TSMC manufactures chips for Apple, NVIDIA, AMD, Qualcomm, and virtually every fabless semiconductor company - it makes 90%+ of the world's most advanced chips. Its 3nm and 2nm processes are years ahead of any competitor. No advanced AI chip, iPhone, or modern server can exist without TSMC. Taiwan's geopolitical significance is entirely intertwined with TSMC's irreplaceable role in global technology.
Key relationships in depth
TSMC has manufactured every Apple processor since 2012, when Apple moved away from Samsung Foundry amid their patent dispute. TSMC's most advanced 3nm nodes are reserved around Apple's launch cadence, with the A18 Pro in iPhone 16 and the M4 chip in the latest Mac lineup built exclusively at TSMC fabs. Apple is TSMC's single largest customer by revenue.
Why it matters: Apple's silicon advantage depends entirely on TSMC executing its roadmap. The custom silicon strategy, which started with the A4 chip in 2010, lets Apple optimize hardware and software in ways competitors using off-the-shelf chips cannot match. The relationship gives Apple roughly 18 to 24 months of process node exclusivity before TSMC opens that node to other clients.
TSMC manufactures Microsoft's Maia 200 AI inference accelerator on its 3nm process node, making TSMC the exclusive foundry for Microsoft's custom silicon strategy. Maia 200 (deployed January 27, 2026) contains 140 billion transistors, 216GB of HBM3e memory, and delivers 3x the FP4 performance of Amazon's Trainium3 for inference workloads. Microsoft also relies on TSMC for manufacturing Cobalt ARM server CPUs that power a growing share of Azure's internal compute infrastructure.
Why it matters: Microsoft's development of proprietary AI chips reduces its NVIDIA dependency and allows it to optimize inference economics specifically for Azure OpenAI Service workloads, with Maia 200 reportedly cutting AI token costs by 30% versus equivalent NVIDIA GPU configurations. TSMC benefits from Microsoft's multi-generation chip development roadmap, which includes Maia 300 (planned 2027) on TSMC's 2nm or N2P process.
TSMC manufactures every NVIDIA GPU on its most advanced process nodes: H100 on 4nm, H200 on 4nm, and Blackwell B200 on 4nm CoWoS (Chip-on-Wafer-on-Substrate) advanced packaging. TSMC is NVIDIA's sole manufacturing partner for data center GPUs, and no other foundry can produce chips of this complexity at meaningful volume.
Why it matters: NVIDIA is a fabless chip designer that relies entirely on TSMC for the hardware generating $215B+ in annual revenue. Without TSMC's 4nm and forthcoming 2nm processes, NVIDIA cannot produce GPUs that maintain performance leadership. This relationship is the backbone of the global AI hardware supply chain.
TSMC manufactures Google's Tensor Processing Units (TPUs), with Google's 7th-generation Ironwood TPU fabricated on TSMC's advanced process nodes. Ironwood TPUs deliver dramatically higher AI compute density than previous generations and are deployed across Google's AI infrastructure for training Gemini models and serving inference at scale. TSMC is the sole foundry capable of manufacturing Google's most advanced TPU designs at leading-edge nodes.
Why it matters: Google's AI competitiveness depends on having the most capable TPUs available, and TSMC's leading-edge process technology is the only path to achieving the transistor density and power efficiency that Google's next-generation AI training infrastructure requires. TSMC benefits from Google's growing TPU investment, with Anthropic's $10B+ deal to use Google TPUs creating additional cloud-mediated demand for TSMC's advanced process capacity.
Relationships are compiled by VexMap from company filings, announcements and reporting, and last reviewed September 2026. Spotted something wrong? Tell us. Not investment advice.