Annapurna Labs

Annapurna Labs is an Israeli fabless semiconductor company and Amazon subsidiary whose work is focused on cloud and data-center infrastructure. Established in 2011, the company was founded by Hrvoje Bilic, Nafea Bshara, and Ronen Boneh, with backing from a group of technology investors and semiconductor-industry participants. Its name refers to the Annapurna mountain massif in the Himalayas. Amazon acquired Annapurna Labs in January 2015. After the acquisition, the business became part of Amazon’s infrastructure-chip strategy and its technology was integrated into Amazon Web Services. Rather than operating primarily as a retail semiconductor brand, Annapurna Labs develops components that are embedded in AWS servers, networking systems, storage infrastructure, and machine-learning platforms. Its products are therefore usually encountered by cloud customers through AWS services and instance types rather than through direct purchase of chips. The division is associated with several important AWS hardware families. Nitro provides specialized hardware and software for networking, storage, security, and virtualization offload in Amazon’s Elastic Compute Cloud infrastructure. Graviton is a family of Arm-based general-purpose server processors designed for AWS cloud workloads. Trainium and Inferentia are purpose-built machine-learning accelerators intended for training and inference workloads. Together, these product lines support Amazon’s effort to control more of its cloud hardware stack and optimize performance, cost, isolation, and energy use for specific workloads. Nitro was first deployed in fourth-generation EC2 servers before evolving into a broader architecture that included the Nitro Hypervisor. AWS has subsequently described Nitro as a foundational part of its server platform. Graviton has progressed through multiple generations, with later processors offering substantially more cores and newer memory technologies than the initial design. Trainium has likewise developed through several generations for increasingly demanding artificial-intelligence workloads. Annapurna Labs is generally understood as an internal technology organization within Amazon rather than an independent public company with a separate consumer-facing product portfolio. Amazon does not publicly report standalone revenue, profitability, unit sales, or a complete organizational structure for the division. Public descriptions of its scale, manufacturing relationships, and product performance should consequently be treated as company or industry disclosures rather than as independently audited Annapurna Labs financial reporting.

Annapurna Labs was established in Israel in 2011 by Hrvoje Bilic, Nafea Bshara, and Ronen Boneh. The founders pursued a fabless semiconductor model, concentrating on the design of chips and systems for networking, storage, and data-center infrastructure rather than manufacturing silicon themselves. The company attracted investment from Avigdor Willenz, Manuel Alba, Andy Bechtolsheim, Walden International, Arm Holdings, and TSMC. Its board was associated with figures including Willenz, Alba, and Lip-Bu Tan. The company’s early direction aligned with a major change in cloud computing. Large cloud providers increasingly sought specialized hardware that could improve server efficiency, network throughput, workload isolation, and control over infrastructure costs. Annapurna Labs developed technology that became associated with AWS Nitro. Nitro hardware appeared in fourth-generation EC2 servers, including the C4, M4, and D2 families, before expanding into a broader platform for networking, storage, security, and virtualization functions. The later Nitro Hypervisor moved additional virtualization responsibilities from conventional software layers into a purpose-built architecture. Amazon announced the acquisition of Annapurna Labs in January 2015. Following the transaction, Annapurna Labs ceased to function primarily as an independent startup and became a wholly owned Amazon semiconductor operation. Its engineering work was directed toward Amazon’s own infrastructure and AWS services. Nitro became a foundational element of AWS server design, with specialized hardware handling functions such as virtual networking, Elastic Block Store connectivity, and local NVMe storage. The division subsequently broadened its portfolio. AWS Graviton introduced Arm-based general-purpose server processors for cloud instances, allowing Amazon to optimize processor design, system software, and cloud services together. Later Graviton generations increased processing capacity and adopted newer memory interfaces. The product family became an important alternative to conventional x86-based cloud processors and was also used to run parts of the AWS service portfolio. Annapurna Labs also developed machine-learning silicon. AWS Trainium was designed primarily for model training, while AWS Inferentia targeted inference workloads. These accelerators formed part of Amazon’s attempt to provide cloud customers with alternatives to general-purpose processors and externally sourced graphics or accelerator hardware. Later Trainium generations were associated with large-scale artificial-intelligence infrastructure, including systems used for demanding model workloads. Public accounts describe Nitro, Graviton, Trainium, and Inferentia as the principal product families connected with Annapurna Labs. However, the division’s exact legal, managerial, and operational boundaries within Amazon are not fully disclosed. Amazon does not publish standalone financial statements for Annapurna Labs, and the company is not independently listed. It is therefore best characterized as an Amazon infrastructure-semiconductor division and engineering brand rather than as a conventional independent chip vendor.

1. 2017Nitro Hypervisor introduced

   AWS introduced the Nitro Hypervisor as a later stage in the development of its hardware-assisted virtualization architecture.

   Source
2. 2015Amazon acquired Annapurna Labs

   Amazon acquired the Israeli semiconductor company and incorporated its engineering capabilities into Amazon’s cloud-infrastructure hardware effort.

   Source
3. 2014Nitro hardware entered AWS EC2 deployments

   AWS began deploying Nitro-related hardware in fourth-generation EC2 server families, including C4, M4, and D2.

   Source
4. 2011Annapurna Labs founded in Israel

   Hrvoje Bilic, Nafea Bshara, and Ronen Boneh established the fabless semiconductor company to develop infrastructure-oriented chips and systems.

   Source
5. Graviton processor family expanded

   AWS released successive generations of Arm-based Graviton processors, increasing core counts and improving the platform’s suitability for cloud workloads.

   Source
6. Trainium and Inferentia added machine-learning acceleration

   The product portfolio expanded into dedicated silicon for machine-learning training and inference.

   Source

A vertically integrated cloud-infrastructure semiconductor division focused on specialized computing, networking, storage, virtualization, and machine-learning hardware for Amazon Web Services.

• AWS Nitro SystemCloud infrastructure hardware and software2014

  AWS Nitro is a hardware and software architecture for offloading infrastructure functions from customer-facing compute processors. Its components support networking, storage, virtualization, and security functions in AWS servers. Nitro hardware underpins virtual private cloud networking, Elastic Block Store connectivity, and local NVMe storage, while the Nitro Hypervisor provides a lightweight virtualization layer. The system is designed to improve performance, isolation, and control across cloud instances.

• AWS GravitonArm-based data-center processors

  AWS Graviton is a family of Arm-based server processors developed for AWS cloud instances. The processors are designed together with AWS infrastructure and software so that Amazon can tune performance, energy efficiency, and price-performance for selected cloud workloads. Successive generations have increased core counts and adopted newer memory technologies. Graviton is used not only for customer-facing compute instances but also for portions of the AWS service portfolio.

• AWS TrainiumMachine-learning training accelerators

  AWS Trainium is a purpose-built accelerator family for training machine-learning models in the cloud. It is intended to provide an AWS-designed alternative to relying solely on general-purpose processors or externally sourced accelerators. Multiple generations have been developed for increasingly large and demanding artificial-intelligence workloads, including large distributed training systems.

• AWS InferentiaMachine-learning inference accelerators

  AWS Inferentia is a family of custom accelerators for machine-learning inference. The product is intended to help AWS customers run trained models with appropriate throughput, latency, and cost characteristics. It forms part of the same broader Amazon strategy of designing specialized silicon for cloud workloads rather than depending exclusively on general-purpose server processors.

• 2015Amazon acquired Annapurna LabsM&A

  Amazon sought greater control over the semiconductor and systems technology used to operate its rapidly expanding cloud infrastructure.

  What changed. Amazon acquired Annapurna Labs in January 2015 and brought the company into its internal infrastructure-chip organization.

  Aftermath. The company’s technology became associated with AWS Nitro, Graviton, Trainium, and Inferentia rather than with an independent commercial chip portfolio.

  Source
• 2014Initial AWS Nitro deploymentProduct launch

  Cloud servers required more efficient handling of networking, storage, and virtualization functions as instance performance increased.

  What changed. AWS deployed Nitro-related hardware in fourth-generation EC2 server families.

  Aftermath. Nitro evolved into a core AWS server architecture and later incorporated the Nitro Hypervisor.

  Source
• Expansion into custom CPUs and AI acceleratorsStrategy

  Amazon wanted to optimize cloud computing economics and performance through closer integration of hardware, software, and AWS services.

  What changed. Annapurna Labs expanded from infrastructure offload into the Graviton server-CPU, Trainium training-accelerator, and Inferentia inference-accelerator families.

  Aftermath. AWS gained a broader internally designed silicon portfolio spanning core compute, infrastructure services, and machine-learning workloads.

  Source

• 2017AWS expanded Nitro infrastructure hardware

  AWS introduced the Nitro Hypervisor as part of the continuing development of its hardware-based infrastructure offload architecture.

  Product launchProduct generation
  Source
• 2015Amazon acquired Annapurna Labs

  Amazon completed the acquisition of the Israeli semiconductor company, bringing its engineering capabilities into Amazon’s infrastructure-chip organization.

  M&A
  Source
• AWS continued expanding Graviton server processors

  The Graviton family developed through successive generations of Arm-based processors for AWS compute services, increasing core counts and supporting newer memory technologies.

  Product generation
  Source
• AWS developed Trainium and Inferentia accelerators

  Annapurna Labs expanded beyond general-purpose CPUs and infrastructure controllers into machine-learning training and inference silicon.

  Product launchProduct generation
  Source

1. Annapurna Labs