The narrative in the AI hardware market is undergoing a clear structural reversal. Over the past two years, the spotlight has been on expanding GPU clusters for training, with dominance determined by who owns the most chips. However, HPE’s Q2 earnings report released on June 1, 2026, signals a shift: demand for AI inference is converting into real orders for server manufacturers at an unexpectedly rapid pace. HPE’s total revenue for the quarter reached $10.7 billion, up 40% year-over-year, with $1.8 billion in new AI system orders and traditional server orders more than doubling compared to last year. More importantly, management directly attributed this growth to customers "upgrading computing infrastructure and investing in AI inference."
The significance of this shift lies in the fact that server companies—not chip companies—are entering the phase of explosive growth first. The large-scale deployment of AI inference workloads is spreading hardware demand from a handful of major model training firms to thousands of enterprises. System integration capabilities, delivery networks, and compliance qualifications of server manufacturers are becoming more critical competitive factors than individual chip performance. This trend is not only reshaping the supply and demand landscape for AI infrastructure but also provides a key logical anchor for understanding the hardware investment cycle in 2026-2027.
HPE Q2 Earnings Beat Expectations: Hardware Upgrades Driven by Inference
On June 1, 2026 (UTC), HPE released its financial results for fiscal Q2 2026. Core financial highlights include: revenue of $10.7 billion, up 40% year-over-year and far exceeding market expectations of $9.76 billion; non-GAAP EPS of $0.79, up 108% year-over-year, also beating forecasts. Server business revenue was $5.5 billion, up 32.7%. New AI system orders totaled $1.8 billion, bringing cumulative AI system orders to $16.4 billion, with $5.9 billion in AI system backlog at quarter’s end.
Traditional servers delivered particularly notable results. Orders for traditional servers more than doubled year-over-year. HPE CFO Marie Myers stated during the earnings call that the direct driver of this growth was "customers upgrading computing infrastructure to support AI inference." This breaks the common assumption that "AI will replace traditional servers"—in reality, inference workloads at this stage are creating incremental and upgrade demand for existing computing infrastructure.
Looking back, the main storyline for AI infrastructure in 2023-2024 was large model training, with buyers highly concentrated among cloud service providers and a few AI labs. In the second half of 2025, widespread deployment of Llama models, enterprise AI tools like Copilot, and the initial rollout of AI Agents accelerated inference workloads. Deloitte’s November 2025 report noted that inference workloads accounted for half of all AI computing, projected to rise to two-thirds in 2026. Futurum’s May 2026 report confirmed that managed inference accounted for 58.6% of AI infrastructure spending in 2025, whereas training made up about three-quarters in 2022. HPE’s order data now provides clear financial validation of this structural reversal.
Doubling Orders: Supply-Demand Gap and Customer Expansion
Key HPE Q2 2026 Financial Data
| Metric | Value | YoY Change |
|---|---|---|
| Total Revenue | $10.7B | +40% |
| Non-GAAP EPS | $0.79 | +108% |
| Server Revenue | $5.5B | +32.7% |
| New AI System Orders | $1.8B | ~+100% |
| AI System Backlog | $5.9B | — |
| Networking Revenue | $2.7B | +148% |
Two structural shifts emerge from this data.
Inference demand is driving both traditional servers and AI-specific systems. The doubling of traditional server orders is a signal underestimated by most market commentators. NVIDIA CEO Jensen Huang raised AI demand forecasts to $1 trillion at GTC 2026, noting that inference workloads run not only on dedicated AI server racks but also on general-purpose servers for preprocessing, postprocessing, and storage. This indicates that AI inference is penetrating enterprise computing infrastructure far more deeply than the training phase.
Customer base is expanding from a few giants to a broad range of enterprises. HPE management reported that 64% of AI system orders came from enterprise and sovereign customers, not large cloud providers. Dell disclosed that its AI customer count surpassed 5,000, growing over 50% in six months. This expansion means AI inference demand is moving from labs to real enterprise production environments, elevating the value of server vendors’ channel networks and service capabilities.
Comparing HPE and Dell’s AI server businesses helps clarify industry dynamics. Dell’s late-May 2026 data shows annual AI server revenue at about $25.2 billion, up over 150% year-over-year, with $24.4 billion in AI orders and a backlog of $51.3 billion in the quarter. Their strategies diverge: HPE focuses on delivering AI compute as a service via the GreenLake platform, excelling in sovereign AI and hybrid cloud scenarios; Dell leverages the world’s largest enterprise sales channel and a scaled service network to lead in AI server shipments.
Market Perspectives and Narrative Validation: Two Interpretations of Order Backlogs
Discussions around HPE’s Q2 results and the AI server market fall into three main camps, each with its own logic and risk boundaries.
The shift from training-driven to inference-driven demand is the fundamental driver of server market expansion. This view is the strongest consensus. Lenovo’s management at CES 2026 stated, "80% of AI compute will be used for inference, 20% for training." TrendForce data supports this, projecting global server shipments to rise 12.8% in 2026, with AI server shipments up over 28%. However, inference workloads (edge, endpoint, enterprise local) differ fundamentally from traditional training workloads (centralized data centers), requiring server vendors to adjust their product portfolios—not all will benefit equally.
Current supply-demand gaps suggest server vendors are in the early phase of sustained volume growth. The main evidence is the scale of order backlogs. Dell received $24.4 billion in AI orders in the quarter but recognized only $16.1 billion in revenue, leaving a $51.3 billion backlog. HPE similarly consumed much of its backlog but still had $5.9 billion in AI system backlog, with management stating "pipeline capacity far exceeds current backlog." However, order backlogs can be interpreted two ways: they may signal strong demand, or they could reflect supply chain bottlenecks (especially GPU supply) causing delivery delays. The latter cannot be ruled out. If supply constraints ease in the second half of 2026, backlog digestion may accelerate, and the sustainability of revenue growth will need reevaluation.
Market valuations for server vendors may already reflect overly optimistic expectations. Days before HPE’s earnings, Dell’s positive report drove HPE’s stock up 12.76% in a single day, with trading volume reaching 66.7 million shares—about 260% of the three-month daily average. This advance pricing means that even with a strong Q2 report, HPE’s further upside may be limited by already priced-in information—a risk variable to consider.
Three common narrative biases warrant scrutiny. First, the claim that "AI will replace traditional servers" is disproven—traditional server orders doubled, directly driven by inference demand. Second, the notion that "AI is a game for a few tech giants" is challenged by customer growth, but 5,000 customers is still limited, and most may be in pilot stages. Deloitte’s report shows that as of early 2026, only 25% of surveyed enterprises had moved more than 40% of AI experiments into production. Third, the belief that "HPE and Dell’s AI growth is entirely driven by GPU supply" is changing; TrendForce forecasts that ASIC-based AI servers will account for 27.8% of shipments in 2026.
Industry Impact Analysis: Supply Chain Redistribution, As-a-Service Models, and Network Upgrades
The rise in AI inference demand is structurally impacting the server industry in at least three ways.
Supply chain redistribution. After Supermicro personnel were indicted by the US Department of Justice (March 2026), large enterprises and sovereign AI projects are placing greater emphasis on compliance risks when evaluating suppliers. This compliance factor, combined with Dell and HPE’s simultaneous launch of next-generation servers based on the Vera Rubin platform at GTC 2026, has opened a key window for reassessing AI server supply chain allocations. For the market, this means top OEMs have more leverage in share battles, and compliance capabilities are becoming an explicit competitive advantage.
Amplification of as-a-service value. HPE’s GreenLake platform managed about 6.7 million systems in Q2, serving roughly 15,000 customers, with managed system count up about 26% year-over-year. For enterprises seeking a cloud-like experience in their own data centers, as-a-service models are a strong fit for AI inference deployment. The core advantage is reducing upfront capital expenditure for AI infrastructure while maintaining compliance with data localization requirements. As sovereign AI demand rises, GreenLake-type services may become HPE’s key differentiator from other server vendors.
Synchronous upgrade demand for network infrastructure. HPE’s networking revenue grew 148% year-over-year in the quarter, with network order growth far outpacing revenue growth. Management highlighted "network as an AI-first business" as a strategic priority. Inference applications’ need for low latency and high bandwidth is driving enterprises to upgrade data center networks, accelerating the transition from 100G to 400G/800G. This is an incremental market beyond traditional server upgrades, and network equipment typically has a shorter replacement cycle than servers, potentially providing more sustained revenue contributions.
Conclusion
The core takeaway is this: AI inference demand is becoming an independent growth engine for the server hardware market, fundamentally different from the training phase. The distributed, large-scale nature of inference workloads makes system integration, channel networks, and compliance qualifications of server vendors more scarce and valuable than chip compute power. HPE’s Q2 report—doubling traditional server orders, high AI order backlog, and expanding customer base—offers early financial evidence of this structural shift.
In the medium term, the second half of 2026 through 2027 will be a critical window to validate the sustainability of inference demand. Key metrics to track include: the conversion rate of enterprise AI deployments from pilot to production, changes in ASIC server shipment share, and the speed at which leading server vendors digest backlog orders. If conversion rates keep rising and supply bottlenecks ease, the server industry could enter a stable 2-3 year volume expansion cycle.
Investors should shift focus from "is there demand for AI servers?" to "which server vendors have differentiated advantages in customer structure, geographic reach, and product offerings?" Compliance capabilities (especially for sovereign AI projects), penetration of as-a-service models, and synergistic growth in networking may be critical variables distinguishing long-term value among vendors.
FAQ
Why does AI inference demand boost server company performance ahead of training demand?
AI inference is highly distributed, requiring large numbers of traditional and specialized servers to be deployed locally across thousands of enterprises. Training is concentrated among a few cloud providers, so server vendors’ channel networks and delivery capabilities are more valuable during the inference phase.
What distinguishes HPE and Dell in the AI server market?
HPE emphasizes the GreenLake as-a-service model and sovereign AI scenarios, while Dell relies on the world’s largest enterprise sales channel and scaled service network to lead in shipment volume.
What’s the logic behind doubling traditional server orders?
AI inference workloads—preprocessing, postprocessing, storage, and lightweight inference tasks—run extensively on general-purpose servers. Enterprises need to upgrade existing infrastructure rather than fully replace it.
Does the high backlog of AI server orders indicate sustainable demand?
Backlogs reflect both strong demand and GPU supply chain bottlenecks. Sustainability depends on new order growth after supply constraints ease in the second half of 2026.
How does growing inference server demand affect crypto hardware?
Rising data center energy consumption and compute competition may indirectly impact crypto mining hardware costs and power allocation, but there’s currently no direct substitution.
How will rising ASIC server share change the competitive landscape?
ASIC solutions are expected to reach 27.8% of shipments in 2026, potentially reducing reliance on single GPU suppliers and giving server vendors more room for cost control and product differentiation.
Under what circumstances could the server industry’s growth cycle end early?
If enterprise AI ROI falls short of expectations or macroeconomic downturns lead to IT spending cuts, server order growth could slow in 2027.
Which server vendors have long-term advantages in the inference phase?
Vendors with global service networks, strong compliance (especially sovereign AI credentials), and robust as-a-service product portfolios are best positioned to maintain pricing power as inference demand spreads.
