Counterpoint: Broadcom Inc. (AVGO.US) is expected to lead the AI ASIC design market, with a projected market share of 60% by 2027.

According to Counterpoint Research's analysis, Broadcom Inc. (AVGO.US) is expected to continue to maintain its leading position in the field of AI server computing ASIC design partners in 2027, with a market share reaching 60%. Counterpoint predicts that the shipment volume of AI server computing ASICs will double in 2027, driven by the demand for Alphabet Inc. Class C's TPU infrastructure (to support the Gemini project), the continued expansion of Amazon.com, Inc.'s Trainium cluster, and the capacity increase brought by Meta's MTIA and Microsoft Corporation's Maia chips as their internal product lines expand. Counterpoint analyst Neil Shah said, "The growth in internal AI server computing ASIC design is validating the era of custom XPU for internal use, where AI accelerators are tailored for specific workloads (training or inference) and no longer solely rely on general-purpose GPUs. As power and space become bottlenecks, moving some AI workloads to vertically integrated chips not only gives hyper-scale data center operators greater control and advantages, but also brings a lot of software support to optimize AI workloads and enjoy performance and efficiency improvements." Additionally, Counterpoint expects the shipment volume of AI server computing ASICs to exceed 15 million units by 2028, surpassing the shipment volume of data center GPUs. Shah pointed out, "The top ten AI hyper-scale data center operators are expected to deploy over 40 million AI server computing ASIC chips from 2024 to 2028. Supporting this unprecedented demand are AI hyper-scale data center operators who have built massive rack-level AI infrastructure based on their own technology stacks, such as Alphabet Inc. Class C's TPU Pod and AWS's Trainium UltraCluster, allowing them to operate like a supercomputer." Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR (TSM.US) holds nearly 99% of the wafer manufacturing share in the top ten companies shipping AI server computing ASICs. Although Alphabet Inc. Class C and Amazon.com, Inc. still dominate the shipment volume of AI server computing ASICs in 2024, the market landscape is rapidly diversifying with the entry of Meta, Microsoft Corporation, and other companies into the field. By 2027, Alphabet Inc. Class C's market share is expected to decrease from 64% to 52%, while Amazon.com, Inc.'s market share is expected to decrease from 36% to 29%. Shah stated, "Due to the continuously expanding market size and competitors in the hyper-scale data center space partnering with companies like Broadcom Inc., Marvell Technology, Inc. (MRVL.US), and Alchip to adopt proprietary chips, Alphabet Inc. Class C's market share is expected to drop to 52% in 2027. However, its TPU chip group will remain the undisputed industry core pillar and guiding light. This benchmark is built on the massive and sustained computing intensity required to train and run next-generation Gemini models, which necessitates continuous and active upgrading of proprietary chip infrastructure." This shows that hyper-scale data centers are gradually moving away from over-reliance on NVIDIA Corporation (NVDA.US) and seeking custom-designed chips to meet some computing needs. By 2027, Broadcom Inc. and Alchip are expected to continue to hold the majority of the share of these hyper-scale data center ASIC design service partners, with 60% and 18% respectively. However, during this period, Marvell Technology, Inc.'s share is expected to decrease from 12% to 8%. Counterpoint analyst Gareth Owen said, "Nevertheless, Marvell Technology, Inc.'s end-to-end custom chip product portfolio looks more solid than ever, thanks to its custom silicon innovation technologies, such as custom HBM/SRAM memories and PIVR solutions, as well as the acquisition of Celestial AI, which expand Marvell Technology, Inc.'s potential market in the field of scale-out connectivity. Celestial AI not only brings in billions of dollars in revenue for Marvell each year, but also has the potential to propel Marvell to a leading position in the optical scale-out connectivity field in the coming years."