Tag: Ampere Computing

Bloomberg News reports that SoftBank's previously announced $65 billion acquisition of Ampere Computing has been placed under a "Phase Two investigation" by the U.S. Federal Trade Commission (FTC). This means the merger may face a lengthy review and could even be rejected. The FTC's "Phase Two Information Request" is a relatively rare in-depth investigation process at the U.S. federal level, targeting potential monopoly risks. Once this investigation begins, it can take months, even more than a year, and could potentially lead to lawsuits to block the transaction. Neither SoftBank nor Ampere Computing has commented on the investigation rumors. SoftBank's acquisition faces further regulatory challenges, potentially repeating the Arm experience. SoftBank acquired Arm for $320 billion in 2016, and its subsequent acquisition of Ampere Computing was seen as strengthening its strategic puzzle in artificial intelligence chip development. Ampere Computing, founded in 2018, has focused on cloud-native computing and has recently expanded into artificial intelligence chip design. If the deal passes review, SoftBank will simultaneously control three influential companies in the AI ​​chip field: Arm, Graphcore, and Ampere Computing. However, this could also raise concerns among regulators about the competitive structure of the market. Previously, SoftBank's attempt to sell Arm to NVIDIA was blocked by US and European regulators in 2022 due to similar antitrust investigations. With Stargate and Qualcomm lawsuits looming, SoftBank faces increasing pressure. Besides the challenges of the Ampere Computing deal, SoftBank has also encountered obstacles in several other areas. For example, the Stargate supercomputer project, led by SoftBank, is currently hampered by US tariff policies, resulting in slower-than-expected progress. Meanwhile, SoftBank's Arm is embroiled in licensing and market monopoly disputes with Qualcomm in Europe, the US, and South Korea. It remains unclear whether the US Federal Trade Commission will block SoftBank's acquisition of Ampere Computing, but given the escalating global AI chip war, the success or failure of this deal will undoubtedly influence SoftBank's next phase of its global AI strategy.

Following recent reports of potential acquisition talks between Softbank and Ampere Computing, Softbank has confirmed its acquisition of Ampere Computing for $65 billion. Ampere Computing will become a subsidiary of Softbank while maintaining independent operations, with its headquarters remaining in Santa Clara, California. Ampere Computing founder Renée James will remain as CEO. Founded by former Intel president Renée James, Ampere Computing primarily focused on building server processors based on the Arm architecture. However, it later launched its own custom architecture processor brand, AmpereOne, to reduce its heavy reliance on Arm. In mid-May 2023, it announced a custom architecture processor with up to 192 computing cores, designed for cloud-native applications. It also collaborated extensively with Qualcomm to reduce the AI ​​market's reliance on GPUs with a new data center processor. Ampere Computing's market valuation was $8 billion in 2021, but has since decreased to $65 billion due to related debt, allowing Softbank to acquire it at the same price. Following the acquisition by Softbank, the shares of Ampere Computing previously held by US private equity firm Carlyle Group and Oracle will be held by Softbank. After being integrated into the Softbank Group, it's easy to predict a deeper collaboration between Ampere Computing and Arm, accelerating the development of Arm-based data centers. However, whether this will affect other companies already using or planning to use Arm-based processors for their data centers remains to be seen.

Rumors have recently surfaced that Arm will launch its first in-house processor targeting the server market this summer, with adoption expected from companies like Meta. SoftBank, Arm's largest shareholder, is also reportedly in acquisition talks with Ampere Computing, which could further benefit Arm's in-house processor development. Sources also suggest that Arm has poached senior executives from its technology licensing clients to participate in processor design. However, Arm has not responded to these rumors. Current market speculation suggests that Arm's investment in in-house processors will focus on server applications, primarily in data centers, cloud services, and large-scale AI acceleration computing, rather than targeting the consumer market. This move will not only put Arm in closer competition with NVIDIA, AMD, and Intel, but also with numerous partners building server processors based on the Arm architecture. However, Arm's considerations might be similar to Google's approach of developing its own Pixel series. While there will be competition in the market, the primary purpose is to demonstrate market applications, providing design references for more partners and showcasing the potential of new designs. Meanwhile, the market believes that Arm's involvement in processor product design might allow it to access larger market opportunities. However, some speculate that once Arm successfully launches its own server-side processor, it may target the mobile and even PC markets with its own processors. This approach, however, could also carry risks, causing existing partners to switch to licensing other processor technologies, such as the open-source RISC-V architecture.

The Financial Times reports that Arm may launch its first in-house processor as early as this summer for data center operations. The processor will be manufactured by TSMC or other foundries, and initial customers will include companies like Meta. This rumor may be related to recent reports of Softbank's acquisition talks with Ampere Computing, potentially helping Arm establish a foothold in the data center market with its in-house processors. If Arm does indeed launch its in-house processor for data centers, it could become a reference point for other companies building data center processors based on the Arm architecture, similar to how Google's Pixel series phones serve as a reference design for most Android phones. However, neither Softbank nor Ampere Computing has responded to the reports of Softbank's planned acquisition of Ampere Computing, and these discussions may change later. Therefore, it remains uncertain whether Softbank will acquire Ampere Computing.

Bloomberg News reports that SoftBank appears to be in talks to acquire Ampere Computing, with an announcement expected in the coming weeks. Ampere Computing, founded by former Intel president Renee James, primarily focuses on Arm-based server processors. However, it later launched its own custom architecture processor brand, AmpereOne, to reduce its heavy reliance on Arm. In mid-May 2023, it announced a custom architecture processor with up to 192 cores, designed for cloud-native environments. It also collaborated extensively with Qualcomm to reduce the AI ​​market's dependence on GPUs with a new data center processor. Ampere Computing's market valuation was $8 billion in 2021, but has since decreased to $65 billion due to related debt. Oracle is currently a major investor in Ampere Computing and uses Ampere Computing processors in its server products. According to data released last year, Oracle holds approximately 29% of Ampere Computing's equity, plus options and convertible notes, giving it the ability to further control Ampere Computing. Regarding the current rumors of Softbank's plan to acquire Ampere Computing, neither Softbank nor Ampere Computing has responded. These discussions may change later, so it's currently impossible to confirm whether Softbank will acquire Ampere Computing. However, if Softbank does acquire Ampere Computing, it might benefit the development of its Arm business. Whether CEO Renee James will remain in her position or move to other areas if Softbank acquires Ampere Computing is also attracting considerable market attention.

Ampere Computing announced a collaboration with Qualcomm to leverage their respective Ampre Altra CPUs and Cloud AI 100 Ultra AI inference chips to create a new data center processor. This collaboration aims to help more businesses reduce their reliance on GPU-accelerated computing. Ampere Computing CEO Renee James stated that the increasing power consumption requirements and energy consumption challenges of artificial intelligence make Ampere Computing's focus on performance and energy efficiency chip design a significant advantage. The company has also demonstrated over the past six years that low-power processor designs are not necessarily inefficient. James emphasized that current AI technology trends indicate that low-power and high-performance designs will be the future of processor design, ensuring environmental sustainability – a key development goal for Ampere Computing. Ampere Computing Product Chief Jeff Wittich further stressed that the company's products meet the needs of cloud-native computing and AI applications, addressing computational requirements such as data processing, network services, and media delivery. Taking the Llama 3 open-source large-scale natural language model recently announced by Meta as an example, in the Oracle Cloud environment using Ampere Computing's CPUs, the performance of the Ampere Altra CPU (which does not include a GPU architecture and uses a 128-core design) was the same as that of an NVIDIA A10 GPU with an x86 architecture CPU, but with only one-third the power consumption. Further explanation from Ampere Computing revealed that its upcoming product will use the same air-cooling design as its current CPU with 192 cores and 12-channel memory, but with 256 cores, offering 40% higher computing performance. It is expected to enter the market this year using TSMC's 3nm process and will initially be used in Supermicro server products, while also collaborating with other companies. Ampere Computing also explained that it will leverage the UCIe design specification to integrate other companies' chip designs with more flexible interfaces, thereby creating more powerful CPU products.

Following its announcement in late May last year of breaking away from Arm architecture and unveiling its new self-developed processor brand, AmpereOne, Ampere recently provided details on the first wave of Ampere One series processors in the Chinese market. These include a single-threaded CPU design with up to 192 cores and utilize the Arm instruction set, making it compatible with existing Ampere Altra and Ampere Altra Max processors based on the Arm Neoverse architecture. The AmpereOne series processors are not intended to replace existing Ampere Altra series products, but rather are designed for cloud-scale computing applications with higher computational density and high data throughput, thereby meeting the demands of large-scale cloud computing applications. The series utilizes Ampere's proprietary core design, offering 136, 144, 160, 172, and up to 192 core designs, with a maximum operating clock speed of 3.0GHz. Manufactured using a 5nm process, each core is equipped with 2MB of L2 cache, providing 16KB of cache for instruction set applications and 64KB of cache for data applications. The L3 cache is 64MB. Memory support is 8-channel ECC DDR5, supporting up to 16 memory slots and 8TB of capacity. It also supports up to 128 PCIe Gen 5 lanes, or an optional 64 CCIX lanes. Furthermore, it incorporates network congestion management, power management, and memory tagging features to optimize cloud computing for faster response times in AI environments. Ampere emphasizes that its AmpereOne series processors offer higher computing performance for cloud-native workloads while maintaining high scalability and computing density. Furthermore, through low latency and high data throughput, they deliver more than twice the performance of other brands' processors for cloud-based AI inference. Currently, cloud service providers including Google Cloud, Microsoft Azure, Oracle, Alibaba, and Tencent, as well as server providers such as HPE and Supermicro, are all collaborating extensively with Ampere.

Google Cloud announced that it will integrate Ampere Computing's Altra processors into its services, thereby increasing the proportion of Arm-based processors on its cloud service platform. Last year, Google Cloud added a virtual machine project called Tau to its Compute Engine service, which included the Tau T2D virtual machine, running on AMD's third-generation EYPC server processors based on the x86 architecture. The newly announced Tau T2A virtual machine, powered by Ampere Computing's Altra processors, will further enhance the efficiency of cloud virtual machine operations while reducing power consumption. Google Cloud also announced a deeper collaboration with Ampere Computing, becoming a key partner. The Tau T2A virtual machine can emulate up to 48 CPUs in a single virtual machine, with each CPU configured with 4GB of memory, up to 32Gbps of network bandwidth, and storage capacity that can be expanded according to service needs. This will help users build various network services, containerized microservices, and data storage applications. Besides Google, companies including AWS, Microsoft, Oracle, Alibaba, Tencent, and Baidu have begun to adopt Arm architecture processors to improve server operating efficiency and reduce the previously significant power consumption.

In addition to companies like Qualcomm resuming their own architecture development, Ampere Computing, which currently uses the Arm Neoverse N1 architecture to build its server processors, has officially announced that it will use Ampere One as the brand name for its future self-developed processors. Ampere One will adopt a completely self-developed architecture design, not using the standard Arm architecture, but only using an Arm instruction set license. It will be manufactured using TSMC's 5nm FinFET process and supports PCIe 5.0 ports and DDR5 memory. Previously, Ampere Computing consistently used the standard Arm architecture to design its server-side computing processors. Last year, however, it announced the launch of its self-developed processor products, thereby reducing its reliance on the Arm architecture and providing more demanding performance for specific computing applications. Besides Ampere Computing, companies including Qualcomm and Apple are currently favoring self-developed architectures to create different market demands for processor performance and achieve greater hardware and software integration benefits with their own products or related applications. Apple has already developed many custom-designed A-series processors using Arm instruction set licensing, and has also developed Apple Silicon processors for PC computing needs. Qualcomm, on the other hand, plans to acquire Nuvia's assets to develop custom-designed processors with higher computing performance for markets such as PCs.

Ampere Computing, known for its Arm-based server processors, recently announced that it has filed documents with the U.S. Securities and Exchange Commission, anticipating a future IPO. Headquartered in Santa Clara, California, Ampere Computing was founded in 2017 by Renée James, who worked at Intel for over 28 years, rising from engineer to president. The company builds its server processors using Arm architecture and mass-produces them using TSMC's manufacturing process. Its main clients are Oracle and Microsoft. Recent news indicates that Ampere Computing received a $4.26 million investment from Oracle, and Oracle has also acquired a significant stake in the company. Previous reports even suggested that Oracle planned to acquire more Ampere Computing shares from Softbank, demonstrating Oracle's confidence in Ampere Computing's future development. A future IPO is expected to allow Ampere Computing to raise more funds through the IPO, enabling it to compete with semiconductor companies like Intel and AMD in the server market.