Semiconductor Supercycle 2026: Chip Sales Cross $1.5 Trillion for the First Time — Three Forces That Make This Cycle Different

The semiconductor market is famous for swinging between boom and bust every few years. But what is happening in 2026 is not a simple repeat of that pattern. Here we trace the three forces behind the surge, with sourced numbers throughout. For the industry's overall structure — size, regions, segments — see our market overview (in Japanese).

Why has the semiconductor market reached $1.5 trillion?

Start with the numbers. Per SIA and WSTS data, 2025 global semiconductor sales came to about $791.7 billion, up 25.6% year over year. The WSTS Spring 2026 forecast calls for about $1.51 trillion in 2026, expanding further to roughly $1.91 trillion in 2027.

By segment, 2025 saw logic semiconductors at about $301.9 billion (up 39.9% year over year) and memory at about $223.1 billion (up 34.8%) — extraordinarily high growth in both. The simultaneous surge of these two segments is what pushed the total market into $1.5 trillion territory.

The monthly data since the start of 2026 suggests the forecast may even be conservative. SIA reported in July that global semiconductor sales for May 2026 reached $120.6 billion, up 9.2% from April and up 104.1% — more than double — from a year earlier. It was the highest monthly total on record, and the fifteenth consecutive month of month-over-month growth (SIA, July 6, 2026). Note that SIA's monthly figures are three-month moving averages of WSTS data, not raw single-month sales. The abnormality of this phase is precisely that a smoothed series still shows sales doubling year over year.

How has AI inference demand transformed logic semiconductors?

The biggest force lifting logic semiconductors is demand for AI "inference" — the processing in which a trained AI actually produces answers. Deloitte estimates that in 2026, roughly two-thirds of all AI processing will be inference, and that the inference chip market will exceed $50 billion.

Where earlier AI investment concentrated on the training phase, 2026 is seeing accelerating investment in inference infrastructure — putting trained models into live services. That means continuously compounding demand for Nvidia, AMD, and the custom ASICs (application-specific integrated circuits) that hyperscalers design in-house. Inference demand is not a one-off spike: as long as the services keep running, the chips keep drawing power. It is structural demand.

HBM has pushed the memory market to unprecedented levels

The force behind the memory segment's surge is the supercycle in HBM (High Bandwidth Memory). HBM is a specialized memory mounted directly beside GPUs and AI accelerators, delivering several times the bandwidth of standard DRAM — an indispensable component for large-scale AI inference and training.

Per TrendForce data, DRAM prices in Q1 2026 recorded a 90-95% quarter-over-quarter surge, followed by another 58-63% in Q2. NAND flash also jumped 70-75% in Q2. SK hynix maintains its outlook that the HBM-led memory supercycle will continue, with supply-side constraints still severe. This price surge underpins the memory segment's revenue explosion and feeds directly into the market's overall expansion. For the structural reasons memory prices swing this violently, see our explainer on the memory price cycle (in Japanese).

Why does advanced logic keep concentrating at TSMC?

The third driver is the extreme concentration of demand at advanced foundries — the contract manufacturers of chips. Per TrendForce, the total foundry market in 2025 was about $169.5 billion, of which TSMC took about $122.5 billion — a 69.9% share.

Leading-edge AI chips can currently be mass-produced only on TSMC's 3nm and 2nm-class processes, and Nvidia, Apple, and AMD are competing to lock up wafer starts at those fabs. On the supply side, building a new advanced fab takes years, so capacity cannot expand quickly. Demand surging while supply stays constrained — that configuration keeps pushing up both logic prices and revenue.

What makes 2026 different from past silicon cycles?

Past silicon cycles fell mainly on inventory corrections: overstocked components for PCs and smartphones. Supply chains would build inventory in unison, then halt orders in unison when demand cooled — that repetition produced the industry's downturn every few years.

The 2026 cycle has a different structure. Logic demand is underpinned by the continuous real demand of AI inference; memory is dominated by HBM, a product with hard supply constraints; and advanced logic is concentrated at a handful of fabs — all three working at once. That said, risk has not vanished. Broadcom's cautious guidance triggered a synchronized selloff on June 5, 2026 — AMD down 10.86%, Intel down 11.28%, Micron down 7%. That selloff was not an AI bubble bursting — it was a correction driven by concerns over slowing AI demand growth, memory market conditions, and a slump in smartphone demand. Short-term supply-demand corrections recur even within structural growth.

How far can the cycle run toward 2027?

WSTS forecasts 2027 global semiconductor sales of about $1.91 trillion — roughly 26% further growth over 2026. Investment in AI inference infrastructure is not a one- or two-year affair: as long as the cloud giants keep expanding data centers, demand for logic, memory, and packaging keeps stacking up. Japan is part of that buildout too, with 38 data center projects planned through 2030.

There are caveats. If HBM prices keep climbing, cloud providers' cost burden rises and investment could pause. Weakness in general-purpose markets such as smartphones can also weigh on industry sentiment. Reading this market requires holding two things at once: conviction in the long-term trend, and vigilance toward short-term supply-demand waves. Note the geopolitical overlay as well — export controls and tariffs are reshaping who can sell to whom, as the near-zero China sales of Nvidia's cleared-for-export H200 show. This article explains facts and context; it is not a recommendation to invest in any specific stock or sector.