Hewlett Packard Enterprise closed its third fiscal quarter of 2026 on July 31 with $12.2 billion in revenue, up 34% year-over-year. Two days after the books closed, Antonio Neri's company booked a $3.5 billion inferencing-server order from a hyperscaler — a single deal that now shapes the next twelve months of guidance.

"We delivered record revenue, orders, and profitability, fueled by surging customer demand across our portfolio," Neri, HPE's president and CEO, said on the earnings call. "AI is becoming a multi-year growth driver for HPE, and our differentiated portfolio positions us to capture that opportunity at scale."

What did the quarter actually deliver?

GAAP net income hit $1.5 billion, up 405% from a year ago. Diluted EPS set a record at $1.06, against $0.21 in the year-ago quarter. Free cash flow reached $1.0 billion, also a record, versus $790 million a year earlier. Gross margin climbed to 40.1%, from 29.9%.

Cash and equivalents closed at $6.2 billion. Total net debt dropped to $14.0 billion from $16.0 billion the prior quarter, helped by $1.4 billion in proceeds from exiting the China-based H3C joint venture.

Where did the growth come from?

Two segments carry HPE's P&L: Intelligent Edge (networking) and Cloud + AI.

Networking revenue hit $2.9 billion, up 74.9% Y/Y on Juniper Networks consolidation. Orders climbed 36% to $700 million. Inside networking:

• Campus and Branch: $1.4 billion, up 31% Y/Y, a record.
• Data Center Networking: $382 million, down 6% Y/Y.
• Security: $281 million, up 12% Y/Y.
• Routing: $788 million, up 23% Y/Y.

Neri flagged a supply ceiling. "Orders are growing 3.5 times faster than the revenue," he said. "What has limited us is the availability of supply. We expect that supply to become more aligned to our order bookings as we go forward, starting Q4."

Cloud + AI pulled in $9.0 billion, up 25.4% Y/Y and a record. Server revenue did the heavy lifting:

• Server: $6.8 billion, up 35.3% Y/Y.
• Storage: $1.3 billion, up 10.2% Y/Y.
• Financial Services: $0.9 billion, down 0.3% Y/Y.

CFO Marie Myers pointed at pricing power. "Higher average selling prices drove server revenue to an all-time high," she said. The average server now sells for more because customers want AI-ready systems, not commodity boxes.

Why is supply the bottleneck?

Memory and NAND remain tight. Neri said HPE is "collaborating very closely with our partners to secure additional multi-year supply agreements." That phrasing — multi-year — signals HPE is locking in allocation the same way DRAM and HBM buyers do, with long-term wafer commitments in exchange for guaranteed starts.

What about storage and Oracle?

HPE's storage line passed Pure Storage in quarterly revenue this period, trailing only NetApp and Dell. Alletra MP orders and revenue climbed strong double digits, Myers said, with the X10K object-and-file platform positioned for unstructured AI data.

Oracle is buying more Juniper routing and switching to knit its AI data centers together. The scope covers PTX and MX routing platforms and QFX and EX switches across Oracle Cloud Infrastructure globally, plus multi-year support services. The latest QFX silicon targets high-density RoCEv2 (RDMA over Converged Ethernet v2) backend networks for AI clusters. HPE and Oracle are also co-developing telemetry for fabric visibility.

Where does GreenLake and internal AI fit?

GreenLake customer count hit 52,000, up 18% from 44,000 a year ago. Inside HPE, an internal agentic AI platform built on the company's Private Cloud AI stack routes routine tasks on-prem and reserves frontier models for harder work. HPE's internal analysis claims Private Cloud AI cuts token cost by up to 60% versus public cloud.

What's the outlook?

Myers raised the bar:

• Q4 revenue: $14.4 billion ± $450 million, a 48.8% midpoint rise Y/Y.
• Full FY2026: lifted to $46.5 billion ± $515 million, from $44.6 billion ± $186 million.
• FY2027 growth: now 13–17%, up from the prior 8–12%.

"The market is telling us that demand continues to be exceptionally strong," Neri said.

The $3.5 billion hyperscaler order, the supply-constrained networking backlog, and the raised FY2027 growth range all rest on one bet: that inferencing and agentic workloads keep ordering more silicon than the supply chain can deliver.