TSMC Q2 Earnings July 16: Three CoWoS Signals That Test AI’s Spending Ceiling

A typhoon that shut down Taiwan’s financial markets on Friday forced TSMC to postpone its June 2026 monthly sales data release — a routine disclosure that investors had been counting on as their last clean read of the quarter before the company reports its second-quarter 2026 earnings in six days. That data now arrives July 13, giving investors roughly 72 hours to absorb it before Thursday’s print. The delay is a small thing in the scheme of a trillion-dollar company, but it illustrates a structural reality that every investor in TSMC, Nvidia, or any AI infrastructure play should understand before July 16: nearly every number that matters in this industry runs through a single point in Hsinchu, Taiwan — and one storm can move a market.

TSMC will release its Q2 2026 results before U.S. markets open on July 16, with the earnings conference call beginning at 2:00 a.m. ET (2:00 p.m. Taipei time) at the Mandarin Oriental in Taipei, according to the TSMC investor relations Q2 2026 page. Wall Street expects quarterly revenue of roughly $40 billion, compared to $30.07 billion in Q2 2025 — a year-over-year increase of about 33%. Analysts put Q2 earnings per ADR unit at approximately $3.80-$3.83, up from $2.47 a year ago. TSMC has beaten those consensus estimates in each of its four most recent quarters, with an average positive surprise of 8.34%.

But the revenue figure is almost beside the point. What July 16 actually tests is a bigger question: whether the AI infrastructure buildout that has consumed the world’s most advanced chip manufacturing capacity for three years is decelerating, plateauing, or still accelerating. The answer will ripple through Nvidia, AMD, Broadcom, the hyperscalers, and every semiconductor-adjacent position in the market.

Why TSMC Is the Only Company That Can Answer This Question

To understand why this single earnings call commands more attention than most quarterly prints, it helps to understand TSMC’s singular position in the AI supply chain.

TSMC is the world’s largest contract chipmaker, fabricating advanced semiconductors for Apple, Nvidia, AMD, Broadcom, Google, and Amazon, among others. That list barely captures the scope. Nearly every leading AI accelerator on the market — Nvidia’s Blackwell GPU series, AMD’s MI-series chips, and the custom accelerators designed in-house by the major hyperscalers — is fabricated on TSMC’s advanced process nodes and, critically, assembled in TSMC’s packaging facilities.

In Q1 2026, advanced technologies at 7-nanometer and below generated 74% of TSMC’s total wafer revenue. High-performance computing — the category that includes AI accelerators — accounted for 61% of total Q1 revenue, up 20% quarter-over-quarter, according to TSMC’s Q1 2026 earnings results. In short: if an AI chip exists, there is a high probability TSMC fabricated and packaged it.

The cloud giants — Amazon, Microsoft, Alphabet, and Meta — are collectively on pace to spend roughly $700 billion on AI infrastructure capital expenditure in 2026. Most of that money eventually reaches TSMC, in the form of wafer starts and packaging reservations. TSMC is not the only supplier in the chain, but it is the first to feel the order pressure — and the first to signal when that pressure is changing.

Signal One: Will TSMC Raise Full-Year Revenue Guidance?

TSMC’s management guided Q2 revenue between $39.0 billion and $40.2 billion on the Q1 earnings call, representing approximately 10% sequential growth at the midpoint, as detailed in TSMC Q1 2026 earnings guidance. Q2 gross margin is projected at 65.5% to 67.5%, reflecting high fab utilization, early pricing increases on advanced nodes, and ongoing cost improvement — partly offset by the ramp-up costs of overseas manufacturing and 2-nanometer production.

The real signal investors are waiting for is not whether Q2 hit guidance, but what management says about the rest of 2026. TSMC already raised its full-year 2026 revenue growth forecast on the Q1 call to above 30% in U.S. dollar terms. Citi analyst Laura Chen expects a further upward revision, driven by strong demand for advanced AI chips and higher wafer prices across the 2nm and 3nm nodes. If TSMC raises that guidance on Thursday, it amounts to an explicit statement that the AI capex cycle has additional runway — a green light for the entire sector.

Another signal embedded in the revenue discussion: pricing. TSMC has informed major customers — including Apple, Nvidia, and Qualcomm — of price increases of 5% to 10% on leading-edge chipmaking processes at 7nm and below, according to reporting by supply chain journalist Tim Culpan. These hikes are now rolling out. Higher prices at the same volume means higher revenue even without demand growth, but they also test whether AI chip buyers will absorb the cost rather than reduce orders. The Q2 gross margin print will offer an early read.

Signal Two: CoWoS Capacity Update — the Bottleneck That Determines AI’s Pace

The number that may matter most on Thursday is not revenue and not margin. It is what C.C. Wei says about CoWoS.

For most of the last twenty years, the binding constraint in chip manufacturing was the ability to shrink transistors. That problem is largely solved. TSMC can now produce 2nm and 3nm logic dies with relatively high yields. The bottleneck that prevents hyperscalers from deploying AI data centers at their desired pace is no longer silicon fabrication. It is advanced packaging — specifically, the 2.5D and 3D packaging technologies required to connect high-performance logic chips with High Bandwidth Memory.

The technology at the center of this story is CoWoS — Chip-on-Wafer-on-Substrate. TSMC’s proprietary 2.5D advanced packaging architecture places a logic die (the GPU or ASIC itself) and multiple stacks of High Bandwidth Memory side-by-side on a silicon interposer — a passive silicon die fabricated with through-silicon vias (TSVs), which are copper-filled vertical channels that allow connections between the stacked dies at bandwidths measured in terabytes per second. Every Nvidia Blackwell GPU, every AMD MI-series AI accelerator, and every major hyperscaler custom ASIC depends on CoWoS capacity allocation to become a shippable product, as detailed in TSMC’s CoWoS technology overview.

What makes CoWoS strategically important — and what the draft framing of TSMC as “contract chipmaker” tends to obscure — is that CoWoS production is captive at TSMC. There is no merchant market for CoWoS interposers. A fabless company that designs an AI accelerator and needs HBM integration has no alternative to TSMC’s packaging lines. This means TSMC holds a near-monopoly not just at the logic-die fabrication level, but at the packaging level as well — a double concentration that today’s typhoon-triggered delay illustrates in miniature. The same geographic reality that postponed a routine monthly data release is the same one that would slow AI chip deployment in any more serious disruption.

At the June 4 annual shareholders meeting, CEO C.C. Wei told shareholders that CoWoS capacity remains “extremely tight and sold out through 2026.” Nvidia’s management has described CoWoS assembly as “oversubscribed through at least mid-2026,” and the Silicon Analysts CoWoS allocation tracker found lead times of 52 to 78 weeks at TSMC’s advanced-packaging backend facilities, with bookings extending into 2027.

TSMC’s response has been aggressive. CoWoS capacity has been expanding at roughly 80% per year — growing from approximately 35,000 wafers per month at the end of 2024 to approximately 75,000 by the end of 2025, with a target of 125,000 to 130,000 wafers per month by the end of 2026, according to TrendForce CoWoS capacity analysis. That is a near-quadruple expansion in under two years. Industry estimates suggest the gap between packaging supply and demand, which ran as wide as 20% earlier this year, could narrow to roughly 10% by the end of 2026 as new capacity comes online. On Thursday, investors will be listening for any update on whether that timeline is holding — and whether any language about 2027 capacity changes the outlook.

Signal Three: Capital Expenditure Trajectory

TSMC’s 2026 capital budget is $52 billion to $56 billion — the largest in the company’s history, approximately 37% above 2025 capex levels — and management has said it intends to spend at the high end of that range, as confirmed in TSMC Q1 2026 capital expenditure guidance. When a company that operates on 18-month to 24-month planning cycles commits money at that scale, it is making a bet on where demand will be two to three years from now, not six months from now.

Any upward revision to this figure on Thursday — TSMC essentially committing more of its own capital to sustained AI demand — would carry significant weight with semiconductor investors. William Li, a senior analyst at Counterpoint Research, told CNBC that 2026 is shaping up to be a “breakout year” for AI servers backed by growth in both chip manufacturing and packaging technology. Goldman Sachs research noted that TSMC’s production capacity will likely continue to lag surging demand through the near term, which is precisely why the capex figure matters: it tells you whether management sees the demand horizon extending or contracting.

Arizona’s Real Story: Packaging Capacity, Not Just Logic Fabs

TSMC’s domestic U.S. expansion is simultaneously its most ambitious project and its most complex execution risk.

TSMC’s Arizona campus has expanded from an initial $12 billion two-fab plan to $165 billion for the current six-fab phase, representing the largest foreign direct investment in a greenfield project in American history, as confirmed in the TSMC official U.S. investment announcement. Fab 21’s first phase entered high-volume 4nm production in Q4 2024 and turned an approximately $514 million profit in its first full year — a milestone that challenged the long-dominant view that advanced chipmaking outside Taiwan cannot be economically viable. The second fab’s structure was completed in April 2026, with 2nm commercial production targeted for H2 2027, per the TSMC Arizona expansion overview.

The more consequential element of the Arizona buildout may be what comes later: two advanced packaging facilities, designated AP1 and AP2, scheduled to begin construction in early 2026 and targeting mass production by 2028. These will be the first CoWoS-capable facilities on U.S. soil. Until they are online, every AI accelerator that uses TSMC’s most advanced packaging must still route through Taiwan — which means the geographic concentration risk that today’s typhoon illustrated is not fully addressed by the logic-fab investment alone.

Execution risks at the Arizona campus are real. TSMC reported that water availability in the Arizona desert remains a challenge — the company is pursuing 90% water reclamation through an industrial water treatment facility. Visa processing delays have complicated the rotation of Taiwanese engineers. In Q3 2025, an unexpected gas supplier outage caused hours of downtime, scrapping thousands of wafers and sharply reducing quarterly profits — a reminder that operational maturity at a new facility takes time, as reporting on TSMC Arizona challenges documented.

The geopolitical case for Arizona was confirmed in TSMC’s annual SEC 20-F filing, which explicitly flags U.S. export controls, tariff policy, and customer concentration as ongoing risks. The Arizona investment is partly a hedge against all three — a domestic supply option that serves U.S. national security interests alongside the commercial case.

What Happens When AI Chip Prices Rise: The Consumer Angle

There is a downstream story that connects TSMC’s pricing power directly to the device sitting in your pocket.

TSMC has informed major customers of price increases of 5% to 10% on all leading-edge 7nm-and-below chip manufacturing processes, which account for approximately 75% of TSMC’s total revenue. For TSMC’s largest AI customers — Nvidia, AMD, Broadcom — the increases are largely absorbable. Nvidia’s margins remain wide enough to pass costs downstream, and data center buyers have shown minimal sensitivity to accelerator pricing as long as the models that run on them remain commercially valuable.

For smartphone and PC chipmakers operating on thinner margins, the dynamic is different. Qualcomm, MediaTek, and others who rely on TSMC’s advanced nodes for consumer-facing chips are expected to pass a meaningful portion of these increases to device makers — and from there, to consumers. Analysts expect flagship smartphone and PC prices to move higher in the second half of 2026, with the magnitude dependent on how much of the cost increase suppliers can absorb before it reaches retail. If you are planning to buy a flagship phone later this year, TSMC’s pricing power is a factor in what you will pay.

Should Investors Buy TSM Before Earnings?

This is the question most investors are actually asking, and it deserves a direct answer.

Analysts are broadly bullish heading into the print. Based on TSM analyst price targets ahead of Q2, the average analyst price target implies roughly 12% upside from current levels, while the most optimistic targets suggest the stock could climb roughly 38% from here. Barclays analyst Simon Coles raised his price target to $625 from $470 and reiterated a Buy rating, describing TSMC as one of the best AI investment opportunities in the semiconductor sector, while Citi analyst Laura Chen raised her Taiwan-listed target to T$3,800 from T$2,875 and placed the stock on a 30-day upside catalyst watch. TSM surged 39.9% in the April-through-June quarter alone, and the stock has gained more than 97% over the past 12 months.

The bull case is clear: TSMC holds a near-monopoly on advanced chip fabrication and advanced packaging simultaneously, its customers are locked in for years by the long lead times required to qualify new nodes, and the transition from generative to agentic AI is driving a new wave of compute demand on top of the wave that already consumed available capacity. C.C. Wei has earned credibility through a consistent pattern of making bullish capital commitments only after extended channel checks with major customers.

The bear case — or at least the cautious case — is also worth stating plainly. Seeking Alpha analysts covering the name have flagged that TSMC’s valuation reflects peak margins and peak utilization, making profitability and free cash flow highly sensitive to even modest demand fluctuations. The Ferrante Capital research note from April 2026 made a useful structural point: TSMC’s HPC revenue mix is a lagging indicator, not a leading one. Foundry revenue lags customer orders by one to three quarters depending on the node. If hyperscaler capex begins to soften in Q2 or Q3 2026, TSMC would see the deceleration toward year-end, not on Thursday.

Earnings announcements are binary events. A strong beat paired with raised full-year guidance could send the stock sharply higher; any sign that the AI capex cycle is cooling — even modestly — would likely trigger a significant pullback in TSM and across the semiconductor sector. This article does not constitute investment advice. Conduct independent research and consult a qualified financial advisor before making any investment decision.

What Makes This Earnings Call Different From Any Other

TSMC’s Q2 print will be dissected for revenue, margin, and guidance. But the more consequential question it is being asked to answer is this: how durable is the AI infrastructure buildout? When C.C. Wei said on the Q1 2026 earnings call that cloud service providers’ spending “massively reinforces the idea that the AI semiconductor cycle is a multiyear opportunity rather than a temporary spending wave,” he was describing what his order book showed at that moment — not a forecast, per TSMC’s Q1 2026 earnings conference.

On Thursday, his language will either confirm or qualify that statement. If his tone on AI demand remains “extremely robust” without caveat, the market will interpret it as confirmation that the next chapter of the buildout is financed and on schedule. If it carries any qualification — any acknowledgment of customer pushout, any softening in second-half order visibility — the sector will re-rate accordingly.

The CEO who runs the only company on Earth that makes both the AI chips and the packaging required to ship them will speak at 2:00 a.m. ET on Thursday, July 16. The June 2026 sales data, postponed by today’s typhoon, arrives Monday. Between the two, investors will have more clarity about where the AI infrastructure cycle stands than from almost any other source available in 2026.

---

Frequently Asked Questions

What is CoWoS, and why is it more important than the chip fabrication itself in 2026?

CoWoS — Chip-on-Wafer-on-Substrate — is TSMC’s 2.5D advanced packaging technology. It integrates a logic die (such as an Nvidia GPU or a custom AI accelerator) and multiple stacks of High Bandwidth Memory on a shared silicon interposer using through-silicon vias, enabling the terabyte-per-second memory bandwidth that large AI models require. No alternative method can achieve comparable bandwidth at the same power envelope. Critically, CoWoS production is captive at TSMC — there is no merchant market for CoWoS interposers — which means TSMC controls both the logic die fabrication and the final assembly step. With lead times of 52 to 78 weeks at TSMC’s packaging facilities and bookings extending into 2027, CoWoS availability is now the primary constraint on how fast AI accelerators can be deployed, even though silicon fabrication itself has largely cleared its earlier bottlenecks.

What three signals should investors watch when TSMC reports on July 16?

First, whether management raises its full-year 2026 revenue growth outlook above the current above-30% target — an upward revision would signal that the AI capex cycle has measurable additional runway. Second, what C.C. Wei says specifically about CoWoS capacity in the second half and into 2027 — whether his language on packaging tightness modifies or confirms “extremely tight and sold out.” Third, whether the 2026 capital expenditure budget is revised upward from the current $52-56 billion range, since that figure represents TSMC’s own bet on where demand will be in 2028 and beyond.

Will rising TSMC chip prices affect what consumers pay for smartphones and PCs?

Yes, with a lag. TSMC has begun notifying major customers of price increases of 5% to 10% on all advanced chipmaking processes at 7nm and below. For AI data center buyers, these hikes are largely absorbable because AI chip margins are wide and buyers have shown limited price sensitivity as long as AI model returns justify the hardware cost. For consumer-facing chipmakers — Qualcomm and MediaTek in particular — thinner margins mean more of the cost increase flows downstream. Analysts expect a meaningful portion of the increases to appear in flagship smartphone and PC prices in the second half of 2026, though the exact magnitude depends on how much each supplier can absorb before reaching retail.

How does TSMC’s geographic concentration in Taiwan affect the AI supply chain?

Taiwan accounts for the vast majority of TSMC’s advanced logic fabrication capacity and, currently, all of its CoWoS advanced packaging capacity. As today’s typhoon-triggered postponement of TSMC’s June sales data release illustrates, operational disruptions in Taiwan affect every company in the AI chip supply chain. A more serious disruption — whether from weather, a power event, or geopolitical escalation — could slow AI accelerator deliveries globally, given that no alternative packaging facility operates at comparable scale. TSMC’s Arizona expansion will add logic fab capacity over the 2027-2029 window and eventually add U.S.-based packaging capacity through its AP1 and AP2 facilities, but those facilities are years from offsetting Taiwan’s role as the single point of assembly for the world’s AI chips.