Habana

Habana Labs, an Intel company, designs programmable deep learning accelerators for data center applications. The company was founded in 2016 by David Dahan, Ran Halutz, and chairman Avigdor Willenz, all veterans of fabless semiconductor companies. Habana Labs launched its Goya inference processor in 2018, followed by the Gaudi AI training processor in 2019. In December 2019, Intel acquired Habana Labs for approximately $2 billion to strengthen its AI portfolio. Habana operates as an independent business unit within Intel's Data Platforms Group.

The company's core products are the Gaudi and Goya processor lines, which are designed to offer a cost-effective alternative to GPUs for AI training and inference. The Gaudi architecture features a heterogeneous design with a Matrix Multiplication Engine (MME) and a cluster of programmable Tensor Processor Cores (TPC). A key differentiator is the on-chip integration of RDMA over Converged Ethernet (RoCE), enabling efficient scaling across thousands of nodes using standard Ethernet switches. Habana's SynapseAI software suite supports popular deep learning frameworks like TensorFlow and PyTorch, aiming to simplify model migration and development. The processors are available to customers through cloud services, such as Amazon's EC2 DL1 instances, and for on-premise deployment in servers.

Habana's second-generation processor, Gaudi2, was launched in 2022, manufactured on a 7nm process with increased memory and a higher number of Tensor Processor Cores. The company's strategy focuses on providing high-performance, cost-efficient compute for AI workloads, targeting hyperscalers and supercomputing centers. However, following the acquisition, there have been reports of challenges in gaining market traction against dominant competitors. Intel announced plans to merge the Gaudi line into its future Falcon Shores project, a combined CPU-GPU platform, which signals a strategic shift from Habana's initial roadmap. Reports in late 2025 and early 2026 indicated that the Falcon Shores project faced negative feedback and would not be marketed commercially, with Intel shifting focus to a new chip design.

Keywords: AI accelerators, deep learning processors, data center, AI training, AI inference, semiconductor, Gaudi processor, Goya processor, SynapseAI, Tensor Processor Cores, RDMA over Converged Ethernet, RoCE, fabless semiconductor, high-performance computing, cloud AI, on-premise AI, machine learning hardware, Intel, hyperscale computing, AI chips