The chip market is set to double in a year. Memory, not processors, is doing the lifting.

For most of the past decade, the semiconductor industry grew in single or low double digits a year, with the occasional memory-driven boom and bust. 2026 is not that. The World Semiconductor Trade Statistics organisation (WSTS), whose figures are compiled from the sales of member chip companies, now expects the world market to more than double in a single year.

A forecast that keeps moving up

In June, WSTS put 2026 sales at about US$1.51 trillion, already 90% above 2025. Two months later it raised that again. First-half sales came in at around US$702 billion, roughly double the same period a year earlier, and the full-year projection moved to about US$1.66 trillion. WSTS now pencils in close to US$2.1 trillion for 2027.

World semiconductor sales

US$ billion, calendar years

Memory is doing the lifting

The headline growth is narrow. In WSTS's June breakdown, memory alone was forecast at about US$804 billion for 2026, more than half the whole market, after growth of nearly 250%. The August update lifted memory growth to roughly 300%. Logic, which includes the processors and accelerators most people associate with AI, was set to grow a still-strong 37%. Analog, sensors, discretes and optoelectronics, the parts that go into cars, factories and phones, were forecast to grow between 3% and 10%.

The reason is the way AI servers are built. Each accelerator is packaged with stacks of high-bandwidth memory, and the companies building AI capacity are buying that memory faster than it can be made. When supply is that tight, price does the work, and the revenue line swells.

2026 forecast sales by product

US$ billion, WSTS Spring 2026 forecast

Who is paying for it

The buyers are a short list. In their April 2026 results, Alphabet, Amazon, Meta and Microsoft between them guided to more than US$700 billion of capital spending for the year, around twice what they spent in 2025, most of it on data centres and the servers inside them. Demand at the other end looks real so far: Google Cloud revenue rose 82% year on year in the second quarter, to US$24.8 billion.

The suppliers' numbers move in step. NVIDIA reported US$89.0 billion of data centre revenue in the quarter to late July, up 117%, and guided to about US$108 billion of total revenue for the next quarter without assuming any data centre sales to China. TSMC, which manufactures most of the world's leading-edge chips, raised its 2026 capital budget twice, to US$60–64 billion, and now expects revenue to grow slightly more than 40% in dollar terms. Between 10% and 20% of that budget is going to advanced packaging and testing, which is where the memory and the processor are joined together and where capacity has been tightest.

Where the opportunity sits

By region, WSTS expects the Americas and Asia Pacific to lead, with Asia Pacific sales forecast at about US$824 billion and the Americas at about US$544 billion in its June numbers. The memory makers of South Korea, the foundries of Taiwan and the system builders in the United States take most of the value.

The more interesting openings are one step down the chain. Packaging, testing, substrates, power delivery and cooling are all short of capacity, and they are less concentrated than leading-edge fabrication. That is the space India is building into. Under the India Semiconductor Mission, the government has approved 12 projects with about ₹1.64 lakh crore of investment: one silicon fab, two compound-semiconductor fabs and nine assembly and packaging units. The second phase of the mission, announced in the 2026–27 Budget, shifts attention to equipment, materials and home-grown chip design, where 24 design projects have already been supported.

Power is the other constraint, and it is one we covered separately in AI's scarcest input is no longer chips. It is power.

What this means for leaders

• Memory is now a cost line, not a commodity. Any business that buys servers, PCs, phones or vehicles with significant memory content should expect higher and less predictable component costs into 2027, and should lock in supply where it can.
• The boom is narrow. Suppliers selling analog chips, sensors and power parts into cars and industry are not riding this wave. Valuing them as if they were is a mistake.
• Bottlenecks move. Last year it was accelerators. This year it is memory and packaging. Next it may be power equipment or cooling. Investment cases should name the constraint they are betting on.
• Plan for both halves of the cycle. Memory has always swung hard. A capacity plan built on 2026 prices alone will be wrong in one direction or the other.

A market that doubles in a year on the back of one product family is a cycle with an AI story attached. Plan for the downturn while you enjoy the upturn.

Figures are as published by the sources above at the time of writing. Forecasts and company guidance are labelled as such and will change.

← Back to all insights