Test report DSG-6867 · Rev D · tested October 10, 2026

Foundries & ManufacturingDevice under test

TSMC AI Chips Drive One-Third of Record $40.2B Q2 Revenue

TSMC posted $40.2B in Q2 2026 revenue with AI chips driving an estimated $13.31B—one-third of the total—as capex guidance climbs to $60-64B and Arizona expansion grows by $100B.

Read
4 min
Words
748
Node
65nm
Operator
Elena Vasquez

Spec summary

  1. Q2 2026 revenue: $40.2 billion, up 33.7% year-on-year
  2. AI chip revenue estimated at $13.31 billion, roughly one-third of total sales
  3. 2026 capex guidance raised to $60-64 billion, a 56.5% year-on-year increase at the high end
  4. Arizona expansion: +$100 billion for four 2nm fabs, bringing total US commitment to $265 billion
  5. 2nm node recorded its first-quarter revenue of $1.21 billion, led by AMD Venice Epyc 9006 CPUs
AI Chips Drive Around A Third Of TSMC Revenues
Fig. AAI Chips Drive Around A Third Of TSMC Revenues — AI-generated

Taiwan Semiconductor Manufacturing Co. posted record Q2 2026 revenue of $40.2 billion, with AI training and inference chips generating an estimated $13.31 billion—roughly one-third of total sales.

Net income reached $22.37 billion, up 74.7% year-on-year and equivalent to 55.6% of revenue. Wafer shipments totaled 4.34 million 12-inch equivalents, up 16.6% year-on-year and 3.9% sequentially. Revenue per wafer hit $9,271, up 14.6% year-on-year and 7.8% sequentially. Four years earlier, revenue per wafer was half that level; seven years earlier, a third.

How does TSMC size its AI forecast?

CEO and Chairman CC Wei declined to update TSMC's January projection that AI business will compound at a mid-to-high 50% annual rate from 2026 through 2030, but confirmed growth is accelerating. The trajectory is getting "stronger and stronger and stronger," he told analysts.

Wei said TSMC pulls demand signals not only from direct customers but from their biggest customers, then applies top-down judgment. Each customer reports their own truth, Wei explained, but the aggregate is not the truth.

"Now remember that I believe every customer tells me the truth – every one. You put all the truths together, it's not the truth," Wei said. "All the customers are very aggressive, right? That's the CEO's job. The CEO has got to be aggressive. So they give me the number for their demand, and I believe they try their best to tell me the truth. So I put all together, all the truths together is not a truth."

What changed in Arizona?

TSMC will add $100 billion in spending for four additional Arizona fabs producing 2 nanometer and smaller geometries, lifting total US investment to $265 billion over several years. The site is currently building out 3nm capacity, mirroring plans in Taiwan and Japan. Existing 5nm fabs are being converted to 3nm to expand output.

The 2nm node recorded its first revenue this quarter at $1.21 billion, driven primarily by AMD's "Venice" Epyc 9006 CPUs and the upcoming MI450 GPUs expected at AMD's Advancing AI 2026 event. Wei noted that fab construction takes five to seven years, so annual capex lags announced commitments.

How high is capex going?

TSMC spent $29.8 billion on capex in 2024, rising 37.4% to $40.9 billion in 2025. The 2026 range, initially $52-56 billion, has shifted to $60-64 billion. The high end implies a 56.5% year-on-year jump and would consume more than half of TSMC's $110.22 billion in cash and equivalents. Wei estimated slightly less than half of the capex increase reflects equipment and facility price inflation; the remainder funds real capacity expansion.

Roughly 70% of capex funds front-end wafer production:

  • 70%: front-end wafer production across CPUs, GPUs, DPUs, smartphone chips, and embedded devices
  • 10-20%: packaging facilities, including CoWoS and the newly ramping COUPE co-packaged optics line
  • 10-20%: specialty products

Wei said he was "pleased" Intel's EMIB-T interposer technology is gaining traction, which relieves pressure on CoWoS packaging supply.

Where is the HPC growth coming from?

TSMC's HPC segment—covering datacenter XPUs, switch ASICs, FPGAs, desktop and laptop CPUs and APUs—generated $26.53 billion, up 47.1% year-on-year and now exceeding smartphone revenue. Smartphone sales slipped 5.2% sequentially but rose 8.9% year-on-year to $8.84 billion. IoT, automotive, DCE, and other segments combined for $4.82 billion, up 23.4% year-on-year.

The author's model estimates AI training and inference chips at $13.31 billion, up 68.3% year-on-year—well above the mid-to-high 50% CAGR guidance. Non-AI HPC chips came in at roughly $13.23 billion, up 30.5%. AI thus represents about one-third of overall revenue and roughly half of HPC segment revenue. AI products are also likely the most profitable segment outside of mature processes with extremely high yields.

What does the competitive picture look like?

Intel Foundry and Samsung Foundry remain constrained for now, but Samsung has capital from the DRAM and HBM booms to expand its CPU and GPU lines. China's Semiconductor Manufacturing International Corp holds 6% of the foundry market and will grow as advanced TSMC supply stays locked out of China. Wei told investors not to hold their breath: meaningful competition will likely take another four or five years.

via The Next Platform (Source)

Filed under

  • tsmc
  • ai-chips
  • semiconductor-capex
  • arizona-fabs
  • 2nm
Share this article:

More from Elena Vasquez

Elena Vasquez

Show full bio

Senior reporter covering industry trends and analytics at Die Signal.

251 articles

Same lot · LOT-C1C6

« Previous articleNext article »