Hewlett Packard Enterprise CEO Antonio Neri said he does not see traditional supercomputing workloads moving to Amazon Web Services’ just-announced UltraCluster AI supercomputer.
“I don’t see anytime soon the traditional supercomputing workloads moving there because the cost is prohibitive,” Neri told CRN when asked about the new AWS UltraCluster supercomputer. “Now we know hyperscalers, including AWS, tend to focus on disaggregating the architecture so they have more control of their stack. But that’s true for maybe generative AI, but I’m not sure it’s true for the traditional supercomputer workloads.”
AWS announced this week that it is building the UltraCluster AI supercomputer — named Project Rainier — which will be released next year – with AWS’ own Trainium2 chips. Those Trainium2 chips go head-to-head against Nvidia’s popular GPUs, which are part of the ‘Nvidia AI Computing By HPE’ portfolio including HPE Private Cloud AI.
AWS has said its UltraCluster supercomputer will contain hundreds of thousands of Trainium2 chips with more than 5x the number of exaflops used to train their current generation of leading AI models.
AWS declined to comment.
Neri said it remains to be seen what “use cases” the AWS supercomputer will address for customers. “If you think about exabytes of data it is hard to say what customers, particularly the most sensitive customers that do a lot of simulation, modeling or very large training models, will do with it,” he said. “We will see what use cases they can address.”
Neri said HPE’s unique supercomputing “expertise” packs a big punch given that it holds seven of the top 10 supercomputing systems on the Green500 list including the top three supercomputers. The Green500 list ranks the world’s most energy-efficient supercomputers.
HPE, Neri said, also has the industry’s first 100 percent fanless direct liquid cooled systems architecture, which is aimed at providing breakthrough energy and cost efficiency for large-scale AI deployments.
Just last month, HPE announced that it had delivered the “most powerful supercomputer in the world” – the 100 percent fanless direct liquid-cooled EL Capitan to the U.S. Department of Energy’s Lawrence Livermore Laboratory. The El Capitan system runs 1.742 exaflops with 58.89 gigaflops performance per watt.
“The key there is that you need more than chips,” said Neri of HPE’s supercomputing expertise. “You need a networking fabric that can drive coherence. Then you need the storage, and that data needs to be co-located next to the supercomputer.”
Neri’s comments came after HPE reported better-than-expected results for its fourth fiscal quarter ended October 31 with quarterly AI systems revenue growth of 16 percent to $1.5 billion. The HPE AI systems backlog now stands at more than $3.5 billion with a sales pipeline that is a multiple of that.
HPE sees continued AI momentum with AI model builders, hyperscalers and service providers, said Neri. What’s more, HPE sees “continued momentum” in sovereign AI projects.
As for the enterprise market, Neri said, HPE’s Private Cloud AI– which only became available on September 9 – is already gaining a foothold. “We closed the first deals in Q4 and now we see POCs (Proof of Concepts) and pipeline growing very rapidly,” he said.
Overall, HPE reported “record” quarterly sales of $8.5 billion for its fourth fiscal quarter ended October 31, up 15 percent in constant currency from $7.71 billion. That was well above the Zacks consensus estimate of $8.23 billion. HPE also reported a “record” non-GAAP operating profit of $938 million.
HPE reported non-GAAP earnings per share of 58 cents per share, up 12 percent from the year-ago quarter, and three cents above the Zacks consensus estimate of 55 cents per share.
HPE shares were up nearly 11 percent or $2.31 at the end of trading Friday to $23.95, a new 52-week high. This after the strong fourth quarter results and a Morgan Stanley upgrade for HPE shares with a target price raise from $23 to $28.
Neri said he is “excited” about the outlook for 2025 given HPE’s $14 billion acquisition of Juniper Networks is set to close early next year.
“We expect the pending acquisition of Juniper Networks to further enhance our portfolio providing customers with complete edge to cloud solutions,” he said. “I am extremely excited about the significant opportunity we have in the coming fiscal year to drive increased value for our shareholders. I am optimistic about what we can achieve and look forward to the year ahead.”
What were the highlights of HPE’s fourth-quarter performance?
We had an exceptional fourth quarter, capping off a super, super strong fiscal year 2024. You can see it in the numbers. Obviously we had record-breaking revenue of $8.5 billion. That’s up 15 percent year over year and nine percent sequentially. That’s a result of the continuous momentum we see in the market because of our innovation.
Second, we delivered again record-breaking performance on operating profit of $938 million and we exceeded our high end of the guidance with non-GAAP diluted earnings per share of 58 cents.
Underneath that, every business segment grew both on an orders and revenue basis sequentially. In many of the segments, we grew orders double-digits year over year. That includes traditional servers, which grew in double-digits for order growth for the third consecutive quarter.
The Intelligent Edge business grew sequentially for the third consecutive quarter. (Intelligent Edge) revenue looks flattish because a lot of the growth comes from subscription based software and services with solutions like SASE, SSE and SD-WAN.
Now we see momentum in the Wi-Fi 7 refresh cycle and also the data center networking business. The only area still going through some digestion is the campus switching. But as you know, as we go through the Wi-Fi 7 refresh, as you refresh the access point you also will have to refresh the switch because you need more power for the access point to power over Ethernet.
What were the key AI metrics in the quarter?
When you look at AI first of all on the revenue side we converted more revenue in the quarter, up 16 percent to $1.5 billion. Today we have a backlog of more than $3.5 billion and a pipeline that’s a multiple of that.
When you look underneath what’s happening with the orders obviously there is a lot of momentum continuing with the model builders and service providers, whether it’s hyperscalers or tier-two and tier-three (service providers). Now we see continued momentum in sovereign AI although sales cycles are longer.
In the enterprise, we have our (HPE) Private Cloud AI which is the perfect offer designed and tailored for the enterprise and our channel. That’s because of the simplicity of the (HPE Private Cloud AI) offer in terms of sales and deployment. We closed the first deals in Q4 and now we see POCs (Proof of Concepts) and pipeline growing very rapidly. Understand we made the offer available on September 9. That is all positive.