Summary: The quarter TSMC reported on January 15, 2026 was the best in company history. Stack the 2026 capital expenditure guidance of US$52–56 billion on top, and the takeaway writes itself — the AI demand everyone talks about exists as signed purchase orders, not slideware. Revenue for the final three months of 2025 came to NT$1.046 trillion (US$33.73 billion), 20.5% more than a year earlier — no quarter had ever carried the company past NT$1 trillion before.
The quarter TSMC reported on January 15, 2026 was the best in company history. Revenue for the final three months of 2025 came to NT$1.046 trillion (US$33.73 billion), 20.5% more than a year earlier — no quarter had ever carried the company past NT$1 trillion before. Net income climbed 35% to NT$505.74 billion (about US$16.05 billion), profit having now grown year over year for eight quarters in a row, another all-time high. Measured against the analyst estimates compiled by LSEG, both figures came out ahead.
Anyone looking for evidence of the AI buildout that cannot be manufactured should read TSMC’s books: a data center GPU, an accelerator, any AI-capable device — each winds up as an advanced-node wafer order. Stack the 2026 capital expenditure guidance of US$52–56 billion on top, and the takeaway writes itself — the AI demand everyone talks about exists as signed purchase orders, not slideware.
The NT$1 Trillion Quarter
Size wasn’t the quarter’s only strength. Gross margin landed at 62.3%, topping a 60.5% to 62.5% guidance band the company had set for itself — pricing power that exists because so few foundries anywhere can mass-produce leading-edge AI silicon. In a note published after the results, JPMorgan argued that US AI demand will keep working in TSMC’s favor and pull the company further ahead of rivals like Samsung and Intel. On the earnings call, CFO Wendell Huang said AI chip demand keeps supply tight, with visibility continuing to improve.
The guidance carries the same momentum: for the first quarter of 2026, TSMC expects revenue of US$34.6–35.8 billion, which works out at the midpoint to roughly 4% growth over the previous quarter and about 38% over the same period last year — a remarkably steep line for what is normally the slow season.
2-Nanometer Volume Production and the Advanced-Node Shift
2-nanometer chips entered volume production during the fourth quarter, and wafer starts are set to climb further across 2026. Wafers at 7 nanometers and below — the tiers where AI processors live — generated 77% of fourth-quarter wafer revenue. Across all of 2025 the share reached 74%, against 69% in 2024, and this one mix shift sits behind both the record margins and the surging capex plan.
Broken down by platform, high-performance computing — the bucket that holds AI and 5G chips — delivered 55% of fourth-quarter sales, while smartphones came in at 32%. The world’s most advanced factories now take their cue from AI silicon rather than handsets.
A $52–56 Billion Budget for 2026
TSMC’s 2026 capital expenditure is budgeted at US$52–56 billion, against the US$40.9 billion actually outlaid in 2025 — an increase of roughly one-third at the midpoint. Jake Lai of Counterpoint Research described 2026 as a “breakout year” for AI server demand, though he flagged the global memory shortage as the biggest near-term variable, one that could squeeze the supply of consumer electronics.
In effect the plan is a pre-placed order for the whole AI economy: equipment vendors, construction, power, and advanced packaging capacity all get pulled through by it. Once the foundry at the bottom of the stack commits at this scale, scarcity has to migrate — toward memory, power, and advanced packaging, precisely the spots where shortages have already begun to surface.
Gross Margin Against the US Buildout
Speaking on the earnings call, Chairman C.C. Wei conceded that memory is short across the industry: the squeeze is sharpest in mainstream Node B and embedded NVM capacity destined for consumer devices, while premium products built around AI silicon are less exposed — and with memory supply constrained, some customers have responded by raising their orders for AI-focused products. Tariffs, he added, stay on the list of risks to watch.
In Arizona, TSMC has bought additional land for its US operation, and Wei repeated his picture of the site growing into a “gigafab” cluster, with overseas manufacturing capacity due for significant expansion in 2026. The bill lands in gross margin: dilution from overseas fabs ran about 2 percentage points in 2025 and will run higher in 2026 — the cost of spreading production across borders, charged straight to the income statement.
Three Takeaways for the AI Supply Chain
Three things for builders to carry away. One: advanced-node supply should stay tight for all of 2026, so teams whose roadmaps rest on leading-edge silicon should secure capacity sooner rather than later. Two: the risk next door is memory — AI accelerators are not the ones taking the hit, mainstream consumer memory parts are, but the BOM of an edge AI device still gets reshuffled. Three: nothing public tracks AI infrastructure demand more faithfully than TSMC’s capex, and a commitment at this scale keeps the 2026 spending supercycle argument from our opening-of-year outlook standing.