Hewlett Packard Enterprise closed Wednesday up 4.5% after disclosing a $1.2 billion AI systems contract with Vultr and lifting its long-term networking revenue outlook by approximately 15% above prior guidance. The dual announcement marks the clearest signal yet that enterprise networking is escaping its decade-long margin compression as AI infrastructure demands purpose-built fabric architecture.
The Vultr contract — HPE's largest single AI systems booking on record — centers on GPU-dense compute clusters and high-speed InfiniBand switching fabric for Vultr's public cloud inference workloads. HPE's networking division, which generated $4.8 billion in fiscal 2024, had been guiding to low-single-digit growth through 2026. The revised outlook now implies $5.5 billion to $5.7 billion by fiscal 2027, driven entirely by AI-adjacent switching and optical transport revenue. CFO Tarek Robbiati told analysts the pipeline for similar deals exceeds $8 billion, with three additional contracts in final legal review.
This matters because HPE is now positioned as the Western alternative to Nvidia's vertically integrated AI stack. Vultr — backed by $400 million from Khosla Ventures and backed by Constant Contact founder Gail Goodman — competes directly with AWS and Azure in inference-optimized cloud. By choosing HPE's Slingshot interconnect over Nvidia's proprietary NVLink fabric, Vultr signals that sovereign compute buyers and hyperscalers outside the Mag Seven are willing to pay for architectural independence. The gross margin on AI switching runs 12 to 18 points higher than commodity enterprise networking, and HPE's backlog conversion cycle is shortening from 18 months to under 12 months as customers prioritize speed to deployment.
The second-order effect is visibility into non-Nvidia AI capex. If Vultr is spending $1.2 billion with HPE, the implied GPU allocation — likely AMD MI300X or Intel Gaudi 3 — suggests $2 billion to $3 billion in total cluster cost. That math only pencils if inference margins hold above 60% at current H100 pricing, meaning Vultr expects enterprise AI workloads to shift from training to production faster than consensus models. It also means Broadcom's Tomahawk and Jericho switching ASICs are now design wins in AI fabrics, not just hyperscale leaf-spine topologies. Watch Arista and Cisco networking guidance in the next 30 days — if they don't raise, HPE is taking share cleanly.
Operators should track three items: HPE's fiscal Q1 2025 earnings on February 25, where management will detail whether the $8 billion pipeline converts at historical 30% rates or accelerates; Vultr's customer wins in the 90 days following this deployment, particularly among European AI labs subject to data residency rules; and AMD's data center GPU revenue guidance on January 28, which will clarify whether non-Nvidia silicon is earning production workloads or remaining pilot-stage.
The market priced HPE like a legacy server vendor until Wednesday. The Vultr contract proves AI infrastructure spend is fragmenting beyond the Mag Seven, and enterprise network architecture is repricing accordingly.