Global Chip Sales Climb to $403.3 Billion in Q2 as AI Demand Reshapes the Semiconductor Market
SEOUL — The world bought $403.3 billion worth of semiconductors in the second quarter of 2026, up 35.1% from the first three months of the year, according to fresh figures from the Semiconductor Industry Association (SIA). June alone brought in $134.5 billion in sales, a 123.6% jump from June 2025 and 9.7% more than May.
The numbers, compiled by the World Semiconductor Trade Statistics (WSTS) organization as a three-month moving average, put the industry on pace for a year few in the sector dared to predict. SIA president and CEO John Neuffer said global chip sales are now expected to clear $1.5 trillion for all of 2026, with the second quarter running well ahead of the first.
Every region grew at once
What stands out about the June data is that no region sat out. Year over year, the Americas led with a 160.9% gain, followed by Asia Pacific and other markets at 124.4% and China at 112.8%. Europe posted a 75.2% increase, and Japan rounded out the field at 39%.
Month to month, the picture looked similar. China rose 10.4% from May, Asia Pacific and other markets added 9.8%, the Americas gained 9.6%, Japan climbed 8.6%, and Europe advanced 7.7%.
Neuffer credited broad demand rather than a single buyer. Sales in the Americas, Asia Pacific and China are all expanding quickly, he said, which shows how much of the world now depends on chips for everything from phones to factory robots to data centers.
The AI machine is eating silicon
The raw numbers reflect a market that has effectively doubled in twelve months. Analysts point to one outsized customer: the AI data center. Training clusters need graphics processors by the thousands, high-bandwidth memory stacked beside them, and networking silicon to tie the racks together. Every one of those chips starts as a wafer in a fab that costs tens of billions to build and years to bring online.
The supply side is reacting in real time. TSMC posted record July revenue as AI orders filled its 2nm lines, a sign that the industry's most advanced nodes are sold out well into next year. Memory makers are pouring money into new plants too — SK Hynix's board recently approved $38.3 billion for two new fabs, one for DRAM and high-bandwidth memory at Yongin and one for NAND at Cheongju, both aimed squarely at AI server demand.
That spending spree shows up in the SIA numbers. Memory has been the fastest-growing slice of the market through 2026, and the June figures suggest the cycle still has room to run. Contract prices for DRAM have climbed for several consecutive quarters, and NAND suppliers have followed suit after a long stretch of oversupply.
What $1.5 trillion actually means
To put the forecast in perspective: the industry took decades to cross the $500 billion mark, and 2025 closed at well under $1 trillion. A $1.5 trillion year would be roughly double the total from just twelve months earlier — growth that has no real precedent in the modern chip era.
Part of that is price, not just volume. Foundries have pushed up pricing on advanced nodes as demand outstrips capacity, and memory makers have followed with higher contract prices for DRAM and NAND. The mix is also richer: a single AI accelerator carries more silicon value than a dozen smartphone chips, and systems now ship with hundreds of gigabytes of high-bandwidth memory attached.
The shape of the market has changed too. Eighteen months ago, the largest slice of industry revenue still came from smartphones and PCs. Today, data-center silicon is the biggest single pool of demand, and it is growing faster than the consumer segments that used to set the industry's rhythm. That shift matters for how chipmakers plan capacity: a data-center surge behaves differently from a phone upgrade cycle, with longer design wins, fatter orders and more concentrated buyers.
Europe shakes off its slump
The European numbers deserve a closer look. The region's 75.2% annual growth in June trailed the Americas and Asia, but it marks a real turn after years of flat or shrinking sales. European chipmakers and fab-equipment suppliers have been beneficiaries of the continent's push to build local capacity, and the SIA data suggests that investment is starting to show up in actual revenue.
Japan's 39% gain is quieter but steady. The country remains the world's biggest supplier of the lithography tools and specialty chemicals every fab depends on, and its equipment makers are riding the same wave as the fabs they serve. Tokyo Electron and the other Japanese tool makers have reported record backlogs, and the SIA numbers confirm the demand feeding those order books.
The bottlenecks that could spoil the party
The forecast comes with caveats. Capacity remains the binding constraint: leading-edge fabs are running at utilization rates that leave almost no slack, and new plants take three to four years from the first shovel of dirt to the first wafer out. Equipment lead times for extreme-ultraviolet lithography tools stretch well past a year. If demand keeps climbing at this pace, the industry risks a supply crunch that pushes prices even higher and slows AI buildouts.
There is also the geopolitical layer. China's 112.8% annual jump in June reflects both genuine demand and a heavy push to stockpile and localize production ahead of export controls. Washington's restrictions on advanced tooling to China have forced Beijing to lean on domestic equipment makers, and the SIA numbers show Chinese buyers still absorbing enormous volumes of chips — much of it older-node silicon for cars, appliances and industrial gear.
Labor is a quieter constraint. Fabs do not run themselves, and the industry is short of the technicians and engineers needed to operate the plants now under construction. Every new gigafab competes for the same small pool of trained process engineers, and the shortage is already showing up in project timelines.
Who wins and who watches
The winners in this environment are the companies that own capacity. Foundry leader TSMC is effectively sold out of its most advanced nodes through 2027, and it has told customers to expect higher prices. Samsung and SK Hynix are racing to add high-bandwidth memory lines, where margins are fat and demand is essentially unlimited for now. On the equipment side, ASML, Tokyo Electron and Applied Materials are selling every tool they can build.
The watchers are the fabless designers. Nvidia, AMD, Qualcomm and the rest are paying more for wafers and passing the cost along, which keeps their revenue climbing even as unit volumes flatten. The risk for them is a demand reversal: if AI spending pauses, the long-term capacity deals signed at today's prices start to look expensive.
What to watch in the second half
The SIA report covers sales, not orders, and the market's direction over the next two quarters will hinge on a few known quantities. TSMC's 2nm ramp, already filling up, will determine how much leading-edge supply exists in 2027. Memory pricing, which has been climbing for months, will tell whether the AI buildout can absorb another round of increases. And the pace of new fab construction — from Texas to Yongin to Dresden — will decide whether capacity finally catches up with demand or falls further behind.
For now, the industry is enjoying a moment it has not seen in its history. Sales are up across every region and every product category, the order books are full, and the forecast keeps getting revised upward. The risk is no longer finding demand — it is finding the wafers to meet it.
The full SIA release is available on the Semiconductor Industry Association site, with additional analysis from eeNews Europe and EE Times Asia. For more on the supply side of this boom, see our coverage of TSMC's record July revenue and the Semiconductors section of the site.
