CEO Antonio Neri On Why HPE Has The ‘Most Cost-Effective Infrastructure’ To Run On-Premises AI Workloads
‘HPE is uniquely positioned because we have the entire portfolio across networking, cloud and AI,’ said Neri. ‘It is not just compute alone. You need to have the right cost per token through the compute side and storage. We have a very complete stack with compute and storage and Private Cloud AI.’
“We can prove to customers and partners that we can offer them the most cost-effective infrastructure to run AI workloads on-premises because we know how to optimize the full stack,” he said.
The HPE portfolio advantage is poised to deliver in general upward of 60 percent cost savings versus public cloud when deploying agentic AI and AI inferencing solutions in the enterprise market, said Neri.
“HPE is uniquely positioned because we have the entire portfolio across networking, cloud and AI,” he said. “It is not just compute alone. You need to have the right cost per token through the compute side and storage. We have a very complete stack with compute and storage and Private Cloud AI. Also, you need to have the networking capabilities to be able to connect this environment in a secure, self-driving way, which we now have with Juniper.”
Neri said another critical component of the HPE stack is a “cloud control plane to govern” AI workloads in a hybrid cloud environment. “That is what we have built inside [HPE] GreenLake with our software and the Private Cloud AI control plane,” he said. “Then you can consume it the way you want it. We have all the components to deliver that and also the financing for enterprises through HPE FS [Financial Services], which is a big asset for enterprises.”
As for AI infrastructure rivals, Neri said: “Other competitors, which I don’t like to speak to, some have some parts, some have other parts, but none of them have the complete portfolio we do.”
Neri’s comments came in an interview with HPE preceding the company’s record results for revenue, gross margin, operating profit and non-GAAP earnings per share for its third fiscal quarter ended July 31.
For the quarter, HPE reported non-GAAP diluted earnings per share of $1.11 on a 34 percent increase in sales to $12.2 billion. That compares with non-GAAP diluted earnings per share of 67 cents on sales of $9.1 billion in the year-ago quarter.
The results were above the Zacks consensus estimate of 94 cents per share on revenue of $12.1 billion.
“The headline for the quarter is that we are turning exceptionally strong demand into durable, profitable growth,” said Neri. “We are finishing 2026 very strong and entering eventually 2027 with strong momentum.”
As for HPE’s blockbuster deal to provide cloud behemoth Oracle with multi-gigawatt HPE Juniper networking infrastructure across Oracle AI data centers, Neri said: “That shows that we have a terrific portfolio both at the silicon level and at the operating system and software level with our AI.”
HPE also disclosed that after the quarter closed it was awarded a $3.5 billion AI inferencing deal with a hyperscaler provider.
Neri said that accelerated agentic AI and AI inferencing in the enterprise market is “opening up” the booming AI market opportunity in traditional servers, storage and networking for partners.
In fact, Neri said HPE partners are well suited to deliver the “right cost-effective” infrastructure to capture the growing agentic AI and AI inferencing opportunity. “Up to now it was a much more limited [opportunity for partners],” he said. “Now it is opening up, and we should expect that to continue.”
Neri’s no-holds-barred advice to partners: “Get aggressive. This is the time to scale AI deployment and inferencing with our customers.”
Below is more of CRN’s conversation with Neri.
What’s the big takeaway from the third fiscal quarter results?
The headline for the quarter is that we are turning exceptionally strong demand into durable, profitable growth.
We are finishing 2026 very strong and entering eventually 2027 with strong momentum.
Obviously, we posted record-breaking results across the company and across all metrics: orders, revenue, gross margin, operating profit, operating margin, non-GAAP EPS, [earnings per share] as well as free cash flow. That’s the main takeaway, and that’s driven by the demand and the exceptional execution by our teams.
On the demand front, networking demand is super strong. Orders grew three and a half times faster than revenue, which resulted in a record backlog for networking. Underneath that, we saw double-digit order growth in in campus and branch with record revenue for the quarter, and then in routing and data center switches we saw high-double-digit order demand, and obviously we are limited by the supply.
Security had another very good quarter with year-over-year double-digit growth. So we are very pleased with that because that basically says that our integration [of Juniper Networks] has been executed very thoughtfully and that we have a portfolio that’s perfectly aligned to the demands of the market. Ultimately, there is tremendous runway ahead of us because of the pipeline we see in the market.
How did the channel perform during the quarter in the networking business?
Our networking indirect business grew 24 percent year over year this past quarter.
Talk about what you are seeing in the Cloud & AI business.
So in Cloud & AI we had another exceptional quarter. We grew revenue 25 percent, but very importantly we had record-breaking profitability with 17 percent operating profit.
What drove that is the traditional server business because now we see an acceleration of use cases with agentic AI and AI inferencing. Our traditional server business grew 75 percent year over year on an order basis.
Our storage business grew 10 percent on revenue, and we grew twice as fast on orders. And then, obviously, our Private Cloud AI portfolio grew high double digits in excess of 60 percent year over year.
GreenLake [HPE’s pay-per-use hybrid cloud platform] continued to be a major differentiator for us. We added 8,000 new customers on the platform on a year-over-year basis. Now we have 52,000 customers, up from 44,000 customers.
How did the channel perform in the quarter on Cloud & AI?
Our Cloud & AI [channel business] grew 45 percent year over year with our revenue up 25 percent. Clearly outstanding growth for the channel. It represented 55 percent of the Cloud & AI business. The channel also accounted for over 75 percent of our total storage business.
What are you seeing with regard to agentic AI and AI inferencing in the enterprise?
As we see growth in AI inferencing and adoption of agentic AI, that will be a tailwind for the channel because obviously that is traditional infrastructure they are very used to and accustomed to selling versus selling a large AI rack-scale solution, which tends to be very concentrated on a very few customers.
After the close of the quarter, we were awarded an AI inferencing [contract] from a strategic hyperscaler which is worth $3.5 billion. That will be for their own internal usage. Think about this hyperscaler as an enterprise customer who’s going to use our traditional infrastructure to run their own AI inferencing.
What’s the Fiscal Year 2026 and 2027 outlook going forward?
The record-breaking backlog we had in Q3 basically is giving us the opportunity to first increase the 2026 guide. The 2026 guide now says that we’re going to grow revenue between 21 [percent] and 23 percent normalized for the Juniper [Networks acquisition].
Profit will grow over 100 percent year over year. That’s five times faster than revenue. Revenue this year in the first three quarters has grown twice as fast as 2025.
For 2027, we are raising the initial framework we shared last quarter for the full 2027 [forecast] on a bigger base. So we are guiding revenue between 14 [percent] and 17 percent and profit between 16 [percent] and 20 percent.
Most importantly, we are raising free cash flow to at least $5 billion. The profitability that we drove in 2026, including $2 billion of profit in Q3, allowed us to pay down the debt faster. So we returned to below two times net leverage five quarters ahead of schedule, which will allow us to return higher capital returns to shareholders starting in the fourth quarter.
We have momentum. Our strategy is proving itself again. This is durable. The partners are benefiting from this because they are drafting behind our momentum. We are excited about what comes next.
We need to navigate the supply constraints, which will be here for a longer period of time, at least through 2028, but the demand is exceptionally strong, and that’s what us makes us very confident [about the future].
Talk about the expansion of the strategic partnership with a large strategic cloud provider [later revealed as Oracle].
That is an expansion of a strategic partnership, where HPE will build multi-gigawatt AI infrastructure over a multiyear period for the networking part of that infrastructure. So this particular partner will leverage our HPE Juniper scale-out and scale-across [network infrastructure] to build multi-gigawatt [network infrastructure] over a multiyear period. That shows that we have a terrific portfolio both at the silicon level and at the operating system and software level with our AI that is a clear advantage for this particular partner.
What kind of impact do you expect the AI momentum you are seeing to have on partners?
This is good for partners because where the demand is accelerating is where the strength of the partners is, which is traditional infrastructure: storage, servers, networking versus training [of large language models], which is large infrastructure. That [large infrastructure] will continue obviously, but by the end of the decade much of the infrastructure will be consumed by AI inferencing. That’s a position of strength for us and the partners because, generally speaking, when you run that infrastructure as an enterprise customer on-prem you reduce the cost per token upward of 60 percent.
Because you’re going to use a number of multiple models with frontier open source and open-weight models, you have to have an AI factory to have the agility to basically deploy and scale AI faster with cost, control and governance. A lot of that does not require massive amount of infrastructure for the enterprise. It requires the right cost-effective infrastructure under their control. That’s why this is good for the partners. Up to now it was much more limited. Now it’s opening up, and we should expect that to continue.
What do partners need to do to continue the AI momentum with HPE?
It’s simple: Build the [AI] labs and bring those customers into your labs so they can accelerate from proof of concept to deployment and see the value of this technology, which is basically business workflow transformation. That’s the benefit of it. When customers see that, they go all in, no different than what we are doing as a company.
We are scaling AI deployments everywhere. Now we are controlling and governing the tokens and the cost per token. But HPE can offer that because we are doing it ourselves. We can do it together with our partners because they have their own capabilities to enable that process. So get aggressive. This is the time to scale AI deployment and inferencing with our customers.
What is your assessment of the enterprise AI inferencing market with regard to where HPE is versus the competition?
HPE is uniquely positioned because we have the entire portfolio across networking, cloud and AI. It is not just compute alone. You need to have the right cost per token through the compute side and storage. We have a very complete stack with compute and storage and Private Cloud AI. Also, you need to have the networking capabilities to be able to connect this environment in a secure, self-driving way, which we now have with Juniper.
You also need to have a cloud control plane to govern this. That is what we have built inside GreenLake with our software and the Private Cloud AI control plane. Then you can consume it the way you want it. We have all the components to deliver that and also the financing for enterprises through HPE FS [Financial Services], which is a big asset for enterprises.
Other competitors, which I don’t like to speak to, some have some, some have other parts, but none of them have the complete portfolio we do.
Does that give you the ability to do lower cost per token than the competition?
We can prove to customers and partners that we can offer them the most cost-effective infrastructure to run AI workloads on- premises because we know how to optimize the full stack.