Mediatek: Next Winner powering AI?

The current AI boom is being driven primarily by a handful of large AI research labs and hyperscalers with the technical depth and capital to build and scale frontier models. Developing and deploying these systems requires enormous investment, advanced GPUs and accelerators for training and inference, massive data center infrastructure, high-bandwidth networking, and reliable energy supply to power these compute-intensive workloads.

As a result, a disproportionate share of capital is flowing toward the “picks and shovels” layer of the AI stack, the companies that supply the critical hardware, cloud infrastructure, semiconductors, networking equipment, and supporting technologies that make AI possible. These players sit upstream in the value chain, enabling the ecosystem regardless of which specific models or applications ultimately win.

In essence, while model level innovation captures headlines, the bulk of early economic value in this cycle is accruing to the foundational infrastructure providers that power the AI revolution.

However, true economic value of AI would come to fore when AI gets embedded in consumer devices. This is when inference happens on device near you and not solely on cloud.

Currently, using AI like ChatGPT or Claude or Gemini involves a digital “round trip.” Your request travels to a massive data center, processes on high-end hardware (GPUs or TPUs), and returns to you. While this takes only a second or two, it’s a massive delay in computing terms.

Edge AI eliminates this journey. By hosting the model directly on your device, your data never leaves your phone, and responses are generated locally and instantaneously.

"What I'm trying to say is that in 10 years time, the amount of energy necessary for artificial intelligence for most people will be minuscule, utterly minuscule. And so we'll have AI running in all kinds of things and all the time because it doesn't consume that much energy."

Jensen Huang

There is a well-known semiconductor company that already powers many of the smartphones and connected devices we use every day. It is now strategically pivoting toward AI-driven opportunities.

If executed well, it has the potential to become a meaningful player across both ends of the AI compute spectrum, designing high-performance chips for data centers that power cloud-based AI workloads, while also building efficient, on-device silicon for Edge AI applications that run locally on smartphones, PCs, and other connected devices.

In other words, it is positioning itself to participate in the full AI stack, from hyperscale cloud infrastructure to intelligent devices at the edge. The name is Mediatek

MediaTek operates as a fabless semiconductor company. This means they design the complex architecture and software of chips but outsource the actual manufacturing (printing the silicon) to foundries like TSMC.

MediaTek's chips (Systems-on-a-Chip, or SoCs) are found in roughly 2 billion devices annually. Mediatek roughly did $19B revenue in FY25, Their portfolio is divided into three main pillars:

1. Mobile Phones - 59% of revenue

This is MediaTek’s largest segment, where they design the main processors for millions of phones.

• Dimensity Series: These are high-performance chips designed for 5G smartphones. They range from “Flagship” models (like the Dimensity 9500) for the most expensive phones to more affordable versions for everyday users.

• Helio Series: These chips are built for 4G smartphones, offering a balance of performance and battery life for budget-friendly devices.

• Key Features: These chips manage everything from the camera (taking clear photos and 8K videos) and gaming (smooth visuals) to AI features like voice assistants and smart photo editing.

2. Smart Edge Platforms - 37% of revenue

This segment focuses on chips that make non-phone devices “smart” and keep them connected to the internet.

• Smart TVs: The Pentonic series powers high-end TVs (like those from Sony or Samsung), handling ultra-sharp displays and AI-enhanced picture quality.

• Connectivity (Wi-Fi/Bluetooth): The Filogic brand provides the chips for Wi-Fi routers and broadband gateways, ensuring fast and stable internet throughout a home or office.

• Computing: The Kompanio series is designed for laptops and tablets, specifically powering many Chromebooks with long battery life.

• Smart Home & IoT: These chips are found in Amazon Echo speakers, fitness equipment, and even “Internet of Things” (IoT) devices like smart appliances and industrial sensors.

• Automotive: The Dimensity Auto platform provides the technology for “smart cockpits” in cars, managing infotainment systems and driver assistance features.

3. Power IC - 5% of revenue

This is a smaller but vital segment that focuses on chips designed to manage electricity within a device.

• What they do: These chips act like a “traffic controller” for electricity, making sure every part of a device, from the screen to the processor, gets exactly the right amount of power.

• Where they are used: They are essential in everything from smartphones to electric vehicles and industrial machinery, helping to prevent overheating and extending battery life.

While all this sounds like a legacy stuff which had higher value in smartphone era of 2010s, but whats changing in Mediatek excites me.

What makes Mediatek interesting?

MediaTek is trying to establish itself as a leading innovator in the “ubiquitous AI” era, successfully pivoting from a mobile-focused company to a key player in the global AI compute backbone. Its expertise may span from the “edge” (local devices) to the “cloud” (massive data centers).

1. Cloud AI and Custom Silicon (ASIC)

MediaTek has rapidly expanded into the high-end data center market by designing custom Application-Specific Integrated Circuits (ASICs) for AI accelerators.

• Advanced Packaging & Interconnects: They utilize world-class interconnect IP and advanced packaging technologies (like 3.5D packaging and custom HBM integration) to manage the intense data and power density required by modern AI workloads.

• Google TPU Partnership: A major part of their cloud expertise is their collaboration with Google on Tensor Processing Units (TPUs), including the 3nm TPU v7 design. Mass production for this project is expected in the FY2026.

• Revenue Projections: MediaTek expects its first cloud ASIC project to contribute approximately $1 billion in revenue in 2026, scaling to multiple billions in 2027 and 2028. Management said that my 2028 cloud ASICs market could be $70B and they can get 10-15% market share in this. This suggests that ASICs revenue from Google alone can reach $10B in 3 years which is almost negligible today.

2. Edge AI Leadership

MediaTek is the top chipset supplier for Android tablets and Arm-based Chromebooks, using its expertise to bring powerful AI directly to personal devices.

• Flagship Mobile Chips: The Dimensity 9500 series features a 3nm architecture with an advanced NPU (Neural Processing Unit) optimized for generative and multimodal AI tasks.

• NPU Architecture: Their chips employ a dual-core NPU architecture, with one high-performance core for complex instructions and one efficient core for “always-on” AI experiences.

• Broad Ecosystem: Their Genio platform provides a development environment that integrates NVIDIA TAO model-training tools, enabling AI vision and voice capabilities in IoT devices, smart retail, and healthcare.

These capabilities positions Mediatek well in the market where usecases of Edge AI could explode.

3. Strategic AI Partnerships

MediaTek leverages a deep collaborative ecosystem to accelerate its technology development.

• NVIDIA Collaboration: MediaTek co-designed the GB10 Grace Blackwell Superchip for the NVIDIA DGX Spark, which is currently the world’s smallest AI supercomputer. This device can run inference on models with up to 200 billion parameters locally.

• Automotive AI: Through the Dimensity Auto platform, MediaTek and NVIDIA are developing AI-defined “smart cockpits” that feature generative AI virtual assistants and advanced driver assistance systems (ADAS).

Working with Nvidia validates the thesis that Mediatek is potent player in AI market.

4. Next-Generation Connectivity for AI

MediaTek integrates AI with advanced connectivity to support “Hybrid AI,” where tasks are split between the device and the cloud.

• AI-Enhanced Modems: Their M90 5G-Advanced modem uses built-in AI to optimize connection speeds and reliability intelligently.

• Optical Communication: The company recently mastered self-developed Micro LED light source technology for active optical cables (AOCs), aimed at solving high-speed interconnect bottlenecks within AI data centers.

Currently, MediaTek is largely valued as a cyclical mobile handset semiconductor supplier. That perception is reflected in its multiple, roughly ~25x P/E and 4–5x P/S in line with companies exposed to the volatile smartphone cycle.

In the near term, that cyclicality remains a real overhang. Smartphone production in 2026 could be constrained by elevated DRAM prices, which pressure OEM bill of materials and unit volumes. Some reports even suggest MediaTek’s handset business could decline 10–15% next year, reinforcing the market’s cyclical framing of the company. This is key near term risk as Mediatek’s customers cannot produce enough volumes when their raw materials are at sky high prices.

However, taking a longer-term perspective, the mix of MediaTek’s incremental growth is beginning to resemble companies like Broadcom and Marvell Technology, which trade at materially higher multiples (10–15x P/S and 60–70x P/E). These firms have successfully pivoted toward high-value custom ASIC programs, partnering with hyperscalers such as Google and Amazon to design AI-focused silicon.

MediaTek now appears to be emerging as a credible entrant in this custom ASIC ecosystem. The core thesis is that MediaTek could potentially secure up to $10B in annual Google ASIC revenue within the next three years. Meanwhile, its existing ~$19B business, even if it compounds at a modest 5–6% CAGR, would provide a stable base. Under this scenario, total revenue could scale toward ~$32B, with operating leverage driving disproportionate growth in earnings.

Importantly, additional ASIC design wins with players such as Amazon or Microsoft would represent pure upside optionality, not embedded in the base case, but capable of meaningfully expanding both revenue and valuation multiples.

In essence, the debate is whether MediaTek remains a smartphone cyclical or transitions into a structural AI infrastructure enabler with a fundamentally different earnings profile and valuation framework. With some valuations comfort, this looks like decent risk-reward bet.