Fueled by booming demand for artificial intelligence chips, TSMC expanded its commanding lead in the global foundry market in 2025 as most competitors ceded share. TSMC’s market share in the overall foundry business reached 69.9% in 2025, up from 64.4% in 2024, according to TrendForce, while the top ten foundries together generated around $169.5 billion in revenues in 2025, up 26.3% over 2024. In 2025, Samsung held second place with 7.2% share, followed by SMIC at 5.32%, UMC at 4.35%, GlobalFoundries at 3.87%, HuaHong Group at 2.6%, Tower at 0.89%, VIS at 0.89, Nexchip at 0.86% and PSMC at 0.8%. Intel’s emerging foundry push and Japan’s Rapidus remain outside the top tier. TrendForce described a mixed outlook for 2026, with robust artificial intelligence related demand but lingering weakness and uncertainty in mainstream, non artificial intelligence markets as rising memory prices could affect device demand and fab utilization later in the year.
TSMC’s momentum was driven by surging artificial intelligence workloads and smartphone chips. TSMC’s sales hit $122.543 billion in 2025, up 36.1% over 2024, with strong orders from customers such as AMD, Broadcom and Nvidia for artificial intelligence accelerators. In 4Q25, TSMC’s foundry share was 70.4%, down slightly from 71% in the previous quarter, as slightly lower wafer shipments were offset by demand for 3nm application processors powering the iPhone17 series. TSMC is ramping 2nm in 2026 and is positioned at the center of a shift toward custom artificial intelligence silicon. Broadcom CEO Hock Tan projects AI chip revenue exceeding $100 billion by 2027, and Broadcom reported $20 billion in AI sales in 2025, with its $100 billion roadmap spanning custom accelerators, networking silicon, DSPs, DPUs and Ethernet switching ICs. Broadcom co designs custom artificial intelligence chips with hyperscalers including Google, Meta, Anthropic, OpenAI, Fujitsu and ByteDance, fabricated at TSMC, while Nvidia continues to rely on TSMC for advanced GPUs. This move toward parallel, application specific compute stacks for large language model and inference workloads is fragmenting the artificial intelligence silicon landscape and elevating the importance of advanced packaging, networking, and deep foundry partnerships.
Other foundries are navigating a more uneven transition. Samsung suffered yield challenges on some nodes but in 4Q25 its foundry revenue (excluding System LSI) grew 6.7% QoQ to nearly $3.4 billion, lifting market share from 6.8% to 7.1% on the back of new 2nm products and logic dies for HBM4 memory, as it also secured orders from Apple, Qualcomm, Tesla and RISC V startup Ubitium. SMIC, entrenched in China’s domestic market and supplying 7nm system on a chip designs for Huawei, saw 4Q25 revenues rise 4.5% QoQ to almost $2.49 billion despite subpar but improving yields at 7nm and early work on a challenging 5nm node. UMC’s utilization held steady with 0.9% QoQ revenue growth to around $2 billion in 4Q25, and it expanded into thin film lithium niobate photonics through Wavetek and HyperLight to boost data center optical modules and artificial intelligence cluster power efficiency. GlobalFoundries posted 8.4% QoQ revenue growth to $1.8 billion in 4Q25 and is tapping demand from emerging humanoid robots, where global sales crossed the $500-million revenue mark in 2025 and China’s AGIBOT alone achieved revenue of over $140 million, while HuaHong maintained 106.1% average capacity utilization in 2025 and accelerated 12 inch fab expansion. Meanwhile, Tower climbed to seventh place on strong silicon photonics demand, ahead of VIS, Nexchip and PSMC, underscoring how specialized technologies and artificial intelligence centric workloads are redrawing the competitive map in the foundry sector.
