TSMC Q2 2026 Earnings: Net Profit Jumps 77% to a Record as AI Servers Hit 66% of Revenue

The star of the numbers is HPC — high-performance computing for AI servers and data centers. It accounted for 66% of revenue by platform and grew another 20% from the prior quarter. This article walks through how strong the results actually were, what produced them, and what to watch in the next quarter.

How strong were the numbers?

Start with the results set against the guidance the company had issued in advance (TSMC, July 16, 2026).

Revenue landed exactly at the top of the guidance range (US$39.0–40.2 billion), while gross margin and operating margin broke clean through the tops of their ranges. Net profit of NT$706.56 billion is a quarterly record — roughly US$22.4 billion. Against market expectations, it beat the LSEG-compiled analyst consensus of NT$632.6 billion by about 12% (Trading Key, July 16, 2026).

Why did net profit grow 77.4% when revenue grew "only" 36.0%? Because profitability improved dramatically over the year: net margin rose from 42.7% in the year-ago quarter to 55.6%. In other words, TSMC did not just sell more — every sale carried a much thicker layer of profit.

What drove the growth? AI servers are now 66% of revenue

The answer is AI demand. By platform, HPC — high-performance computing chips for AI servers and data centers — accounted for 66% of revenue, up 20% from the prior quarter. Smartphone revenue, meanwhile, was 22% of the mix, down 4% quarter over quarter. A company that only a year ago was still, in part, a smartphone chipmaker now looks like an AI infrastructure company on the inside of its income statement.

The revenue mix by process node backs up the same story.

The headline is that 2nm was booked in the revenue mix for the first time. A 3% share may look small, but 3% of US$40.2 billion in quarterly revenue is roughly US$1.2 billion — an unusually large figure for a node's first quarter in production. CFO Wendell Huang cited the "steep ramp-up" of 2nm as a growth driver ahead (TSMC, July 16, 2026). For what the 2nm generation actually is and why customers are lining up for it, see TSMC's 2nm generation and its nanosheet ramp (in Japanese) and 2nm Capacity Sold Out Before Mass Production Even Began: Where TSMC, Intel, Samsung and Rapidus Stand.

Monthly revenue suggests the momentum is accelerating rather than fading. June revenue alone was NT$442.68 billion, up 67.9% year over year, and first-half revenue totaled NT$2,404.48 billion, up 35.6% (TSMC, July 13, 2026). June's +67.9% is far above the quarter's +36.0% — meaning the most recent months are the strongest. The monthly detail and the "N3 sold out" situation are covered in the Japanese-language version of this earnings article: TSMC earned about ¥3.3 trillion in three months (in Japanese).

What is the Q3 guidance telling us?

For the July–September quarter, TSMC guided revenue to US$44.6–45.8 billion — up 11–14% quarter over quarter on top of a record quarter. That is the company's numerical answer to anyone worried about an AI demand slowdown.

Margin guidance, however, was set slightly lower: gross margin of 65–67% and operating margin of 56–58%, versus Q2 actuals of 67.7% and 60.3%. The FX assumption is NT$32 per US dollar, weaker for the Taiwan dollar than Q2's actual rate of 31.60. Since a stronger Taiwan dollar squeezes margins for a company that earns in US dollars, the FX assumption alone should be a margin tailwind. That the guidance is still lower suggests it bakes in ramp-up costs at overseas fabs — Arizona above all — and the early costs of the rapid 2nm expansion. Why overseas fabs are structurally more expensive is explained in the economics of leading-edge fabs that now cost over US$20 billion each (in Japanese).

There is one more thing hiding in that gross margin line. In June, TSMC was reported to have notified customers of price increases across all advanced nodes from 7nm onward (Culpium, June 23, 2026). The details and caveats of that report are laid out in the reported TSMC price hike covering 74% of wafer revenue (in Japanese) — but if the hikes are real, their effect would show up mainly from 2027 contract negotiations onward. Q2's 67.7% gross margin is a pre-hike figure. If price increases land on top of it, the margin ceiling is still higher than what we have seen.

How should investors read this?

The single most important message of this report is that AI demand keeps converting from expectation into results. HPC at 66% of revenue, advanced nodes at 77%, and the first 2nm contribution — the AI server investment boom is flowing straight into the income statement not only of chip designers like NVIDIA but of the one company that manufactures for all of them. Why TSMC holds that position is explained in why the world cannot build smartphones or AI without one Taiwanese company (in Japanese).

The checkpoints are equally clear. First, as the 65–67% Q3 gross margin guidance signals, the weight of overseas fab costs will keep growing. This is a company that has repeatedly beaten its own guidance, so it is not a reason for pessimism — but the question of when margins peak remains open. Second, what actually caps AI chip output is not wafer capacity but CoWoS, the advanced packaging technology that combines chips at high density on a substrate. That supply constraint is explained in why the bottleneck for AI chips is the package, not the transistor (in Japanese).

Third, smartphone revenue fell 4% quarter over quarter. With AI pulling the train, this goes unnoticed — but a pillar that still accounts for over a fifth of revenue is shrinking, and that matters for resilience if AI investment ever pauses. The more the growth story leans on a single AI leg, the more the repeatability of this quarter becomes inseparable from the durability of AI data center capex.

What are the caveats?

The figures in this article are based on TSMC's IR disclosures (primary sources), but as a same-day earnings article, some things remained unverified and should be flagged.

First, the details of management commentary from the analyst conference call: updates to the full-year revenue growth outlook, the 2026 capital expenditure plan, the pace of CoWoS expansion, and any comments on US tariffs (Section 232) had not been cross-checked at the time of writing. These are all worth following up.

Second, this article does not cover the market reaction — moves in the ADR or Taipei-listed shares, or analyst target-price revisions. There is also some noise in the US-dollar conversion: this article uses TSMC's officially reported US$40.20 billion, while some outlets recompute revenue at market rates to roughly US$39.6 billion. The rough yen conversion in the Japanese edition (about ¥3.3 trillion at ¥147 per dollar) is an approximation, not a precise figure.

Even so, the core numbers — revenue, profit, margins and revenue mix — are all confirmed against primary sources. To the question "is AI demand real?", TSMC's first half of 2026 answered with its financial statements. The next test comes with the Q3 report in mid-October (TSMC, July 16, 2026).