MediaTek has doubled its 2026 AI chip revenue target to $2 billion and is gunning for hyperscaler custom silicon business that Broadcom has dominated for years, as investors rotate out of TSMC and into the Taiwanese chipmaker at a pace not seen in over a decade.
The numbers are hard to dismiss. MediaTek's stock has gained roughly 150% year-to-date, outperforming TSMC by the widest margin since 2009, and its April announcement doubling the 2026 AI ASIC revenue target from $1 billion to $2 billion landed with real force. But the more interesting question is not whether MediaTek has had a good year. It is whether the company, long identified as a budget-tier smartphone chip supplier trailing Qualcomm, has genuinely earned a seat at the table where Broadcom and Marvell have held court for years.
The primary evidence that it has: Google. MediaTek's first major AI accelerator program for a U.S. hyperscaler, widely understood to sit within Google's TPU development pipeline, is on schedule, with the company now targeting $2 billion in AI ASIC revenue for 2026. A second AI accelerator project is already in development, aimed at volume production by end-2027. At the company's 2026 analyst day, management described multi-year order visibility for its data center line and laid out a target of 10 to 15 percent of what it estimates will be a $70 to $80 billion AI ASIC total addressable market by 2027. That is not a roadmap slide. That is a commitment.
Broadcom's position in this market makes clear what MediaTek is actually up against. As CNBC's analysis of Broadcom's latest earnings made clear, the company reported $10.8 billion in AI semiconductor revenue in Q2 fiscal 2026 after posting $8.4 billion in the prior quarter, and together with Marvell controls an estimated 95% of the custom AI ASIC co-design market. Those relationships run deep. Broadcom has co-designed Google's TPUs for years and holds Meta's MTIA chips as another flagship program. More than brand loyalty, Broadcom and Marvell own the networking-to-compute interconnect fabric installed in many hyperscaler data centers, which means switching partners is not a simple procurement call. It is an architectural overhaul.
Still, Google is doing something deliberate. Reports suggest the company is introducing MediaTek into its supply chain not to replace Broadcom but to use it as competitive pressure on pricing. That is a real foothold. Even playing second fiddle in a Google program at $2 billion in annual revenue is a meaningful business, and Google's willingness to run parallel development tracks tells you something about how much leverage Broadcom currently enjoys in contract negotiations. MediaTek's cost structure, honed over years of competing in price-sensitive smartphone markets, may be precisely what hyperscalers want when they push into inference workloads that do not demand Broadcom's most sophisticated training silicon. Custom AI chip shipments are forecast to grow at triple the rate of GPU shipments in 2026, and the inference tier is where that growth concentrates.
The RTX Spark announcement at Computex 2026 reinforces MediaTek's credentials. Nvidia's new AI PC chip, unveiled June 1, pairs a Blackwell GPU with a MediaTek-linked Arm CPU design on a single TSMC 3nm system-on-chip, with 70 billion transistors, 6,144 CUDA cores, and up to 128GB of unified memory running at 300GB per second. Devices from Microsoft, Dell, HP, Lenovo, ASUS, and MSI are set to debut this fall. Having Nvidia lean on MediaTek's CPU and system-on-chip expertise for its first serious push into the Windows PC market is not an accidental endorsement. It means MediaTek is credentialed at the leading edge of silicon design, co-producing chips with the dominant AI hardware company on a platform targeting mass deployment.
The capital rotation behind the stock move
Bloomberg flagged in May that TSMC's weighting in global semiconductor indices and single-stock concentration limits at major funds has made continued accumulation structurally difficult. As AI enthusiasm has broadened beyond Nvidia and its foundry relationship with TSMC, concentration-limited funds are actively seeking other places to deploy capital in the semiconductor supply chain. MediaTek's share of that overflow is not purely speculative; the AI ASIC revenue ramp gives it fundamental support that many AI-adjacent stocks lack. That said, the 150% year-to-date gain already prices in substantial execution, and a Q1 2026 profit decline of 18% year-over-year is a reminder that the smartphone business, still the core of MediaTek's revenue, is under real pressure.
The challenge ahead is credibility at scale. Broadcom's data center moat is structural, not reputational, and no analyst day optimism changes the interconnect fabric already installed inside Google's, Meta's, and Microsoft's infrastructure. MediaTek's realistic path is not displacement but addition: capturing inference-tier programs where cost efficiency matters more than peak training performance, and using the Nvidia relationship to demonstrate manufacturing depth at 3nm. If the company hits its $2 billion 2026 target and converts its second hyperscaler program into a confirmed order, the debate about whether this is a real AI infrastructure company or a smartphone vendor with a good story will resolve itself. Until then, the stock has done the talking.
Also read: Palantir's NHS battle reveals the political risk enterprise AI investors ignore • Software Stocks Are Posting Record Earnings While Investors Price In Their Obsolescence • LG Electronics is bringing its $65 billion hardware business to Arbitrum to build an advertising platform that puts the entire ad supply chain on-chain
