Omdia puts global semiconductor revenue above $425 billion for the second quarter of 2026, up 31.4% on the first quarter. That beats the sequential record of 29.2% set only three months earlier. Across the 97 quarters Omdia has tracked since 2002, sequential growth above 10% has happened ten times. The last four quarters account for four of them.
The headline describes one market growing. The detail underneath describes two. Omdia says memory ICs made up more than half of second-quarter revenue, and that the non-memory market grew by a little over 10%. Those two disclosures are enough to take the quarter apart.
First-half revenue of $752 billion leaves about $325 billion for the first quarter. With memory just above half of Q2, it came to roughly $215 billion, leaving about $210 billion for everything else. Back out 10% growth and non-memory was near $191 billion in Q1, which puts Q1 memory around $134 billion. Memory revenue therefore rose about 60% in one quarter; its share of the market went from roughly 41% to over 50%. Of the $100 billion or so added between Q1 and Q2, about $80 billion was memory. Push the non-memory growth figure to 12% and memory’s share of the gain still lands near 78%. Four new dollars in five came from DRAM, NAND and NOR.
Omdia is plain about the cause. AI demand has changed what memory makers choose to build, which tightens supply for every other buyer and lifts average selling prices. DRAM, NAND and NOR each posted their biggest second-quarter sequential gain on record, and each hit its highest quarterly revenue ever. Wafer output can’t rise 60% in a quarter. Prices can. This is a pricing print.
The non-memory figure is the cleaner read on real demand, and it is strong on its own. Omdia’s history from 2002 through 2025 puts typical Q2 sequential growth for non-memory chips at just over 3%. This quarter did better than triple that. Microprocessors grew 16% against a seasonal norm of about 1%, which Omdia ties to their growing role in AI systems. Stripped of memory, the industry still had one of its best second quarters in a generation.
Omdia’s own PC desk, reporting five days earlier, shows where the memory price goes after it leaves the fab. US PC shipments rose 1.0% year on year in Q2 to 18.8 million units, after a 7.0% fall in Q1. The analysts credit the rebound to retailers and channel partners buying ahead of price increases driven by DRAM and NAND shortages. Average sell-in prices crossed $1,000 for the first time, up 12%. Business ASPs rose 19%, consumer ASPs 9.7%. Revenue climbed 13.1% on flat units.
That is pull-forward, and part of it sits inside the $425 billion. Some of the memory revenue booked in the second quarter is now channel inventory. Omdia expects US PC shipments to fall 18.4% in the second half and 10.7% for the full year.
The semiconductor desk, meanwhile, forecasts Q3 revenue above $500 billion, taking the first nine months past $1.25 trillion. That would be 50% more than all of 2025, which implies a 2025 total of about $833 billion; the first half of 2026 alone already reached roughly 90% of it. Getting from $425 billion to $500 billion is sequential growth of about 18%, a slower rate on a bigger base. It also requires memory pricing to hold while one of the largest end markets for DRAM and NAND in the US loses nearly a fifth of its volume. That only works if AI servers absorb every bit PCs give up, at the same price or higher. So far they have.
The PC release carries one more number worth isolating. For 2027 Omdia forecasts US units down 4.9% and PC market revenue down 6.0%. Revenue falling faster than units means the average PC gets about 1% cheaper. After a year of double-digit price gains, this is the first place in either release where Omdia pencils in the cost pass-through reversing, whether through cheaper memory or buyers moving down to lower-spec machines. Both routes reduce memory revenue per PC.
For equities, the split matters more than the total. Producers collecting a 60% quarterly repricing are earning scarcity rent, the most profitable revenue in the sector and the least durable. Non-memory designers growing at three times the seasonal rate carry less price risk and less upside per quarter. PC OEMs sit on the wrong side of the same trade (their Q2 revenue gain was customers front-running cost inflation, which borrows from second-half demand).
The moat behind the memory rent is allocation discipline. Samsung, SK Hynix and Micron steering wafers toward HBM and server DRAM is what created the shortage in PC-grade parts. The pricing behind 80% of this quarter’s growth holds for as long as none of the three moves capacity back to commodity nodes.
The number to watch is memory’s share in Omdia’s 3Q26 data. If revenue clears $500 billion with memory share still rising while US PC units drop 18%, AI demand is taking the diverted bits at higher prices. If share slips, the second quarter was the channel buying early.