By Wei Sheng-feng (魏聖峰)
MediaTek reported consolidated revenue of NT$64.18 billion for August, up 44.1% from a year earlier and 32.4% from July, as the Taiwanese chip designer's expanding push into cloud computing chips reshapes a business once defined by the boom and bust cycles of the smartphone market.
The results mark what analysts describe as one of the most consequential inflection points in MediaTek's history: a shift from a company whose fortunes rose and fell with handset demand to one increasingly anchored by long term contracts to design custom processors for the world's largest cloud computing companies.
Smart Device Platform Overtakes Phone Chips As Top Revenue Driver
MediaTek's revenue is built on three pillars: smartphone chips, a Smart Device Platform segment, and power management integrated circuits. In the second quarter of this year, the Smart Device Platform segment — which spans connectivity chips, WiFi 8, PC processors developed with Nvidia's RTX Spark platform, and application specific integrated circuits, or ASICs — overtook smartphone chips as MediaTek's single largest source of revenue for the first time.
The shift traces back to MediaTek's collaboration with Google on custom tensor processing units, or TPUs, used to train and run artificial intelligence models in data centers. Citing progress on that program, MediaTek told investors on earnings calls in April and July that it was doubling its full year ASIC revenue forecast to more than $2 billion, or roughly NT$63.2 billion.
Chips developed for the Google partnership are expected to enter mass production in the fourth quarter of this year and begin contributing meaningfully to revenue, a milestone executives have pointed to as the primary engine behind the company's push toward record margins. By reusing processor and communications intellectual property first developed for smartphones — including CPU, GPU, and NPU designs, along with low power engineering — across routers, vehicles, and data center hardware, MediaTek has been able to spread the cost of developing at advanced process nodes across a wider base of products, lifting the Smart Device Platform's overall gross margin.
Two Decades Of Andes Technology Bets Underpin The Shift
MediaTek's transformation did not happen overnight. It is the product of a two decade long, step by step investment in owning more of its own chip architecture rather than relying solely on licensed designs.
In 2005, MediaTek made a strategic investment in Andes Technology, a company focused on developing embedded processor cores. When the RISC-V Foundation was established in 2015 to promote an open, royalty free chip architecture, Andes was one of 37 founding members. That early bet has let MediaTek deploy RISC-V widely across microcontrollers and Internet of Things chips without paying the licensing fees charged for proprietary architectures, holding down its research and development costs.
Andes-licensed chips have now shipped more than 10 billion units worldwide, according to the company, giving MediaTek's broader organization deep experience in fine tuning microarchitecture and designing high performance instruction sets — expertise that underpins its ability to take on large scale ASIC projects for cloud computing customers today.
Phone Grade AI Chips Now Power Cars And Data Centers
The boom in generative artificial intelligence has created a parallel opportunity. As AI workloads strain data center capacity, network bandwidth, and user privacy expectations, chipmakers have raced to move more AI processing onto local devices rather than the cloud, a shift known as edge AI.
MediaTek's years of low power chip design for smartphones gave it an early foothold in this mid tier computing market. Its processors have evolved from early neural network accelerators into a seventh generation neural processing unit, or NPU, optimized specifically for the transformer architecture that underlies today's large language models. That hardware lets MediaTek-powered smartphones run generative AI models with tens of billions of parameters entirely offline, the company says, outperforming comparable CPU or GPU based approaches while extending battery life.
The company's NeuroPilot software platform allows developers to port AI models built for smartphones onto smart TVs, vehicles, and edge Internet of Things devices with minimal rework, extending the reach of technology first proven in handsets into an increasingly wide range of connected products.
MediaTek's High Speed Interconnect Breaks Into Broadcom And Marvell Territory
The cloud ASIC market has long been dominated by Broadcom and Marvell Technology, which built commanding leads supplying custom chips to hyperscale cloud providers. Breaking into that market required MediaTek to solve one of the hardest problems in chip design: moving massive volumes of data between chips at extremely high speed, with minimal delay and signal loss.
MediaTek says it has achieved that with SerDes, or serializer deserializer, technology capable of transmitting data at 224 gigabits to 336 gigabits per second, addressing data transmission bottlenecks that emerge when AI clusters operate at their computational limits. In its partnership on Google's TPU v8t and v9 chips, MediaTek is responsible for the demanding input output die design, working closely with Taiwan Semiconductor Manufacturing Co. on three nanometer and two nanometer process nodes and CoWoS advanced packaging technology.
This article is authorized for republication from Wealth Investment Weekly, Issue 2423. （Related： How MediaTek Reinvented Itself: The Secret Behind the Google AI Deal ｜ Latest ）
