The Semiconductor Industry Association reported global semiconductor sales of $403.3 billion in the second quarter of 2026, up 35.1% from the first quarter. The same release put June sales at $134.5 billion, up 123.6% year over year and 9.7% month over month, with the Americas up 160.9% YoY, Asia Pacific up 124.4%, China up 112.8%, Europe up 75.2% and Japan up 39.0%. SIA president and CEO John Neuffer said full-year 2026 sales are expected to exceed $1.5 trillion. Every one of those figures is real and correctly reported (SIA; mirrored at Evertiq).
They also do not mean what a plant manager staring at a quote sheet would reasonably assume they mean. Two corrections come first, because everything else depends on them.
The June number is not June. SIA/WSTS monthly figures are compiled by World Semiconductor Trade Statistics and represent a three-month moving average, a point carried in the release boilerplate (Circuits Assembly). Multiply $134.5 billion by three and you get roughly $403.5 billion — the quarter. There was no single month in which the industry sold $134.5 billion of chips.
These are sales, not billings. "Billings" is SEMI's equipment book-to-bill vocabulary. The 160.9% Americas figure reflects where revenue is booked, not what American factories consumed.
What the 35% actually is
WSTS publishes the decomposition, and it settles the price-versus-units question in one line. For the first half of 2026, the global market reached $702 billion, up 102% year over year — with memory up 305% YoY and logic up 45% (WSTS).
The forecast tells the same story forward. WSTS's Spring 2026 outlook puts the 2026 market at $1.51 trillion, up 90%, and 2027 near $1.9 trillion. By category for 2026: memory roughly +250%, to more than $800 billion — about half the entire market by itself — logic +37%, microprocessors +20%, analog +10%, discretes +8%, sensors and optoelectronics +3% (WSTS Spring 2026 forecast).
Analog and discretes are what machine builders, appliance OEMs and industrial-controls firms buy by the reel. Their growth rates are single-digit and low-double-digit. The 123.6% headline belongs to a product category most industrial BOMs touch lightly.
The arithmetic is worth doing in the open. TrendForce put 1Q26 DRAM industry revenue at $97 billion, up 81% QoQ, driven by conventional DRAM contract prices rising 93–98% sequentially — Samsung $37.32B (+93.4%), SK hynix $27.98B (+62.5%), Micron $21.75B (+81.6%) — and forecast a further 58–63% QoQ contract price increase in 2Q26 (TrendForce, June 1, 2026). Back out Q1 total sales from the SIA figures and you get an implied ~$298.5 billion ($403.3B ÷ 1.351), meaning the sequential dollar increase was roughly $105 billion. That is a derived estimate, not a published figure, and should be treated as one. But a $97 billion product category repricing 58–63% higher against constrained bit growth plausibly accounts for well over half of the entire quarterly increase on its own.
The industrial reality check
If the 123% headline were a broad chip-price event, it would show up in the results of the supplier that sells into machine builders. It does not — and what does show up is more useful.
Texas Instruments reported Q2 2026 revenue of $5.46 billion, up 23% year over year and 13% sequentially, with net income of $1.98 billion, EPS of $2.14, 61% gross margin and operating profit of $2.3 billion (+48% YoY). Industrial was up 30% YoY, automotive up mid-teens, data center roughly doubled. Analog grew 26% YoY; Embedded Processing hit $788 million against $679 million, up 16%. Inventory days fell 13 sequentially to 196 (TI, July 22, 2026).
Set the two side by side: the global market grew 102% in the first half; the analog and embedded supplier that industrial buyers actually depend on grew 23%. The gap is memory. And 30% YoY industrial growth is not a pricing mirage — industrial chip demand is genuinely up in units. Both things are true, and an article that reports only the first is misleading its readers.
What is arriving: the price-increase ledger
The memory event is not staying in memory. Electronics Weekly documented the spread across suppliers, with effective dates (May 12, 2026):
- Texas Instruments — increase scheduled July 1, reported to range from 15% to 85% on lines including digital isolators and PMICs.
- NXP — effective June 1, the company's second increase of 2026.
- STMicroelectronics — effective April 26.
- Analog Devices — averaging 15% since February, 30% on military-spec parts.
- Murata — 15–35% on multilayer ferrite components, April 1.
- onsemi — raised April 1, targeting 38–40% gross margins.
Note the categories: digital isolators, PMICs, discretes, ferrites, passives. Those are the parts in industrial control cabinets and power stages, not in AI training racks. The same report cited DRAM up 90–95% in Q1 and NAND up 70–75% — the wave that started it.
A 10–30% move on the analog, MCU and power content of an industrial machine is a materially different problem from a 90% move on server DRAM. It is also a real problem, because it lands on parts with no second source and long qualification cycles.
The distributor channel is the leading indicator
The two large broadline distributors reported within a day of each other in early August, and both quantified how much of their growth was price rather than volume.
Arrow Electronics (Q2 2026, call August 6) posted revenue of $10.0 billion, up 32% YoY, with Global Components at $7.4 billion. Global Components president Rick Marano said "price inflation contributed roughly one-third of the sequential revenue growth in our Global Components business." Lead times "continue to extend for certain technologies, but they remain lower than a pervasive shortage environment." Customers are "adding buffer inventory back into their supply chains... I don't consider it irrational in any way." Backlog is "building out into the first half of 2027." CFO Raj Agrawal said book-to-bill was "well above one in all three regions." CEO William Austen: "we're in the second inning" (transcript).
Avnet (FQ4 2026, call August 5) put roughly one-third of both sequential and year-over-year sales growth down to memory pricing, and said memory pricing drove more than 50% of a $600 million inventory dollar increase. CFO Ken Jacobson characterized price increases outside memory as "modest or minor." CEO Philip Gallagher said "pricing increases became more prevalent beyond memory" and that "lead times moving higher across most component categories," with book-to-bill above 1 in all regions and the transportation vertical up about 25% YoY in the Americas (transcript).
That is the honest channel read: extending lead times, book-to-bill above one everywhere, backlog reaching into 2027, buffer stock rebuilding — and explicitly not a pervasive shortage. Tightening, not 2021.
The lag: why this shows up in Q4 BOMs, not headlines
Component price increases do not hit finished-goods pricing on the effective date. They hit when the cheaper inventory runs out. TrendForce describes exactly this mechanic on the consumer side — retail notebook prices rising as higher-cost components gradually flow through notebook inventories, and smartphone vendors planning retail increases to offset LPDRAM costs (July 3, 2026). Dell and Lenovo moved first, in December 2025 and January 2026 respectively (TrendForce; treat those specific percentages as reported-at-the-time).
The industrial version runs slower and lands harder. Tier-1 suppliers and machine builders typically own the BOM under fixed-price agreements, so they absorb the increase into gross margin first and reprice at contract renewal. That deferral is why the pain arrives as a margin surprise before it arrives as a price list. TI's inventory days falling 13 sequentially to 196 is a measurable sign the buffer is being drawn down rather than replenished — the flush stage that precedes repricing.
The deceleration nobody is pricing in
TrendForce's 3Q26 outlook has conventional DRAM contract prices rising 13–18% QoQ and NAND flash 10–15% QoQ — still up, but a fraction of Q1's 93–98%. The stated reasons are that "record-high contract prices mean customers from consumer markets, such as PCs and smartphones, are reaching their affordability limit," plus high base effects.
Two things follow, and they are routinely conflated. Prices are not falling; the rate of increase is falling. A buyer signing a 2027 supply agreement will pay the level, not the delta. Deceleration from a repriced base is not relief.
The capacity ledger, done properly
Here is where the headline of this piece needs qualifying against the data rather than repeated.
US greenfield electronics construction is genuinely in retreat. IoT Analytics, working from Census Value of Construction Put in Place data, found construction spending on computer, electronic and electrical manufacturing down 44% from its July 2024 peak — a category that had accounted for more than half of all US manufacturing construction spending at that peak. Excluding electronics, manufacturing construction rose 5.6% between February 2025 and March 2026. Data center construction, meanwhile, reached roughly $47 billion annualized in January 2026, up from about $9.5 billion in January 2020 (IoT Analytics, May 12, 2026). The Census C30 release for June 2026 put total construction at $2,166.5 billion SAAR, 0.1% below a revised $2,168.5 billion in May, with manufacturing down 1.2% for the month (Census C30).
But equipment spending is going the other way, hard. SEMI forecasts global 300mm fab equipment spending of $133 billion in 2026 (+18%), $151 billion in 2027 (+14%), $155 billion in 2028 and $172 billion in 2029. For 2027–29 the split is $228 billion for logic and micro against $175 billion for memory (DRAM $111B, 3D NAND $62B). SEMI tracks 404 facilities and lines globally and has added nine new fab/line projects since December 2025. CEO Ajit Manocha: "AI is resetting the scale of semiconductor manufacturing investment" (SEMI, April 1, 2026).
These datasets do not contradict each other. Tools are going into shells that already exist or are already in progress, and the spending is weighted to leading-edge logic and memory. Capacity is being added. It is simply not being added where industrial buyers source parts.
What it means for 2027
So the concrete headline is directionally right and mechanically wrong. It isn't that nobody is building — it's that the shells being built and equipped are for AI. Mature-node capacity — where industrial analog, MCUs, discretes and power devices are made — appears in neither growth story. It is the residual: it improves only as a side effect of leading-edge investment, if at all.
Frame the 2027 risk accordingly. This is not shaping up as a repeat of 2021's absolute scarcity; Arrow says on the record that lead times remain short of a pervasive shortage environment. The binding constraints are allocation and lead time at mature nodes, against a price level that has already reset and is still climbing, just more slowly. Plan for a base that does not come back down, and for parts that are available but scheduled rather than absent.
What we're still chasing
Four open questions this reporting has not yet closed, stated as questions rather than assertions:
- Have industrial OEMs repriced, or are they absorbing? Tier-1s typically own the BOM and eat the first increase. The answer determines whether this shows up as customer invoices or as margin compression.
- What does a 10–30% analog/MCU/power ASP move actually do to gross margin on a roughly $50,000 industrial machine, given semiconductor content is a modest share of that BOM? The arithmetic may be smaller than the anxiety.
- Is lead time or price the binding constraint for mature-node industrial parts? Distributors report both moving; only one of them stops a line.
- Is mature-node capacity being crowded out by AI-driven allocation at the foundries and IDMs — and what have TI, onsemi, Infineon and ST actually committed to mature-node expansion?
One independent check worth running alongside the trade data: BLS producer price index for semiconductor and other electronic component manufacturing (series PCU3344133441), which reads realized US price inflation rather than announced increases.
Related reading
-
$1.966 Trillion Announced, 22% Less Built: The Reshoring Boom Is a Press-Release Boom
-
Boeing Is Studying 70 Jets a Month. Its New Line Builds One.
Sources
- SIA — Global Semiconductor Sales Increase 35.1% from Q1 2026 to Q2 2026
- Evertiq — Global semiconductor sales hit $403 billion in Q2 (Aug 25, 2026)
- Circuits Assembly — Global Semiconductor Sales Jump 35% in Second Quarter (Aug 7, 2026)
- WSTS — Global Semiconductor Market records exceptional growth in Q2 2026
- WSTS — Global Semiconductor Market Surges Beyond $1.5T in 2026 (Spring 2026 forecast)
- TrendForce — Rapid Contract Price Surge Drives 1Q26 DRAM Industry Up 81% QoQ (June 1, 2026)
- [TrendForce — AI Server Demand Continues to Support Memory Prices in 3Q26 (July 3, 2026)](https://www.trendforce.com/presscenter/news/20260703-13134.html)
- Electronics Weekly — Price rises spread to analogue, discretes and passives (May 12, 2026)
- Arrow Electronics Q2 2026 earnings call transcript (Aug 6, 2026)
- Avnet (AVT) Q4 FY2026 earnings call transcript (Aug 5, 2026)
- Texas Instruments — Q2 2026 financial results (July 22, 2026)
- IoT Analytics — US manufacturing reshoring boom: what the data says (May 12, 2026)
- US Census Bureau — Monthly Construction Spending, June 2026 (C30)
- SEMI — 300mm Fab Equipment Spending to Grow Double Digits in 2026 and 2027 (April 1, 2026)
- TrendForce — Memory Crunch Hits PCs: Dell Hikes Prices 15–20% Mid-December, Lenovo from January 2026