MediaTek’s board approved a $5 billion discretionary financing framework on Friday to chase AI data center chips, and “discretionary” is doing a lot of work in that sentence. It’s not $5 billion spent, it’s $5 billion the company can reach for if the opportunity keeps looking as good as it does right now, stacked on top of a balance sheet that already holds more than $7 billion in cash. CEO Rick Tsai called it “optionality” on the earnings call, a careful word choice from a company that just watched its mobile chip revenue fall 20% year over year.

That contrast is the whole story, and it’s the part that makes me skeptical of the tidy version of this announcement that’s already circulating. Global smartphone shipments dropped 11% last quarter, the lowest for that period since 2013, and a memory chip shortage that’s pushed component costs up more than 90% since late 2025 is squeezing every phone maker in the chain, MediaTek included, the same crunch that killed the CMF Phone 3 Pro’s original pricing a few days ago. MediaTek itself now expects global smartphone units to shrink about 15% this year. So the company whose whole identity is built on selling mobile chips is telling investors, in the same call, that its growth story now lives somewhere else.

Somewhere else means AI accelerator ASICs for hyperscalers, and I’ve tracked MediaTek’s slow institutional pivot away from pure mobile for a while now, but this is the clearest financial commitment yet, and the confidence behind it is real, not just talk. MediaTek raised its estimate of the 2027 addressable market for custom AI chips to a flat $80 billion, up from a range that only touched $80 billion at its high end, and it raised its own target share of that market from 10-15% to 15-20%. The first custom AI chip is finished and enters production in Q4. A second design is tracking for volume production in 2028. MediaTek expects the data center AI chip business alone to clear $2 billion in revenue this year. That’s a shipping number, not a roadmap slide.

What I find more interesting than the dollar figure is why MediaTek gets a seat at this table at all. The ASIC market right now is basically Broadcom’s to lose. I’ve watched how thoroughly Broadcom has wired itself into this cycle: the $200 billion foundry commitment it just signed with Samsung, the Fort Collins fab tied to Apple’s $30 billion deal, and the fact that OpenAI decided the fastest path to its own inference silicon was to become Broadcom’s customer directly. MediaTek’s flank isn’t raw compute IP, it’s SerDes, the serializer and deserializer silicon that moves data between compute dies and HBM stacks inside a modern AI accelerator package. Everyone obsesses over the compute die. The actual bottleneck has quietly become the lanes connecting it to memory, running at 224 gigabits per lane and climbing, and that happens to be a discipline MediaTek spent two decades refining for reasons that had nothing to do with AI. It’s a genuinely clever angle of attack. I’m just not convinced it’s a $5 billion one yet.

Run the number next to what everyone else in this fight is committing and $5 billion, even fully deployed, looks less like a war chest and more like an entry fee. CFO David Ku said on the call that the money is earmarked mostly for locking down advanced packaging capacity and securing wafer allocations, which tells you MediaTek is worried about getting squeezed out of TSMC’s production calendar by bigger checkbooks, not about outspending Broadcom on R&D. That’s defensive capital dressed up as an offensive headline, and the market bought it anyway: MediaTek shares are up nearly 150% this year.

I’m not digging into what a credible MediaTek ASIC business eventually does to Nvidia’s pricing power, that’s a much bigger piece and it depends on design wins MediaTek hasn’t landed yet. What I keep coming back to is the timing. A chipmaker doesn’t raise a $5 billion discretionary fund and push its TAM math upward in the same call where mobile revenue just dropped 20% unless it’s genuinely worried about what happens if the AI pivot doesn’t land before the phone business it built its name on finishes shrinking.

Sources