from $6.3 billion in Q1—a 6.3% decrease. Interpretations split into two opposing logics: optimists see backlog decline as a sign of improved delivery, with backlog quickly converting to revenue; cautious observers worry that new order growth may be slowing, especially as Dell and Supermicro compete for limited GPU allocation. HPE itself raised its full-year AI networking business growth forecast sharply, suggesting strong demand for network equipment. At least two more quarters of order data are needed to determine whether backlog changes reflect delivery cadence or marginal demand shifts.
Revising the "Big Three" Narrative: Market Concentration Is Far Lower Than Media Portrays
Mainstream media often describes Dell, HPE, and Supermicro as the "Big Three" of AI servers, implying they dominate the entire market. This narrative diverges sharply from reality. Dell holds about 20% of the global AI server market, HPE about 15%, and Supermicro about 9%—combined, less than half. Over 50% of market share belongs to vendors such as Huawei, Lenovo, Cisco, and CSPs’ self-developed ASIC servers and original design manufacturers. In 2026, ASIC-based AI servers are projected to account for 27.8% of the market, the highest since 2023. This means the AI server market is highly fragmented and competitive, not monopolized by a handful of giants.
Another narrative needing scrutiny is "AI servers equal NVIDIA GPUs." While NVIDIA’s Blackwell and Rubin platforms remain mainstream for high-end training, AMD’s MI325X, built on fifth-generation Epyc CPUs, entered a phase of intensive release in May 2026, with HPE, Dell, and Supermicro all launching server product lines featuring AMD GPUs. Domestic solutions such as Huawei’s Atlas 950 also hold significant market share in certain regions. Diversification in GPU supply is weakening the logic that "whoever gets more NVIDIA allocation wins," and elevating the importance of system integration, cooling solutions, customer channels, and global delivery networks.
Understanding the true market structure helps avoid overly optimistic or pessimistic linear projections about these three companies. Dell’s scale advantage in the CSP market is real, but it faces fierce competition from Huawei, Inspur, and others in Asia-Pacific. HPE has deep enterprise roots, but enterprise AI deployment typically lags behind cloud providers, resulting in a smoother but slower growth curve. Supermicro leads in liquid cooling, but its overall market share falls short of justifying a "giant" label.
Four Layers of AI Infrastructure Investment Transmission
The surge in AI server demand is not a single-stage boom, but generates cascading effects across the entire data center investment chain. Understanding this chain is key to assessing each server OEM’s prospects.
The first layer transmits from GPU suppliers to server OEMs. When CSPs and large enterprises ramp up compute purchases, revenue first flows to NVIDIA and AMD. But GPU supply is limited, and servers must be integrated by OEMs before final delivery. This explains why Dell and HPE’s backlog far exceeds quarterly revenue—they secure GPU allocations and then schedule production and delivery. The competitive focus here is "allocation acquisition capability," favoring companies with the closest ties to GPU suppliers.
The second layer is shifting from AI training to AI inference. Currently, about 55% of shipments are high-end training models, but the inflection point for inference demand has arrived. Inference servers no longer simply "stack GPUs," but require low-latency, high-concurrency architecture optimizations, demanding new system design capabilities from OEMs. Companies able to offer customized solutions for different inference scenarios—such as real-time dialogue, image generation, or recommendation systems—will capture higher value.
The third layer centers on data center power management and cooling. Single-rack power density is rising rapidly, with mainstream data centers in 2026 reaching 240kW, making liquid cooling a necessity. Supermicro’s early lead in liquid cooling is its key structural moat, but Dell and HPE are catching up quickly by partnering with cooling solution providers or developing their own technologies.
The fourth layer shifts from hardware investment to operational expense optimization. As AI infrastructure moves from build-out to operation, hardware procurement decisions focus less on "acquiring compute" and more on "total cost of ownership per unit compute." This drives demand for higher efficiency and lower power solutions, and shifts OEM pricing power from "can you supply" to "can you help customers save money."
These four layers do not unfold sequentially, but overlap on the same timeline. The market is currently in a complex phase: the first layer is ongoing, the second just starting, the third accelerating, and the fourth beginning to enter decision frameworks. Each OEM’s position in these layers determines its growth flexibility over the next 12–18 months.
Three Scenario Projections: Variables of Inference Demand, GPU Supply, and Market Growth
Based on current market data, technology trends, and policy environment, three main evolutionary paths emerge for the competitive landscape among these companies over the next 12–18 months.
Scenario 1: High growth driven by AI inference demand. If CSPs commercialize AI applications faster than expected and inference compute demand accelerates, Dell—with its deep CSP ties and coverage of over 5,000 customers—is poised to benefit most. With $24.4 billion in new orders and $51.3 billion in backlog in a single quarter, its annual AI server revenue guidance is now raised to about $60 billion. HPE benefits from a diversified customer base; since enterprise AI inference deployment typically lags CSPs by 6–12 months, its growth may be smoother but more stable. If Supermicro’s legal probe does not materially impact its business, its liquid cooling technology will translate into pricing power in high-density inference clusters, with the main risk being whether delivery capacity can match growth expectations.
Scenario 2: Persistent GPU supply constraints shift competition toward allocation acquisition. If NVIDIA and AMD’s GPU production remains insufficient to meet downstream demand, the focus shifts from "sales capability" to "allocation acquisition capability." Companies with the closest relationships to GPU suppliers gain the greatest advantage. Dell’s long-term partnerships and large procurement volumes give it an edge, and Supermicro’s deep collaboration with NVIDIA on liquid cooling also provides strong allocation access. HPE, relying on third-party GPU supply, is at a relative disadvantage for allocations, but its sticky enterprise customers partly offset this weakness.
Scenario 3: Market growth falls short of expectations, intensifying competition and pressuring margins. If market growth slows, competition turns to price and service. All three companies may face margin pressure. Dell’s scale enables the strongest cost-sharing in price wars. HPE’s enterprise focus and GreenLake pay-as-you-go model provide some pricing moat. Supermicro’s high customization means its margins are most sensitive to price fluctuations, making it the most vulnerable. If the legal probe continues during a slowdown, the market may accelerate shifting order share to Dell and HPE.
Each scenario aligns with different macro assumptions. TrendForce’s projection of over 28% AI server shipment growth in 2026 supports scenario one most strongly, but CSP capital expenditure decisions, GPU production ramp-up, and geopolitical trade policies could alter the trajectory.
Conclusion
The core judgment for today’s AI server market is this: the sector is shifting from a "training-driven, supply-constrained" first half to an "inference-driven, operational efficiency competition" second half. Dell leads in scale and CSP customer coverage, but must prove its competitiveness in high-margin enterprise markets and inference scenarios. HPE builds differentiated barriers through stability among enterprise and government clients, but its order growth is more gradual, and the quarter-over-quarter backlog trend warrants attention. Supermicro holds a structural advantage in liquid cooling technology, but legal uncertainty keeps its valuation discounted; the next six months of legal developments will be a key variable.
In the medium term, demand for AI inference servers will become the primary growth driver, with inference models expected to surpass training models in shipments by 2027. Suppliers able to deliver high-density liquid cooling, multi-GPU platform compatibility, and system architectures optimized for inference scenarios will gain stronger pricing power.
For investors, the next 12 months require close attention to the following signals: Dell’s AI server backlog conversion rate and quarter-over-quarter new order changes; whether HPE’s full-year 72–75% networking business growth guidance is realized; progress and interim conclusions of Supermicro’s legal probe; quarterly changes in inference server share of CSP capital expenditures; and the penetration rate of AMD MI325X and subsequent GPU solutions in OEM product lines. These signals will help determine which scenario is becoming reality.
FAQ
How much did HPE’s AI server business grow in the latest earnings report?
HPE’s server revenue in Q2 FY2026 grew 32.7% year-over-year to $5.45 billion, with AI system backlog at $5.9 billion.
Why did Dell’s AI server revenue grow 757% year-over-year?
Dell’s AI-optimized server revenue reached $16.1 billion in Q1 FY2027, primarily driven by large-scale purchases from cloud service provider customers and a customer base exceeding 5,000.
How strong is Supermicro’s liquid cooling technology advantage?
Supermicro holds about 70% market share in data center liquid cooling solutions, and direct liquid cooling has become the standard for high-density AI clusters.
Has demand for AI training servers peaked?
TrendForce estimates AI server shipments will grow over 28% annually in 2026, with training models still accounting for about 55% of shipments, though inference models are growing faster.
Does HPE’s quarter-over-quarter backlog decline signal slowing demand?
HPE’s AI system backlog fell from $6.3 billion to $5.9 billion, which could reflect accelerated delivery or slower new order growth; further quarterly data is needed for confirmation.
Which company has the strongest position in the AI inference server market?
Dell leads in scale and channels with over 5,000 customers and $51.3 billion in backlog, but HPE’s sticky enterprise customer base also offers long-term value.
Will Supermicro’s legal investigation affect its stock price?
Uncertainty from the legal probe has led to a valuation discount for Supermicro relative to Dell and HPE; progress and conclusions from the investigation will be key factors affecting share price.
What is the true market concentration in AI servers?
Dell, HPE, and Supermicro together account for less than 50% of market share; Huawei, Lenovo, Cisco, and CSP self-developed ASIC servers make up the other half.
