On October 30, at its 2026 third-quarter earnings call, MediaTek confirmed that it has secured its second AI application-specific integrated circuit (AI ASIC) customer, with the new project expected to enter volume production and begin contributing revenue in 2028. Market speculation had widely suggested that the new customer was ByteDance, but MediaTek clarified directly on the call that the customer is not a Chinese company, though its name has not yet been officially disclosed.
Public records show that MediaTek’s first AI ASIC customer is Google, with the partnership centered on TPU accelerators; the first data-center AI accelerator has been shipping in small volumes since September 2026. Driven by its AI business, MediaTek expects fourth-quarter revenue to grow 25%-32% quarter-over-quarter, with AI ASIC revenue projected to reach US$2.5 billion in Q4; the company has also set an explicit goal of capturing roughly 20% of the global AI accelerator ASIC market by 2027. From design services to volume delivery, MediaTek is turning “custom AI chips for cloud vendors” from a one-off deal into a full product line.
From an industry-landscape perspective, the continued ramp of AI ASICs marks a further deepening of specialization across the computing-power supply chain: hyperscale cloud providers are no longer content to remain mere buyers of general-purpose GPUs; instead, they are locking in cost-performance and supply security through self-designed or customized ASICs, opening incremental markets for vendors with advanced-node design capabilities such as MediaTek. As large-scale data-center chips move toward customization, the industry’s dividing line between general-purpose and purpose-built computing is becoming clearer — custom ASICs compete in the cloud, while rapidly iterating general-purpose SoCs and their developer ecosystems compete at the edge. The two tracks advance in parallel rather than in conflict.
For Junyi Microelectronics, this trend once again confirms the long-term value of edge-side computing platforms: cloud ASICs address centralized training and inference, while the massive number of industrial on-site scenarios calls for standardized edge-computing foundations built on mature SoCs. Junyi Microelectronics’ J-3568/J-3576/J-3588 core boards, built on the Rockchip platform, together with the K-series edge AI motherboards, provide equipment makers with a continuous ladder of computing power from entry level to flagship. For concentrated deployments, the AI-BOX-3588 edge computing host leverages its 6 TOPS NPU to support multi-channel video structural analysis, forming a complementary “custom in the cloud, general-purpose at the edge” landscape with customized cloud silicon.
Source: https://tech.ifeng.com/c/8xfE4yKvUxQ (October 31, 2026)