How does HPE, this old-school manufacturer, get to ride the AI wave?

$Hewlett Packard Enterprise(HPE.US) closed at $32.07, up 6.16% for the day. This isn't a huge wave—but after-hours, it pulled up to $33.85, another 5.5% gain. Then I saw in the news that activist investor Irenic Capital joined.

This company is Hewlett Packard Enterprise—HPE. It sounds like a name from a bygone era, why would anyone suddenly want to buy it in the AI cycle?

I think this is a second-tier AI story, not the pure leverage kind like NVIDIA, but "enterprise AI implementation." Below, I'll explain in plain language which semiconductor company it's connected to and which part of the AI cycle it's on.

First, what exactly does HPE do?

It doesn't make chips. It does the job of "putting chips into racks and selling them to enterprises." Three main businesses:

First, servers. HPE's ProLiant, Cray AI supercomputing systems, HGX integrated machines—these involve soldering NVIDIA H100/H200/B200 GPUs or AMD MI300X GPUs onto motherboards, connecting them into racks, configuring networking and liquid cooling, and selling the whole set to your company. Dell and SuperMicro are its peers, competing for this business.

Second, storage and networking. HPE has its own storage product lines (3PAR, Nimble, Alletra). In networking, it just swallowed Juniper Networks last year—I think this acquisition is HPE's most critical move in the past two years, discussed separately below.

Third, GreenLake—HPE's "IT resource rental" service. Customers who don't want to buy servers can rent on-demand, paying monthly. This is HPE's attempt to mimic AWS's subscription model, which would have higher gross margins, but the scale is still small.

Second, which semiconductor vendor is it connected to?

Direct answer: It has connections with both NVIDIA and AMD, but the relationship with NVIDIA is the most critical.

NVIDIA's HGX platform (the thing that puts 8 GPUs on a baseboard with NVLink high-speed interconnects) requires OEM integrators for the last mile. HPE is one of the few NVIDIA-certified OEMs—others include Dell, SuperMicro, Lenovo, etc. This means a significant portion of enterprise customers wanting NVIDIA AI servers will go through HPE. This business is growing now—on 5/11, HPE launched a 64TB memory SAP Cloud ERP dedicated server, prepared for large model inference / enterprise database AI workloads.

On the AMD side, HPE Cray EX supercomputers use AMD MI300X GPUs + EPYC CPUs. El Capitan (the supercomputer at Lawrence Livermore National Laboratory) was provided by HPE—this was AMD AI's first billion-dollar order.

Third, which part of the AI cycle?

Need to distinguish the links in the AI chain: large-scale training clusters (cloud vendors + xAI / OpenAI / Meta, hyperscale data centers) → enterprise private AI (companies wanting to deploy large models in their own data centers) → edge and PC AI.

HPE's main battlefield is the middle one—enterprise private AI. This market is different from "hyperscale training clusters." Enterprises can't compete for container-sized H200 GPU orders, but they need small AI clusters of one or two racks in their own data centers to run internal model fine-tuning and inference. This is a secondary benefit transmission from GPUs—after hyperscale vendors grab the first wave, the second wave of orders flows to the HPE / Dell layer.

Fourth, why is the Juniper acquisition key?

Selling AI servers is just the beginning; machines need to be interconnected. Hyperscale training has Infiniband / NVIDIA NVLink systems, but enterprise AI still uses Ethernet. Juniper is one of the Ethernet networking leaders. HPE completed its acquisition in 2024, meaning HPE can now offer customers a bundled "AI server + AI network" package. Cisco (CSCO's earnings exploded +20% today, that's another story I'll cover later) is also competing for this business, but HPE has its own complete server + storage + network stack.

Fifth, what does Irenic Capital's activist entry on 5/13 indicate?

Irenic is an activist fund specializing in "business spin-offs." An activist fund targeting HPE usually means they think some internal HPE businesses (like GreenLake, or the acquired Juniper) are undervalued and should be independent. In the short term, activist fund entry usually brings catalysts—forcing the company to increase dividends, buybacks, or spin off businesses to unlock value.

HPE is not the vanguard in the AI cycle; it's a beneficiary of cycle spillover. NVIDIA sells chips like selling bricks; HPE is the contractor. When brick prices rise, the contractor's profits don't necessarily explode in sync, but the volume is rising. HPE's current P/E ratio is lower than Dell's and lower than SuperMicro's (SuperMicro is highly dependent on AI), so it's reasonable for the activist fund to enter now—valuation has room for a rebound + AI server orders are rising + Juniper integration catalyst + Fiscal Q2 earnings approaching (conference call announced on 5/11, expected end of May).

🎯My play: HPE won't have a one-sided surge, but the probability of volatile upward movement before earnings is not small. The after-hours +5.5% has already reflected some expectations. If it pulls back below 30 after earnings, I'll consider entering in batches. If it surges above 35, I'll take profits first—a second-tier AI story can't command SuperMicro-like valuations.

But if Q2 earnings guidance is raised only modestly, or Juniper integration is slower than expected, the story will cool off. However, Irenic's entry reduces the short-term downside risk somewhat.