There is a theory in Taiwan's semiconductor industry, attributed to MediaTek Chairman Tsai Ming-kai (蔡明介), that goes something like this: the chip design business is brutally like boxing. You can knock out every opponent, hold the belt high, and still find yourself flat on the canvas the moment a hungrier challenger steps into the ring. The only way to survive is not to win one generation — it is to win three.
In April 2026, at Google's annual Cloud Next conference, that theory got its most compelling proof yet. Google announced that its eighth-generation tensor processing unit — the engine powering some of the most advanced AI systems on the planet — would be split across two dedicated chips for the first time. One chip handles training. The other handles inference. Broadcom designed the training chip. MediaTek (聯發科) designed the inference chip.
For anyone tracking the global AI hardware race, the announcement landed like a statement of arrival. MediaTek, a company most consumers know — if they know it at all — as the maker of the chip inside their budget Android phone, had just been handed a seat at the table alongside the most powerful technology companies in the world.
It did not happen by accident.
How MediaTek Conquered the CD-ROM Era — Then Walked Away
To understand how MediaTek got here, it helps to go back to the late 1990s, when personal computers were just discovering multimedia and every household in Asia seemed to be buying a new CD-ROM drive. MediaTek was a small, relatively unknown company in Hsinchu, Taiwan's technology heartland. Its engineers had developed chipsets for optical disc drives that were cheaper, more integrated, and easier to manufacture than anything else on the market.
The company did not just compete in that space — it dominated it. Revenue poured in. The industry took notice. By any conventional measure, it was time to consolidate, defend the position, and enjoy the returns.
Tsai Ming-kai did the opposite. Even as the disc drive business was still generating its best numbers, he quietly began redirecting engineering resources toward the next opportunity. When solid-state storage eventually made optical drives obsolete — a transition that wiped out several competitors who had stayed too long — MediaTek had already moved on. First-generation belt secured. Crisis averted.
The lesson Tsai Ming-kai drew from that experience would shape everything that followed. Dominance, he concluded, is a warning sign as much as a reward. The moment you are winning most comfortably is precisely the moment to start worrying about what comes next.
The Bet on Budget Smartphones That Built a 5G Powerhouse
The smartphone era gave Tsai Ming-kai the chance to test that philosophy on a much larger stage. Around 2005, as mobile phones were beginning their transformation from communication devices into pocket computers, MediaTek started building chips for white-label handsets — the anonymous, unbranded devices flooding markets across China and Southeast Asia.
The move attracted ridicule. MediaTek, critics said, was a company for counterfeit phones. It had no place among the premium players. Tsai Ming-kai responded with a phrase that became something of a motto in Taiwan's tech circles: "Today's copycat, tomorrow's mainstream." He was not defending imitation. He was describing a pattern — how emerging markets absorb technology from the bottom up, how volume creates capability, and how capability eventually produces quality.
He was right. MediaTek's turnkey chip solutions — packages that gave manufacturers everything they needed to build a functional smartphone quickly and cheaply — swept through the mid-to-low-end Android market. The cash that flooded in was immediately reinvested into premium development. By the early 2020s, MediaTek's Dimensity chipset series had become the dominant platform for flagship Android smartphones outside Apple's ecosystem, holding more than 30% market share. Customers including Xiaomi, OPPO, vivo, Samsung, and Transsion all came to depend on it.
In 2025, MediaTek's smartphone revenue exceeded $10 billion for the first time. Full-year revenue reached approximately NT$596 billion (approximately $18.3 billion), up 12.3% year-on-year. From a maker of cheap components for knockoff phones, the company had transformed into the world's leading supplier of 5G mobile processors. Second-generation belt secured.
From Mobile Chips to Google's AI Infrastructure
By 2026, the belt was starting to feel heavy in a familiar way. Global smartphone shipments were forecast to decline. Memory costs were surging. The mobile market that had carried MediaTek through its second transformation was showing the same early signs of fatigue that the disc drive business had shown a generation earlier.
Tsai Ming-kai did not wait. In February, he made his position clear in a public statement that was characteristically direct: "AI is still very hot. There are many opportunities. We must keep investing."
The company's AI strategy runs along two parallel tracks. The first and more immediately consequential is in cloud infrastructure. The Google TPU partnership is the centrepiece, but MediaTek is also working with Nvidia to develop custom chips compatible with the NVLink Fusion architecture — the interconnect fabric that links Nvidia's most powerful AI processors. The company's management has expressed confidence that data centre ASIC revenue will surpass $1 billion in 2026, with projections for 2027 running considerably higher. Some foreign analysts have estimated that figure could approach $12 billion by then, representing close to 40% of total revenue.
To staff the push, Tsai Ming-kai took the striking step of redeploying hundreds of engineers from MediaTek's mobile division into ASIC development — a signal, internally and externally, that this is not a side project. The company has also secured manufacturing capacity at TSMC for 2-nanometer fabrication and CoWoS advanced packaging, positioning itself for high-volume production in the fourth quarter.
The second track is edge AI — the territory where MediaTek has long been strongest and where it is now trying to move upmarket. The latest Dimensity 9500 and 9600 chipsets carry neural processing units capable of running autonomous AI agent applications directly on the device. An automotive intelligent cockpit platform is scheduled for mass production before the end of the year. The company's Smart Edge Platforms division, which spans Wi-Fi chips, tablets, IoT devices, and television processors, posted 80% revenue growth last year and is expected to continue expanding.
Together, the two tracks form what MediaTek has begun calling an edge-to-cloud continuum — a vision in which the company provides the silicon for AI computing at every point in the chain, from the device in your hand to the data centre running the model behind it.
The TSMC Veteran Who Transformed MediaTek's Operations
Executing that vision requires more than strategy. It requires operational capability of a different order — the kind built over decades managing the most demanding supply chains and fabrication processes in the world.
In 2017, Tsai Ming-kai made what many now regard as his shrewdest move: recruiting Rick Tsai (蔡力行), the former president and chief executive of TSMC, to join MediaTek as co-CEO. The two men share a surname but little else in background. Where Tsai Ming-kai is the architect — the long-range strategist who identified three successive waves of industry transformation and positioned the company to ride each one — Rick Tsai is the builder, bringing to MediaTek the operational precision and manufacturing relationships he spent decades developing at the world's most important chipmaker.
Under Rick Tsai's leadership, MediaTek has accelerated the integration of a string of acquisitions that Tsai Ming-kai had assembled over the previous decade and a half: networking chip specialist Ralink, acquired in 2011; display chip maker MStar, in 2012; and analog and power management company Richtek, in 2015. Those purchases had given MediaTek an unusually broad technology portfolio. Translating that breadth into coherent products and margin discipline required a different kind of leadership — and Rick Tsai delivered it.
At the ISSCC conference earlier this year, Rick Tsai put the competitive stakes plainly. In AI, he argued, the race is not won by whoever has the most raw compute. It is won by whoever delivers results at the lowest energy consumption and the lowest cost. That framing — efficiency over brute force — is precisely the argument for inference-optimised ASICs over general-purpose GPUs, and it is the logic underpinning MediaTek's entire third-generation bet.
MediaTek Stock Hits Record High as Investors Back AI Strategy
On April 24, MediaTek's share price hit a record NT$2,435, up roughly 70% from NT$1,430 just two weeks earlier. Market capitalisation reached approximately NT$3.89 trillion (approximately $120 billion) — a new high. Foreign institutional investors have been buying heavily, with price targets revised as high as NT$2,588. Multiple brokerages have upgraded the stock to a top pick.
The numbers reflect something beyond routine optimism about a single product cycle. They reflect a growing belief that MediaTek has genuinely pulled off what most chip companies fail to do even once: a complete reinvention, executed from a position of strength rather than desperation.
Tsai Ming-kai has a line he returns to when asked about career planning. "Don't plan your career," he says. "Just be adaptive."
For a company now designing AI inference chips for Google while simultaneously powering the smartphone in your pocket and developing the processor for next year's intelligent car cockpit, the advice appears to have been well taken.
MediaTek's first-quarter earnings call is scheduled for April 30. Analysts will be listening closely for any update on AI ASIC order visibility. The third round, by all indications, is just beginning.
