TSMC booked NT$1.494 trillion in third-quarter revenue — the Taiwan foundry's highest quarterly figure on record, driven by what a headline report called an "AI boom."

The figure, attributed to Taiwan Semiconductor Manufacturing Company and posted by tech-insider.org, converts to roughly US$48 billion at current exchange rates. TSMC, the world's largest contract chipmaker by revenue, reports in New Taiwan dollars and operates wafer fabs across Taiwan, the United States, Japan, and Germany.

NT$1.494 trillion lands as another notch in a string of record quarters. The foundry has set or matched quarterly highs every period since the generative AI buildout accelerated in late 2022. Each notch up has tracked demand for accelerator silicon and the high-bandwidth memory wrapped around it.

What does the headline actually say?

Only the top line. The disclosure carries revenue and the AI demand framing. Missing from the report:

• Revenue split by node
• Revenue split by platform
• Revenue split by region
• Capital expenditure guidance
• Utilization rates
• Fourth-quarter outlook

Those numbers typically arrive hours later, in TSMC's formal earnings release and management conference call. Watch the node breakdown — 5nm and 3nm have carried the bulk of incremental AI demand for two years running.

What's the broader picture?

TSMC manufactures chips for fabless designers. The customer roster reads like a directory of the semiconductor industry's most valuable names: Nvidia, AMD, Apple, Broadcom, Qualcomm, MediaTek, Sony.

The AI accelerator generation driving the surge runs primarily through TSMC's 5nm and 3nm process nodes. The company's CoWoS line stitches GPU dies together with high-bandwidth-memory stacks. CoWoS has been the binding constraint on accelerator shipments since 2023.

The accelerator customers inside that build include:

• Nvidia's H100, H200, and Blackwell B100/B200 families
• AMD's MI300 and MI325 series
• Google's TPU line
• Broadcom's custom ASIC work for hyperscalers

Capex has stepped up repeatedly over the past 18 months to expand wafer-fab and packaging output. Diversification has run in parallel:

• Arizona, now producing in 4nm, backed by CHIPS Act funding
• Kumamoto, Japan, with Sony and Denso as anchor customers
• Dresden, Germany, with Bosch, Infineon, and NXP

What's the supply-chain angle?

The AI buildout reshaped foundry order patterns starting in late 2022. Hyperscalers — Microsoft, Google, Meta, Amazon — poured capital into accelerator procurement for generative AI training and inference. That demand pulled foundry utilization above 90% on leading-edge nodes and lifted blended average selling prices across TSMC's portfolio.

Second-order effects rippled through the supply chain:

• HBM suppliers (SK Hynix, Samsung, Micron) ran tight
• Substrate and ABF material vendors saw order lead times multiply
• Advanced packaging equipment makers booked multi-quarter backlogs

TSMC's NT$1.494 trillion is the headline. The second-order story is the tightness across every node that touches AI silicon.

What to watch

TSMC's fourth-quarter guidance. If management projects another sequential increase, calendar 2026 closes as the foundry's strongest year on record. The headline points there. The next data point will tell the market whether the run-rate sustains — or whether inventory digestion at the largest AI customers has begun to bite.