A quarter of a trillion dollars of chips, and then some

Semiconductor cycles are usually described in years. This one is being described in quarters, which is itself the most useful piece of information in the numbers.

Global semiconductor revenue passed $425 billion in the second quarter of 2026, a record 31.4% rise on the quarter before. The figures come from Omdia, which puts first-half revenue at $752 billion.

The firm expects the third quarter to exceed $500 billion, which would take the first nine months of the year above $1.25 trillion. An industry that took decades to reach half a trillion annually is now being forecast to do that in a quarter.

Three things are named as drivers: artificial intelligence infrastructure, surging memory demand and higher chip prices. The order matters less than the third item, because price and volume are very different foundations for a record.

A 31.4% quarterly rise cannot be manufacturing capacity. Fabs do not expand a third in ninety days; wafer starts are committed years ahead, and advanced packaging is the constraint everybody has been describing since 2024. Most of that increase is what the same silicon now costs.

Memory is where that shows most clearly. High-bandwidth memory is allocated rather than sold, its supply is concentrated among three manufacturers, and an AI accelerator cannot ship without it, which is the definition of a component that can be priced rather than merely quoted.

That makes the number a measure of scarcity as much as of demand. Revenue rising faster than output is what a shortage looks like in an income statement, and it is not the same signal as a market growing because more customers are buying more parts.

The telecom read-through is the part the industry itself is watching. Mobile networks are becoming computing estates, with 5G-Advanced, AI-RAN, cloud-native cores, optical and edge deployments all adding processors, memory and networking silicon to sites that used to hold radios.

Data centre construction pulls on the same networks from the other end. Connecting clouds, data centres and enterprise customers requires high-capacity transport, which is why the optical module has become a trade-policy object rather than an ordinary component.

The risk in a cycle shaped this way is the one every memory cycle has had. Pricing power built on allocation reverses quickly when allocation eases, and the revenue that arrived through price leaves faster than the revenue that arrived through customers.

Which is the question to hold against the forecast. A $500 billion quarter is plausible on current constraints. Whether it describes an industry that has grown or an industry that is charging more for the same scarce parts will only be answerable when the parts stop being scarce.