TSMC Forecasts AI-Driven Record Profit
TSMC is projecting a record-breaking earnings per share of over NT$100 in 2026, driven by a massive wave of AI chip orders. The foundry expects revenue from AI accelerators, sourced from clients like Nvidia, AMD, and hyperscalers, to grow 54-59%. This forecast provides a strong indicator of the sustained, high-volume demand for advanced AI silicon.
The foundry's future is tied to its most advanced nodes; 3nm and 5nm process technologies already account for 60% of wafer revenue. High-Performance Computing (HPC) is now TSMC's largest platform, making up 57% of sales, a significant shift from 2020 when smartphones dominated. This AI-driven demand is creating a bottleneck in advanced packaging. TSMC is racing to expand its Chip-on-Wafer-on-Substrate (CoWoS) capacity, projecting it will reach 100,000 wafers per month by the end of 2026. Still, this may not be enough, forcing some ASIC vendors and chipmakers to explore alternatives like Intel's Foveros and EMIB packaging technologies. Nvidia's Blackwell B200 and AMD's Instinct MI300 series are central to this demand wave, both relying on TSMC for manufacturing. The architectural race is intensifying, with Blackwell boasting 208 billion transistors and a second-generation Transformer Engine, while AMD's CDNA 3 architecture integrates CPU and GPU chiplets for AI and HPC workloads. Hyperscalers are increasingly designing their own custom silicon to optimize workloads and reduce dependency on third-party vendors, representing a major competitive dynamic. Google's TPU, Amazon's Trainium, and Meta's MTIA are all vying for CoWoS capacity alongside merchant silicon providers, creating a complex build-versus-buy ecosystem. The competitive landscape extends beyond TSMC, with Samsung Foundry gaining traction. Samsung is reportedly producing chips for Tesla and is in talks with ByteDance for its "SeedChip" AI processor. Intel Foundry Services is also positioning itself as a viable alternative, aiming to compete with its 18A process node. As 3nm capacity is fully booked, TSMC is strategically guiding customers to its 2nm node, which is on track for mass production in the second half of 2025. This next generation of chips, featuring nanosheet transistor architecture, has reportedly secured around 15 customers, with 10 focused on high-performance computing. Venture capital is pouring into the AI hardware space, with AI-related startups raising a record $171 billion in February 2026. Funding for AI chip startups alone surged from $822 million in 2022 to over $5.2 billion in 2025, signaling a robust ecosystem of new players challenging incumbents.
