The Trillion-Dollar Headline Hides the Real Story

KEY TAKEAWAYS

• Structural shift: The semiconductor market is on course to cross well beyond USD 1.5 trillion in 2026¹, but the growth traces to one end market: AI infrastructure. Concentration cuts both ways.
• One build-out, many beneficiaries: The same AI build-out drives both memory and logic, and now spans the whole value chain, from foundry and equipment to memory and interconnect, well beyond any single company.
• A sector that keeps reinventing itself: Leadership rotates and the current cycle is price-driven. The durable exposure is the theme, not a single subsector or stock.
• Industry and valuation risk is real: Investing involves risk. Past performance is not a reliable indicator of future results. Read the Key Information Document (KID) before investing.

The Scale of a Structural Shift

The shift is easiest to see in one statistic. Generative-AI chips accounted for roughly 0.2% of global semiconductor unit volume in 2025, yet are set to deliver more than half of industry revenue in 2026². Global revenue is on course for around USD 1.5 trillion this year, up from USD 795.6 billion in 2025¹.

Global semiconductor sales, 2025 to 2027E (USD bn)

Source: WSTS full-year 2025 results and Spring 2026 forecast.

Consensus earnings expectations have re-based accordingly. Bloomberg estimates for the next twelve months (as of 30 June 2026) point to strong but uneven growth across the large US-listed chipmakers: the typical name is expected to grow earnings by above 50 percent, with the average pulled far higher, near 95 percent, by a handful of memory names rebounding from a weak base. Numbers of that size assume the AI capital-expenditure cycle holds, and hyperscaler AI infrastructure spending is running well above 2025 levels. Yet valuations are not as stretched as the "bubble" label suggests: The sector’s forward price-to-earnings multiple has fallen from around 50x in late 2023 to under 20x by mid-2026, because earnings have so far outrun the share price. Demanding is not the same as detached. Execution risk is nonetheless real, and investors should form their own view on whether the trajectory is sustainable³.

Memory: one leg of the story, not the whole story

Memory has driven much of the sector's headline noise in H1 2026. Its revenue is forecast to grow by around 250 percent year over year to more than USD 800 billion, over half of the USD 1.51 trillion market on its own¹.

Forecast 2026 semiconductor growth by product segment (% YoY)

Source: WSTS Spring 2026 forecast.

That figure needs reading with care, because most of it is price, not volume: DRAM prices are estimated to rise around 125 percent in 2026 and NAND around 234 percent, with meaningful relief not expected until late 2027⁴. A revenue line built on a supply-driven price spike behaves very differently from one built on structural volume growth.

The demand behind it, though, is real and specific. Over the past two decades, peak processor performance has grown far faster than memory bandwidth, so feeding data to the accelerators, not raw compute, has become the bottleneck in AI systems⁵. High-bandwidth memory (HBM) is the industry’s current answer, and demand for it is what turns AI spending into memory revenue. That is why an unusually large share of 2026 growth, across memory and logic alike, traces back to a single end market: AI infrastructure^1,5. Samsung, SK Hynix and Micron all reported HBM capacity essentially sold out for 2026 by the end of Q1⁶. Within the US-listed universe, Micron is the most direct way in, with confirmed HBM allocations to both Nvidia's and AMD's next-generation platforms⁶. That US-listed access set is now widening. In July 2026 SK Hynix, the world's second-largest memory maker and Micron's chief HBM rival, listed ADRs on the Nasdaq, in the largest US share sale ever by a foreign company, bringing the segment's dominant supplier within reach of US-listed strategies for the first time⁷.

A Sector That Keeps Reinventing Itself

Semiconductors have one of the most cyclical histories in equity markets. Since 1990, indices tracking US-listed chip companies have suffered multiple drawdowns beyond 40 percent, including two beyond 60 percent. The deepest corrections have typically coincided with shifts in sector leadership. Intel's dominance gave way to mobile, mobile gave way to GPU-driven data-centre computing, and today the profit pool is increasingly shared across custom silicon, advanced packaging and interconnect. Identifying the next cycle's winners in advance has proven difficult, even for specialists with deep sector knowledge.

That rotation has already begun inside the current cycle. HBM allocation decisions finalised in Q1 2026 reshuffled share between the three memory suppliers in a single quarter⁶. Custom-silicon programmes at the largest hyperscalers are compressing the addressable market for merchant GPU designers, a margin risk even if aggregate demand holds. Interconnect moved from unknown to critical infrastructure in a matter of quarters.

That constant change has an upside. The same forces that unseat leaders keep adding new ways to play the theme. Beyond the accelerator itself, the foundry that manufactures nearly every high-end AI chip, the lithography systems that make leading-edge production possible, the process-equipment, materials and testing firms, and the design-automation software through which every advanced-node chip is designed all sit on the same demand curve. Investors focused on the AI theme have a far wider set of names to consider than a single-name accelerator trade would suggest.

What Could Sustain, or Disrupt, the Cycle?

Three forces could decide whether the cycle holds. The first is capacity and pricing: today's memory revenue rests on a supply crunch, and as new capacity comes online that dynamic can reverse. The second is the durability of AI demand itself, which rests on hyperscaler capital-expenditure plans that can be revised, and where one major customer slowing deployment would have outsized impact. The third is efficiency: the energy required per AI task can fall in the future, and if the hardware intensity of AI follows, demand growth could soften faster than expected. The same innovation that created the demand could, in time, temper it.

Projections already assume the cycle cools. Industry growth is estimated to slow from around 90 percent in 2026 to about 27 percent in 2027, with memory decelerating from roughly 250 percent to 32 percent¹. On those numbers, 2026 looks less like a new plateau than a step-change that then levels off, a very different thing for investors to underwrite.

Semiconductor Growth Forecast, 2026 vs 2027 (%YoY)

Source: WSTS Spring 2026 forecast.

Look Past the Headline

The trillion-dollar headline is real, but it is not the insight. The insight is that 2026's growth traces to one end market, AI infrastructure, and shows up most loudly in one segment, memory, which is also the most cyclical and price-driven part of it. That makes the industry easier to understand, and single companies and segments more exposed than the headline suggests. For investors, the lesson is to look beyond headline growth. Holding the theme in a way that does not depend on picking the name that leads next may matter as much as knowing how much of it there is.