While the market widely agrees that generative AI has triggered an explosion in GPU demand, TSMC (2330) Chairman and President C.C. Wei believes the next wave of AI industry growth will be driven not just by GPUs, but by 'Agentic AI'—spurring an upgrade across the entire computing architecture, with CPUs regaining a central role. During the earnings call, he stated that Agentic AI will boost demand for CPUs, GPUs, and ASICs alike, further increasing the need for computing power in AI data centers.

Over the past year, market investment in AI has largely focused on GPUs. However, Wei pointed out that the key difference between Agentic AI and previous generative AI is that Agentic AI doesn't just answer questions—it can autonomously plan, execute, and coordinate multiple tasks. This requires significantly greater system-level computing power, which is why the importance of CPUs is resurfacing.

He said, 'Agentic AI requires more CPU support, so CPU demand will increase.' In the future, AI data centers won't benefit only GPUs—entire computing platforms, including CPUs, GPUs, and various AI accelerators, will grow in tandem.

On the heated competition among current CPU architectures, Wei displayed his characteristic humor. Smiling, he said, 'Whether x86, Arm, or RISC-V, they’re all my customers.' For TSMC, there's no need to bet on which architecture will win—regardless of which ultimately dominates the market, as long as they continue using the most advanced processes, TSMC will be the primary beneficiary.

During the call, analysts asked whether future AI demand would be led by GPUs, CPUs, or custom ASICs (application-specific chips) developed by cloud providers. Wei refrained from commenting on individual customers' products, but noted that all three models currently coexist: some customers continue developing GPU platforms, others are investing in CPUs, and major cloud service providers (CSPs) are actively building their own ASICs.

He emphasized that for TSMC, this makes little difference—'They will all eventually come to TSMC.' The company is more focused on the sustained growth of overall AI computing demand, rather than shifts in market share for any single chip architecture or individual customer.

Wei explained that the demand TSMC sees today comes not only from direct customers but, more importantly, from large cloud service providers (CSPs) behind those customers. He noted that TSMC maintains long-term partnerships with its customers, giving it visibility into product roadmaps for the next several years. As a result, AI demand remains very strong, with no significant change despite market concerns about an AI investment bubble.

'Our customers are still very active, and their customers—the CSPs—are sending strong demand signals,' Wei said. This is one of the key reasons TSMC decided to raise its full-year USD revenue growth forecast to 'slightly over 40%.'

Beyond core AI compute chips, Wei also pointed out that the AI wave is gradually spreading to other semiconductor components, including power management ICs (PMIC) and CMOS image sensors (CIS), which are benefiting from demand for AI servers and related equipment. However, he acknowledged that demand for mature-node processes is not universally recovering—strength remains concentrated in AI-related applications, while the consumer electronics market has yet to fully return to its pre-pandemic growth trajectory.

Regarding concerns that a few large AI customers might take up an increasing share of TSMC’s revenue, creating customer concentration risk, Wei stated he is not worried.

He explained that while some large customers are indeed growing rapidly, the AI industry continues to see new startups and platforms entering the space. Therefore, what TSMC sees is not a few customers getting larger, but the entire AI ecosystem expanding. 'I am not concerned about customer concentration,' he said.

As for whether AI chip shortages mean TSMC will significantly raise prices in the future, Wei clearly stated the company will not suddenly hike prices due to strong demand.

He said TSMC always wants its customers to succeed and aims to fairly reflect technological value—not to gain windfall profits from short-term supply-demand imbalances. 'If we raise prices four or five times today, our customers will lose competitiveness,' Wei said. Only when customers succeed can TSMC build longer-term, more stable partnerships—this is the core of the company’s business philosophy, which enables it to maintain healthy gross margins and long-term investment capacity.

FACT BOX

  • Source: PR Times
  • Category: News
  • Products / services: CPU / GPU