Taiwan Semiconductor ((TSM Quick QuoteTSM - Free Report) ) is the premier chip fabrication company for the world's most advanced digital technologies, from Apple smartphones to NVIDIA GPUs.

In less than five years, TSMC's trailing 12-month revenue has more than doubled to hit $95 billion as of the March quarter. Clearly, this is on the back of NVIDIA's ((NVDA Quick QuoteNVDA - Free Report) ) explosive growth which saw sales grow by ten times in the same period to $148.5 billion as of the April quarter, on its way to nearly $200 billion this fiscal year (ends January).

Of course, demand for NVIDIA GPU systems didn't turn into a hockey stick until its "ChatGPT moment" in the spring of 2023.

I mark that moment in late May at Computex in Taipei City, Taiwan when Jensen Huang unveiled the DGX Grace Hopper 200 system and said that Microsoft, Google, and Meta Platforms ((META Quick QuoteMETA - Free Report) ) were already in line.

To me, this meant that the other 996 corporations in the Fortune 1,000 would want (or need) to build their own LLMs too. I then published this article and video to explain what was coming...

Nvidia DGX: Workhorse of AI Will Drive NVDA to $2 Trillion

And that $2 trillion market cap came quickly. That's when the "AI bubble" talk started. You can imagine how astonished the bubble breathers are now that NVDA is approaching a $4 trillion market cap.

"Honey, I Shrunk the Chips!"

While Apple's business has been significant for TSMC (as much as 25% of revenues), the iPhone maker's sales growth has stalled in the low single digits.

What still makes Apple very important though is how they help drive innovation at TSMC by always demanding the latest and greatest architectures for their ecosystem of devices.

TSMC supplies chips for flagship smartphones and 5G infrastructure, where efficiency and miniaturization are crucial. The company’s 5nm and 3nm transistor nodes are widely adopted in high-end mobile devices and 5G equipment, supporting the global rollout of next-generation connectivity.

Transistor nodes refer to the specific manufacturing processes used to produce integrated circuits, where the "nm" (nanometer = 1 billionth of a meter) like 7nm or 5nm, historically represented a physical dimension of the transistor, specifically the gate length. For reference, the coronavirus is about 50nm.

Now it's more of a marketing term that signifies a new generation of chip technology with improved performance, power efficiency, and transistor density.

But generally, a smaller node number means...

Smaller transistors: More transistors can be packed into the same area on a chip.

Faster switching: Electrons have a shorter distance to travel, leading to higher clock speeds.

Lower power consumption: Less voltage is needed to operate the smaller transistors.

Why the AI Economy is Not a Bubble

Which brings us back to NVIDIA as they now pack 208 billion transistors into the Grace Blackwell "superchip" with further density on the horizon in the Rubin and Feynman series of GPUs expected in 2026 and 2027, following Jensen's nearly annual cadence of development.

I mentioned the early demand for NVIDIA systems in 2023 and have recently seen similar projections from Goldman Sachs, Bank of America, and Marvell ((MRVL Quick QuoteMRVL - Free Report) ) at their Custom AI Investor Event that total datacenter capex by the big "hyperscalers" (Amazon, Microsoft, Google, Oracle) plus Meta, Tesla ((TSLA Quick QuoteTSLA - Free Report) ), OpenAI, Apple, and xAI, will grow from nearly $600 billion this year to over $1 trillion in 2028.

This demand is not slowing down because all these companies realize that the transition to an "AI economy" is built on the infrastructure of NVIDIA GPU systems. I am on record late last year saying that this megatrend would lead to over $500 billion in revenues for NVIDIA by 2030. That's only ~35% CAGR from when the TTM topline crossed $100 billion.

And that's the infrastructure to power not just generative and agentic AI systems (i.e., LLMs), but a whole new world of "physical AI" for autonomous machines, including self-driving cars, humanoid robots, and many other types of 3D automation requiring "embodied" or edge intelligence and vision/spatial perception.

And of course my favorite use cases for these technologies are science and medicine, where researchers can plumb the depths of our existence and come back with wondrous and life-changing knowledge.

The training alone for these worlds requires enormous datasets and compute, using simulation and synthetic models. In this way, Jensen is forging a new universe of possibilities that keep TSMC "fabs" humming.

TSMC is the Backbone of the AI Economy

TSMC holds about 67% of the global foundry market and nearly 90% of advanced chip production, making it the preferred manufacturer for high-performance, custom AI accelerators and GPUs.

While NVIDIA designs the magical GPUs and systems that power the innovations of the AI Economy, it all starts with Taiwan Semi's advanced silicon fabrication. This is the epitome of high-tech manufacturing using lithographic etching to create circuits smaller than the eye can see.

And what's better for crafting new chips than brand new "fabs" as TSMC just broke ground on their 3rd facility in Phoenix, Arizona. In late June, Softbank's Masayoshi Son announced new plans to invest in a $1 trillion industrial complex in Arizona that will build robots and AI technologies.

Son is seeking to create a version of China's vast manufacturing hub of Shenzhen, which concentrates high-tech companies for proximity of resources, logistics, R&D, collaboration, and talent. TSMC was named as a key partner in Son's plans.

Currently for TSMC, the 5nm and 3nm nodes are the backbone for state-of-the-art AI accelerators, powering chips from NVIDIA, AMD, and Apple that are used in generative AI, machine learning, and high-performance computing (HPC) workloads.

In Q1 2025, 58% of TSMC's wafer revenue came from 3nm and 5nm nodes, with 73% from 7nm and below, underscoring the centrality of these advanced processes in TSMC's business and the broader tech ecosystem.

But TSMC wouldn't be itself if it wasn't planning to go deeper. The company plans to begin production of its 2nm node by late 2025, promising significant gains in power efficiency and performance. Looking further ahead, mass production of the 1.4nm (14 angstroms) process is slated for 2028, offering 10–15% higher speeds or 25–30% lower energy use compared to 2nm.

TSMC Earnings On Deck July 17

The reason that TSM is a Zacks #1 Rank (Strong Buy) currently is that analysts have been raising their EPS estimates lately as the already-fierce pace of the AI buildout accelerates.

In the past two months, the Zacks consensus for 2025 has risen from $9.00 to $9.28, representing 32% annual growth, while sales are projected to top $117 billion for a 30% advance. Analysts and investors usually have a good read on their quarterly estimates since TSMC issues a monthly sales report. June numbers are due July 10.

Next year's EPS estimate moved up 5% from $10.34 to $10.89, anticipating profit growth of over 17%, matching the topline advance as it is projected to ascend to $137 billion.

Incidentally, while we have held NVDA shares in my TAZR Trader portfolio since late 2022 from $12.50, we were able to take advantage of a big buying opportunity during the "tariff tantrum" in April and scooped TSM shares at $145.

The fact that TSM hit new highs over $235 last week confirms that large investors are still betting on the AI revolution as being the difference maker in a global economy where trade wars, inflation wars, and even real wars don't matter as much.

Kevin Cook is a Senior Stock Strategist for Zacks Investment Research where he runs the TAZR Trader portfolio service and keeps a close eye on all-things AI.