Taiwan Semiconductor ((TSM Quick QuoteTSM - Free Report) ), aka TSMC, reported another amazing quarter this morning with revenue of $33.73 billion, beating the consensus by over $400 million and representing a 25.5% jump from the year-ago quarter and +1.9% sequentially.

As both gross and operating margins increased substantially, the premier foundry for advanced chip nodes (3 and 2 nanometers) delivered stunning profits of $16.3 billion, up 35% year on year, vs the consensus of roughly $15 billion.

And management guidance was music to investors' ears as they forecast Q126 revenue to fall between $34.6 and $35.8 billion vs the prior Zacks topline consensus of $32.5 billion. As they leverage unprecedented demand, gross profit margin is expected between 63% and 65% and operating profit margin to be between 54% and 56%.

In early October before their Q3 earnings report, when TSM shares were trading below $300, I made the case for a much higher price...

Taiwan Semi is Going to $400

Today I am framing this 2025 wrap-up and 2026 preview through three "equations" that tell me TSM shares can go to $500 in 2027...

1. Pricing Power = Profit Torque: We've been hearing reports for months that TSMC was gaining new business from NVIDIA ((NVDA Quick QuoteNVDA - Free Report) ) and able to raise prices as their capacity became constrained for other customers. 

The company delivered Gross Margin of 62.3%, vs their guide of 60%. As noted, guidance for Q126 GM is even better at 65%-63%, but given the company's conservative nature we can expect that number to be challenged to the upside as well. 

While TSMC is not a pure monopoly in foundry, the demand for their services is record-breaking and allows them to leverage their capacity for higher revenues and profits.

Given their enviable position, the company is still mindful of long-term relationships and it was noted on the conference call by the CFO that “Our pricing will remain strategic, not opportunistic.”

2. Increased Investment = Strong Outlook: TSMC raised their 2026 capex projection to $52-$56B vs $40.9B for 2025, and prior expectations of $48-50B. 

This will serve bringing forward of semiconductor fabrication plant build-outs in Taiwan and new facilities in Arizona where the company already started building its third "fab" and just bought another 900 acres next to its Arizona campus. 

The planned buildout in Arizona could total $165 billion including six fabs and two advanced packaging sites, plus an R&D center. Management said it takes two to three years to build a new fab and so the new capacity won't meaningfully help supply until 2028 or 2029. "Aggressive investment reflects sustained AI-driven demand and tight supply conditions," they noted.

3. Hyperscaler Capex = Shortages to 2028: We've seen projections from Goldman Sachs and Bank of America that the AI infrastructure buildout will cross $1 trillion annually in 2028, with as much as 1/3 going for chips.

While TSMC does not typically report a specific dollar-denominated "backlog" in the same way a defense contractor might, management used extremely strong language during this morning's call to describe an effectively unprecedented order queue.

When Will Supply Meet Demand?

When pressed by analysts on when capacity will finally catch up to the insatiable AI demand, management offered a candid outlook.

The "Three Times" Demand Gap: CEO C.C. Wei stated that for advanced nodes (3nm and 5nm), current demand is roughly three times higher than their available capacity. This gap has reportedly widened over the last quarter, indicating a functional "backlog increase" as more customers (including Tier-1 CSPs like Google and Amazon) move to custom silicon.

CoWoS Bottleneck: Advanced packaging (CoWoS stands for Chip-on-Wafer-Substrate) remains the primary constraint, along with wafer supply. Management said capacity is sold out through at least mid-2026, despite doubling output year-over-year.

CoWoS is TSMC's advanced 2.5D packaging technology that integrates multiple chips (like GPUs and High Bandwidth Memory) side-by-side on a silicon interposer, which acts as a high-density connection hub before being mounted on a substrate, delivering massive bandwidth and low latency crucial for AI and High-Performance Computing (HPC) chips by reducing data travel distance.  

Equilibrium Timing: While TSMC is racing to bring new fabs online (including the record $52B–$56B Capex), they cautioned that this investment won't significantly boost output until 2027 or 2028.

Short-Term Strategy: In the interim, they are focusing on "productivity improvements" at existing fabs to squeeze out extra wafers rather than waiting for new buildings to finish.

A16 Technology Roadmap

In this morning’s Q4 earnings call, management provided specific details on the A16 (angstrom 16 = 1.6nm) roadmap and addressed the ongoing supply-demand imbalance for AI chips.

TSMC confirmed that the A16 node remains on track for volume production in the second half of 2026.

Innovation: It will be the first to feature Super Power Rail (SPR) technology, which moves power delivery to the backside of the wafer to improve efficiency.

Performance: Compared to the N2P (2nm) process, A16 is expected to deliver an 8–10% speed increase or a 15–20% power reduction at the same speed, with up to 1.10x transistor density.

Demand: CEO C.C. Wei noted that early interest for A16 from HPC and AI customers is "extremely high," with NVIDIA reportedly secured as the lead customer.

Apple vs. NVIDIA, While Broadcom Muscles In

Apple ((AAPL Quick QuoteAAPL - Free Report) ) has always been TSMC's customer numero uno, and their share is expected to rise to between 22% and 25% of TSMC sales. They are viewed as hogging 2nm node production, as they reportedly they booked over 50% of all initial 2nm capacity for the iPhone 18 (A20 chip). But they may skip A16 (1.6nm) to wait for A14.

Meanwhile NVIDIA has slowly risen to consume nearly as much capacity (~20%), with the two powerhouses combined representing almost half of TSMC sales. Jensen & Co. are the lead customer for A16 and they secured the bulk of advanced packaging (CoWoS) through 2026 for Blackwell, Rubin, and Feynman GPU systems.

Based on this morning’s Q4 2025 earnings call and recent analyst data, TSMC’s customer mix is undergoing a historic shift. For the first time, High-Performance Computing (HPC) has decisively overtaken the Smartphone segment as the primary revenue engine, as Broadcom's ((AVGO Quick QuoteAVGO - Free Report) ) slice of the pie rises to between 11% and 15% of TSMC sales.

Broadcom may be the "dark horse" now, but they have surged to the #3 spot due to massive ASIC orders (requiring 3nm and 2nm nodes) from OpenAI, Google, and Meta Platforms ((META Quick QuoteMETA - Free Report) ) who just announced the formation of Meta Compute to build hundreds of gigawatts of datacenter capacity in the next 5-10 years.

These demand trends for TSMC technology will likely persist because most of the 2nm demand has nowhere to go but TSMC as Samsung’s yields are still 20-30% behind. TSMC expects 2nm revenue to surpass the total of 3nm and 5nm by Q3 of this year.

Bottom line: In October, I told investors to buy TSM shares under $300 while they could. Now that mark is $350 as we head toward $500 in 2027.

Disclosure: I own shares of NVDA and TSM for the Zacks TAZR Trader portfolio.