Investment bank Bernstein recently raised its price target for the semiconductor giant TSMC to $351, maintaining its 'Outperform' rating.
So, what's behind this optimism? It boils down to three key drivers. First is the powerful, ongoing demand for XPUs—a catch-all term for the specialized chips like GPUs and TPUs that power AI. Second, TSMC is aggressively expanding its CoWoS advanced packaging capacity, which is essential for assembling these powerful AI chips. Finally, and perhaps most importantly, TSMC is expanding its business into a new area: manufacturing the logic base die for HBM, the high-performance memory used alongside AI accelerators.
This isn't just a long-term story; recent events have provided strong validation for this view. The timing of Bernstein's upgrade makes sense when we look at the causal chain. First, TSMC's recent sales figures for January and February were very strong, growing nearly 30% year-over-year. This proved that the company's growth isn't just an AI-fueled fantasy; the non-AI parts of its business are also healthy, giving analysts confidence that the high growth is sustainable.
Second, major cloud service providers like Google and Amazon have announced massive spending plans for 2026, totaling over $710 billion. Much of this capital expenditure is aimed at building out their AI infrastructure, which relies heavily on custom chips like Google's TPU. Since TSMC is the primary manufacturer for these cutting-edge chips and their complex packaging, this spending directly translates into a visible and growing order book for the company.
Third, the move into HBM base dies represents a fundamental expansion of TSMC's role in the AI supply chain. Previously, its AI revenue came from making the processors. Now, by partnering with memory makers like SK Hynix and Micron to produce the logic layer of HBM, TSMC is capturing a whole new slice of the AI hardware pie. This significantly increases its total addressable market, justifying a higher valuation.
In short, Bernstein's upgraded forecast for TSMC isn't based on a single piece of news. It's a response to a powerful combination of factors: strong current performance, clear signals of massive future demand from key customers, and a strategic expansion of its business that positions it even more centrally in the AI revolution.
- XPU (Cross-Platform Processing Unit): An umbrella term for various types of processors like CPUs (Central Processing Units), GPUs (Graphics Processing Units), and TPUs (Tensor Processing Units) that are crucial for AI computations.
- CoWoS (Chip-on-Wafer-on-Substrate): An advanced 2.5D packaging technology by TSMC. It allows multiple chips, like a processor and HBM memory, to be integrated closely together on a single substrate, enabling high speed and performance for AI accelerators.
- HBM Base Die: High-Bandwidth Memory (HBM) consists of stacked memory chips on top of a logic chip called a base die. This base die controls the memory stack and connects it to the main processor. TSMC is now manufacturing this critical logic component.