AMD (AMD.US) teamed up with Hewlett Packard Enterprise and Broadcom! The top three joined forces to build a rack-level AI computing power platform to declare war on the “Nvidia Blackwell series”

The Zhitong Finance App learned that AMD (AMD.US), Nvidia's strongest competitor in the field of PC and data center chips, has recently fully expanded the scale of cooperation with HPE.US (HPE.US), which focuses on AI infrastructure and hybrid cloud platforms, with the aim of building open and rack-scale (rack-scale) artificial intelligence computing power infrastructure for high-performance computing clusters and large-scale AI data centers, and strives to accelerate the progress of global “Sovereign AI” (Sovereign AI) research.

According to information announced by the two parties today, HPE Technology will be one of the first system vendors to adopt the AMD “Helios” rack-level AI computing power cluster architecture, and AMD and HPE will integrate a specially built HPE Juniper Networking in-rack high-performance expansion switch (scale-up switch) based on deep cooperation with ASIC and AVGO.US (AVGO.US), a leader in high-performance network infrastructure. This large-scale artificial intelligence computing power system aims to simplify the deployment of larger AI computing power infrastructure clusters and provide more cost-effective and energy-efficient AMD AI computing power cluster solutions other than Nvidia's Blackwell family.

The collaboration between AMD, HPE, and Broadcom on the AMD rack-level Helios AI computing power cluster can be described as packaging GPU/CPU/NIC, rack system engineering, and high-end network silicon into an open rack-level AI cluster platform. For cloud computing giants and sovereign AI builders such as Microsoft, Google, and Amazon, which are spending huge sums of money to buy AI computing power infrastructure, it is not only a high-end alternative to Nvidia-led vertical integration solutions, but also an underlying template for future “sovereign AI+ frontier HPC fusion”: that is, data centers are no longer stacked GPUs, but rather a standard computing power building block that can run HPC and high-parameter AI models at the same time.

Three parties join forces to fight against “Nvidia”

This latest collaboration can be called a powerful alliance of AMD, HPE Technology, and Broadcom to set another AI computing power infrastructure benchmark with energy efficiency and cost performance advantages in the AI era; for the entire AI computing power infrastructure field, it means that the AI cluster is moving from the “stack card era” to “rack as a product”; for AMD, which strives to continuously encroach on Nvidia's AI GPU market share in the AI era, it is an upgrade from “selling chip stack computing cards” to “selling chip stacks” in the AI era. Sell a complete set of racks “Solution” is expected to significantly increase the scale of revenue generation and bargaining power of AI-based data centers in the medium to long term.

“With 'Helios', we are taking this collaboration to a deeper stage, combining AMD full-stack high-performance computing technology with HPE's system-level innovation to provide customers with an open, rack-level AI computing power platform in the AI era, opening a new level of efficiency, scalability, and breakthrough performance.” AMD CEO Lisa Su (Lisa Su) said in a recent partnership statement.

Antonio Neri, CEO of HPE, added: “With the launch of the new AMd'Helios and our purpose-built HPE (HPE Technology) rack-scale networking solution, we are providing significant help to cloud computing service providers, in terms of the speed of deploying AI applications globally, making rack AI systems more flexible, and reducing computational risk, thereby supporting them to massively expand AI computing power clusters in their business.”

Jointly built by three parties, this latest AI rack-level AI cluster platform uses an AMD Instinct MI455X GPU, an AMD EPYC Venice CPU, and an AMD Pensando Vulcano network card (NiCS) for horizontally expanding networks.

According to information, Helios is also being used to drive the “Herder” supercomputing architecture, a “next-generation supercomputer” at the Stuttgart High Performance Computing Center in Germany, which is one of the sovereign AI systems being built by the German government.

“Our research user community demands that we continue to support traditional high-performance computing (HPC) numerical simulation applications that require high computing power.” Michael Resch, director of the computing center, said. “At the same time, we are also seeing growing interest from some major customers in machine learning and large models of high-parameter artificial intelligence. Herder's system architecture, which we're building with AMD Helios, will enable us to simultaneously support these two extremely demanding models and provide customers with huge computing power capacity to develop and benefit from the new hybrid HPC/AI workflow.”

This new supercomputer at the sovereign AI level is expected to be put into operation by the end of 2027.

Is the “King of the Hill” AMD stock price embarking on a new round of gains?

Through Helios, a rack-level platform that AMD spent years of painstaking work to build, AMD is actually selling a complete set of MI455X + EPYC + Pensando's “combined AI computing power infrastructure”.

This tripartite partnership also helps AMD to bid for the “NVIDIA Blackwell Full Stack AI Computing Power Infrastructure Solution” led by Nvidia in large-scale AI data center project tenders. It is expected to greatly increase AMD's medium- to long-term AI data center chip market share, and give global enterprises or sovereign agencies a “non-NVIDIA vertical integration” high-end AI computing power infrastructure option.

In this major partnership, AMD exclusively provides a complete set of AI core computing power clusters and data channels: Instinct MI455X (GPU), EPYC “Venice” (server CPU), and Pensando “Vulcano” (NIC/DPU), which covers the “core computing power + network intelligent network card” layer; HPE Technology is responsible for server forms, rack engineering (electricity, cooling, wiring), management software, delivery and operation and maintenance, turning traditional piled up AI chips into “rack-level products” systems The manufacturer is tantamount to making an “official full-rack reference design + commercial implementation” for AMD Helios, while Broadcom is the network silicon hidden behind HPE Juniper Networking high-performance switches, providing actual switching chips and data center-level high-performance Ethernet capacity.

AMD's stock price has surged more than 80% since this year, but most of the increase was concentrated in October. The main catalyst for AMD's strong rise was the agreement reached with Saudi Arabia's “sovereign AI system” Humain to purchase an AI chip computing power cluster with 1 gigawatt of electricity power, and the chip company reached a 100 billion dollar important AI computing power infrastructure cooperation agreement with OpenAI, a leader in artificial intelligence applications. These large-scale collaborations not only verified the strong strength of AMD's AI computing power infrastructure technology, but also made Wall Street investment institutions more optimistic about their future financial prospects.

Wall Street financial giant Citigroup called AMD the “King of the Hill”. “Considering the higher-level earnings per share (EPS) growth rate in the 2027 calendar year, AMD is the stock with the strongest buying momentum in the market. Our research feedback shows that recent Analyst Day activities will help consolidate its revenue/profit margin goals and future EPS targets of up to $20.” The team led by Citibank analyst Christopher Danely wrote in a report to clients and reaffirmed the target price of $260.

According to the average price target price of Wall Street analysts compiled by TIPRANKS, analysts expect the average target price for AMD to be as high as 284.67 US dollars, which means a potential increase of at least 32% in the next 12 months. The highest target stock price is the bull market target price of up to 377 US dollars given by the well-known investment agency Raymond James when covering AMD shares for the first time.

AMD CEO Su Zifeng gave a very optimistic forecast for the artificial intelligence (AI) computing power chip market in early November, and it is expected that AMD's performance will show a stronger growth trajectory over the next five years. Su Zifeng announced AMD's financial fundamentals for the next three to five years. She said that AMD's goal is to have a “double digit” share in the data center artificial intelligence chip market. It is estimated that AMD's annual data center chip revenue will reach 100 billion US dollars within 5 years (compared to the current about 16 billion US dollars), and profits are expected to more than double by 2030.