From Earnings to Exascale: Dell, HPE & Nvidia Reshape AI Infrastructure

Table of contents
Modern AI data center with server racks and networking infrastructure

1. Dell FY2026 Full-Year Results: Record $113.5B Revenue, $43B AI Backlog, Eyes $50B in FY2027

Dell Technologies closed fiscal year 2026 (ended January 30, 2026) with record full-year revenue of $113.5 billion, up 17% year-over-year, and delivered a standout Q4 with $33.4 billion in quarterly revenue - 39% above the same period last year and well ahead of the $31.7 billion analyst consensus. The Infrastructure Solutions Group (ISG), which covers servers, networking, and storage, posted $19.6 billion in Q4 alone, up 73% year-over-year.

The headline figure: AI-optimized server revenue reached $9.0 billion in Q4, up 342% year-over-year. For the full fiscal year, Dell shipped more than $25 billion in AI servers and closed $64.1 billion in AI orders, exiting the year with a record AI backlog of $43 billion. Storage revenue was $4.8 billion in Q4, up 2% YoY, with management noting that new PowerStore customers are growing rapidly and demand for unstructured storage is rising in step with AI inference workloads. Looking ahead, Dell guided FY2027 total revenue to approximately $140 billion and AI-optimized server revenue to roughly $50 billion - a further doubling year-over-year. The company also raised its quarterly dividend 20% and added $10 billion to its share repurchase authorization.

2. HPE Emerges as Quiet AI Infrastructure Winner - Stock Up 15.8% YTD as of April 20

Analysis published on April 21 highlights that HPE has been the overlooked infrastructure beneficiary of the AI buildout cycle. As of market close on April 20, HPE stock was up 15.8% year-to-date at $27.81, outperforming the broader tech sector. The driver is unmistakable: HPE's Networking segment - combining Aruba and the newly acquired Juniper - posted 151.5% revenue growth year-over-year in Q1 FY2026 to $2.706 billion. Data Center Networking alone surged 382.6% year-over-year to $444 million, reflecting the rapid ramp of AI-optimized switching and routing demand from hyperscalers and neoclouds.

While Dell's AI server boom has drawn most of the spotlight, HPE's financial profile is shifting in a structurally different way. GAAP gross margin reached 35.9% in Q1 FY2026, up 670 basis points year-over-year, driven by higher-margin networking and software revenue - in contrast to Dell, where AI server volume has compressed gross margins to around 20%. HPE's management guided non-GAAP operating profit growth of 32% to 40% for the full fiscal year, and free cash flow swung from negative $877 million to positive $708 million in a single quarter, with full-year guidance of at least $2 billion.

3. HPE Compute XD700 & Vera Rubin NVL72: Double-Density AI Servers on the Horizon

Announced at Nvidia GTC 2026 in March but now drawing fresh attention as availability timelines firm up, HPE's Compute XD700 is an Open Compute Project-inspired AI server built on Nvidia HGX Rubin NVL8, supporting up to 128 Rubin GPUs per rack - double the density of the previous generation - at reduced space, power, and cooling cost. It targets both AI training and inference workloads and is scheduled for availability in early 2027.

Alongside it, HPE confirmed the Vera Rubin NVL72 by HPE rack-scale system - combining 36 Nvidia Vera CPUs, 72 Rubin GPUs, sixth-generation NVLink, ConnectX-9 SuperNICs, and BlueField-4 DPUs - will be globally available in December 2026, targeting neoclouds and sovereign AI deployments at the frontier model scale (in excess of 1 trillion parameters). Rounding out the portfolio, the HPE Cray Supercomputing GX240 blade brings up to 16 Nvidia Vera CPUs per blade (640 per rack) for leadership-class supercomputing, while Nvidia Quantum-X800 InfiniBand - providing 144 ports at 800 Gb/s per port - has been added to the HPE Cray GX5000 platform for large-scale interconnect. All these systems carry availability dates in late 2026 or 2027, confirming that the Rubin era in production compute is still several quarters away.

4. Dell'Oro: 800G Ports Tripled in 2025, AI-Exposed Switch Vendors Dominate Market Gains

Research firm Dell'Oro Group published its 2025-2026 data center networking outlook, revealing that 800 Gbps port shipments surpassed 20 million units within just three years of first shipment - a milestone that took 400G six to seven years to reach. The explosive pace of 800G adoption reflects the direct pull of AI training and inference clusters, which require ever-wider backplane bandwidth to keep GPU utilization high.

The competitive landscape shifted meaningfully toward vendors with high AI exposure: Accton, Celestica, and Nvidia were among the primary beneficiaries of back-end AI networking demand in 2025, while Arista maintained the leading position in total Ethernet data center switching revenue overall. Looking into 2026, Dell'Oro noted growing skepticism about whether AI infrastructure capex can sustain its current pace, but characterized the industry as still in the early innings of a multi-year investment cycle. High-speed networking components - including optics, advanced substrates, and 800G switches - are also facing rising prices and extended lead times due to supply chain constraints, a factor that enterprise buyers will need to plan for through the rest of the year.

5. Nvidia Targets Space Computing as Next AI Frontier

In a forward-looking announcement on March 19, Nvidia shared plans to extend AI and accelerated computing into space - joining a growing wave of tech companies pursuing orbital and satellite-based computing infrastructure. Nvidia CEO Jensen Huang framed space as the next deployment environment for AI inference, where low-latency processing at the edge of the atmosphere could extend AI capabilities to locations unreachable by terrestrial data centers. The initiative is early-stage, but it signals Nvidia's intent to extend its GPU and networking architecture beyond the data center and into every compute environment on Earth - and above it.

The announcement came in the same week that Huang articulated Nvidia's broader vision for inference at GTC 2026: data centers are no longer storage-and-retrieval facilities but token factories, and every physical environment - from enterprise campuses to satellites - is a potential deployment point for AI inference workloads. For networking hardware vendors, this implies a long-term expansion of the addressable market for ruggedized, low-latency, high-bandwidth switching and compute infrastructure well beyond traditional hyperscale facilities.

Editor's Summary

This week's coverage crystallizes a shift that has been building for months: AI infrastructure has moved from a growth theme into a core financial reality for the industry's biggest hardware vendors. Dell's $43 billion AI backlog and $50 billion FY2027 target make it the most visible beneficiary of enterprise AI deployment at scale. HPE's margin expansion and networking revenue surge show that the Juniper acquisition is already paying dividends in structural profitability. And Dell'Oro's data confirms that 800G Ethernet adoption is running years ahead of historical precedent - a signal that the switch and optics refresh cycle for enterprise and data center buyers is not optional; it is already underway. For IT procurement teams, the clear implication is that planning for 400G-to-800G network upgrades cannot be deferred.

Network-Switch.com is a global professional distributor of networking equipment from Cisco, Huawei, Ruijie, H3C, and our own NSComm brand - including switches, routers, firewalls, wireless APs, optical modules, and fiber patch cables. Our CCIE, HCIE, H3CIE, and RCNP certified engineers provide complete technical support and network solutions for enterprise, campus, and large-scale infrastructure deployments worldwide.

Visit us: https://network-switch.com