While NVIDIA has long been the center of attention in the artificial intelligence investment boom, a fund manager overseeing nearly $10 billion in assets has a different view: Taiwan Semiconductor Manufacturing is the indispensable core of the current AI ecosystem.

Jonathan Cofsky, co-portfolio manager of the $9.35 billion Janus Henderson Global Technology and Innovation Fund, stated that TSMC is his fund's largest holding, surpassing NVIDIA. He described it as potentially the most important company driving AI development today. In his view, regardless of which chip designer ultimately wins, all leading-edge chips are manufactured at TSMC. This irreplaceable position makes it the ultimate beneficiary of AI computing power expansion.

Cofsky's fund has outperformed the S&P 500 on both three- and ten-year annualized returns. He also noted a fundamental difference between the current AI infrastructure boom and the early-2000s dot-com bubble: the physical constraints of real resources and electricity will effectively curb disorderly capital expansion, providing support for this growth cycle.

Taiwan Semiconductor Manufacturing (NYSE: TSM)

Cofsky distilled TSMC's core value into a simple logic: investors need only answer one question—"How much computing power will be built in the future?" The answer is independent of any specific chip designer, "because everyone manufactures at TSMC."

TSMC is the world's only pure-play semiconductor foundry, holding the most advanced process technologies. Cofsky emphasized that no matter the form of AI computing, TSMC is an unavoidable production node, granting it a unique structural advantage within the entire AI supply chain.

ARM Holdings (NASDAQ: ARM)

Beyond TSMC, Cofsky revealed his fund increased its stake in chip architecture designer Arm Holdings earlier this year. The rationale is that AI agentic workflows will rely far more heavily on CPUs than previous AI iterations, and Arm possesses the "broadest architecture penetration."

He used an analogy to explain the GPU-CPU division of labor: GPUs are like an "army of workers" executing tasks, while CPUs are the "managers" responsible for coordination. AI agents involve extensive "orchestration and combination," which plays to Arm's strengths.

Cofsky also pointed out that Arm is transitioning from a pure IP licensing model to designing its own chips, with Meta as its first customer. "By the end of the decade, how successful Arm is as a chipmaker will be a key variable to watch."

Software Stocks: A Challenging Transition

Commenting on the recent rebound in software stocks, Cofsky remained cautious. He believes the software industry faces disruptive risks from AI, as AI fundamentally automates labor.

He cited historical patterns, noting that in each disruptive long-term trend—such as retail shifting from offline to online platforms like Amazon—the market typically takes four to six years to separate winners from losers. Ultimately, only 10% to 20% of the original leading companies successfully complete the transition.

Consequently, his fund focuses on companies that are "extremely difficult to replicate and can layer AI as an incremental revenue stream." This includes Datadog (NASDAQ: DDOG), Snowflake (NYSE: SNOW), and theory/design software firms with unique data assets. "Snowflake and Datadog are already seeing revenue re-acceleration from AI, which is the most positive signal we can see right now," he said, adding, "But we need to see if this momentum is sustainable."

DoorDash, Inc. (NYSE: DASH)

Cofsky also highlighted a currently less-popular stock: food delivery platform DoorDash. He argued that current market concerns focus on its ongoing investment cycle and macroeconomic pressures on consumers, but these factors are overpriced.

In his view, DoorDash is a founder-led company actively expanding into new businesses like international grocery delivery. It is also positioned to benefit from both generative AI and physical AI developments, giving it value for medium- to long-term investment.