Ampere Computing
Last reviewed
Jun 7, 2026
Sources
18 citations
Review status
Source-backed
Revision
v1 · 1,836 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 7, 2026
Sources
18 citations
Review status
Source-backed
Revision
v1 · 1,836 words
Add missing citations, update stale details, or suggest a clearer explanation.
Ampere Computing is an American fabless semiconductor company based in Santa Clara, California, that designs high-core-count, Arm-based server processors for cloud computing and data center workloads. Founded in 2017 by former Intel president Renee James, the company set out to challenge x86 chips from Intel and AMD with energy-efficient central processing units (CPUs) built on the Arm instruction set. Its two main product lines are the Ampere Altra family and the newer AmpereOne family, which scale from 80 cores to as many as 192 cores per socket, with a 512-core design called AmpereOne Aurora announced for 2026. After years of backing from The Carlyle Group and Oracle, Ampere agreed in March 2025 to be acquired by Japan's SoftBank Group for about $6.5 billion in cash, a transaction that closed in November 2025 and folded the company into SoftBank's wider Arm and AI-infrastructure strategy.[1][2][3]
Ampere was founded in 2017 by Renee James, who had spent nearly three decades at Intel and served as the company's president from 2013 until her departure in 2016.[3][4] James assembled the startup around an engineering team drawn from MACOM Technology Solutions, which had earlier absorbed the server-chip group of AppliedMicro, giving Ampere an existing base of Arm server CPU intellectual property to build on.[3] The private-equity firm The Carlyle Group provided the initial funding, and Carlyle remained Ampere's largest shareholder through to its sale.[3][5]
From the start, Ampere positioned itself as a fabless designer: it develops the processor architecture in-house and contracts manufacturing to TSMC, the Taiwanese foundry. As an Arm architecture licensee, the company bet that the power efficiency and density of Arm cores, long dominant in smartphones, could win share in the data center, where electricity and cooling are major operating costs. Ampere's first product, the eMAG, shipped in 2018 with 32 cores built on TSMC's 16-nanometer process and derived largely from the inherited AppliedMicro design.[3] The company branded its subsequent chips "Cloud Native Processors" to emphasize the target market of hyperscale and cloud-service operators rather than traditional enterprise servers.[6]
Oracle became a strategic investor and customer in 2019, joining a funding round alongside Arm Holdings; Oracle disclosed a $40 million stake that September and is reported to have invested roughly $1.5 billion in Ampere over the following years.[4][5] By early 2022 the company was valued at around $8 billion, and it filed a confidential prospectus with the U.S. Securities and Exchange Commission in April 2022 in preparation for an initial public offering. That IPO was later shelved as market conditions for chip stocks weakened, and Ampere never went public as an independent company.[3][5]
Ampere's breakthrough product was the Ampere Altra, unveiled on March 3, 2020, as what the company called the industry's first 80-core server processor. Altra used Arm's Neoverse N1 cores on TSMC's N7 (7-nanometer) process, ran at up to 3.0 to 3.3 GHz within a 210-watt to 250-watt power envelope, and supported eight channels of DDR4-3200 memory and 128 lanes of PCIe Gen4.[7] In 2021 the company followed with Ampere Altra Max, which raised the count to 128 cores on the same platform while keeping the 250-watt design point, the eight DDR4 memory channels, and the 128 PCIe Gen4 lanes.[8][9]
The AmpereOne family, unveiled on May 18, 2023, marked Ampere's shift from licensed Arm cores to its own custom core design (codenamed Siryn). AmpereOne offered up to 192 single-threaded cores on a 5-nanometer-class TSMC process and added eight channels of DDR5 memory and 128 lanes of PCIe Gen5, along with features aimed at cloud and AI workloads such as Bfloat16 support, memory tagging, and single-key memory encryption.[6][10] Ampere later extended the line with a 12-channel-DDR5 platform: a 192-core AmpereOne M in 2024 and a planned 256-core part, widening memory bandwidth for data-heavy and inference workloads.[11]
In July 2024 Ampere announced AmpereOne Aurora, a far larger design scaling to 512 cores and integrating the company's own AI acceleration intellectual property directly onto the chip, together with support for high-bandwidth memory (HBM). Ampere said Aurora would deliver up to three times the performance per rack of the then-current AmpereOne while remaining air-cooled and deployable in existing data centers, and aimed it at AI training and inference as well as retrieval-augmented generation and vector databases. Aurora is slated to arrive in 2026.[11][12][13]
| Product | Announced | Max cores | Core design | Process | Memory and I/O |
|---|---|---|---|---|---|
| eMAG | 2018 | 32 | Skylark (ex-AppliedMicro) | TSMC 16 nm | DDR4 |
| Ampere Altra | March 2020 | 80 | Arm Neoverse N1 | TSMC N7 (7 nm) | 8-channel DDR4-3200, 128 PCIe Gen4 |
| Ampere Altra Max | 2021 | 128 | Arm Neoverse N1 | TSMC N7 (7 nm) | 8-channel DDR4-3200, 128 PCIe Gen4 |
| AmpereOne | May 2023 | 192 | Custom Ampere core (Siryn) | TSMC 5 nm class | 8-channel DDR5, 128 PCIe Gen5 |
| AmpereOne M | 2024 | 192 | Custom Ampere core | TSMC 5 nm class | 12-channel DDR5 |
| AmpereOne Aurora | July 2024 (due 2026) | 512 | Custom core plus AI engine | Not disclosed | HBM and DDR5, air-cooled |
Ampere's most important customer is also a former investor: Oracle, whose Oracle Cloud Infrastructure runs Ampere-based compute instances and which used the chips heavily for its own cloud services.[5][6] Google Cloud adopted Ampere Altra for its Tau T2A virtual machines, and Microsoft Azure offered Altra-based instances for scale-out and cloud-native workloads.[3][14] On the systems side, Hewlett Packard Enterprise built ProLiant servers around Ampere processors, and the chips were also adopted by operators including Tencent Cloud, ByteDance, and Cloudflare.[3]
This customer base illustrates both Ampere's opening and its central challenge. The company helped prove that Arm processors could handle mainstream cloud workloads, but several of the same hyperscalers have since designed their own Arm server CPUs in-house, such as Amazon's Graviton, Google's Axion, and Microsoft's Cobalt. That vertical integration narrows the merchant market that an independent supplier like Ampere can address.[6][2]
On March 19, 2025, SoftBank Group announced an agreement to acquire Ampere in an all-cash transaction valued at $6.5 billion.[1][2] Under the deal, Ampere's two principal backers, The Carlyle Group and Oracle, agreed to sell their positions. SoftBank's later disclosures put the pre-acquisition ownership at 59.65% for Carlyle, 32.27% for Oracle, and 8.08% for an Arm affiliate, Arm Technology Investment.[15] The $6.5 billion price was below Ampere's roughly $8 billion peak valuation from 2022.[5]
SoftBank framed the purchase as part of an aggressive expansion in AI infrastructure, alongside ventures such as the Stargate data center project and its "Cristal intelligence" initiative. SoftBank chairman and chief executive Masayoshi Son said, "The future of Artificial Super Intelligence requires breakthrough computing power. Ampere's expertise in semiconductors and high-performance computing will help accelerate this vision."[1] Renee James said the company was "excited to drive forward our AmpereOne roadmap for high performance Arm processors and AI."[16]
After clearing regulatory review, SoftBank completed the acquisition on November 25, 2025 (U.S. time), buying all of Ampere's equity through a subsidiary named Silver Bands 6 (US) Corp.[15][17] Ampere became a wholly owned subsidiary of SoftBank Group, retaining its name, its Santa Clara headquarters, and its leadership under James, with its roughly 1,500 employees continuing to develop the AmpereOne line.[2][5]
Ampere's strategy centers on Arm-based efficiency for cloud-native and AI-inference workloads rather than the model training that has driven demand for NVIDIA graphics processors. The company argues that high-core-count CPUs can handle a meaningful share of AI inference at lower power and cost than dedicated accelerators, particularly for smaller models and latency-sensitive serving. To extend that pitch to larger models, Ampere announced a partnership with Qualcomm in May 2024 to pair AmpereOne CPUs with Qualcomm's Cloud AI 100 Ultra inference accelerators in a single server, and AmpereOne Aurora later brought AI acceleration and HBM onto the CPU itself.[11][18]
The fit with SoftBank rests on the same theme of vertically integrated AI compute. SoftBank already controls Arm, whose architecture underpins every Ampere chip, and has assembled other AI-hardware and infrastructure assets. Folding in Ampere gives SoftBank a team of server-CPU designers and a shipping product line that can complement Arm's licensable IP, support large buildouts such as Stargate, and reduce reliance on third-party silicon as the group pursues what Son describes as artificial super-intelligence-scale data centers. Industry analysts noted that the deal was as much about acquiring Ampere's engineering talent as its current revenue.[2][5]
Ampere is one of the few independent companies to have pushed Arm into the data center at scale, helping shift an industry long dominated by x86 toward more energy-efficient designs. Its chips demonstrated that cloud providers could run general-purpose workloads on Arm cores, a trend that has since reshaped server AI chip roadmaps across the industry. The 2025 sale to SoftBank, however, ended Ampere's run as a standalone venture and recast it as a strategic component of one of the largest corporate bets on AI infrastructure. Whether Ampere thrives inside SoftBank will hinge on the AmpereOne and Aurora roadmaps and on its ability to keep winning business in a market where its largest customers increasingly design competing chips of their own.[2][6]