The Semiconductor Market Is Not Recovering. It Is Restructuring.

The Number That Changes the Planning Horizon

Global semiconductor sales hit $110.5 billion in April 2026, a jump of 11% from March's $99.5 billion and 93.9% above April 2025's $56.9 billion. That year-over-year figure deserves scrutiny. April 2025 was a trough quarter; the base is soft. But the industry's month-on-month growth streak has extended to 14 consecutive months, which moves beyond bounce territory into structural shift.

SIA endorsed the WSTS Spring 2026 forecast projecting annual global semiconductor sales will grow by 90% to $1.5 trillion in 2026, with 2027 global sales projected to exceed $1.9 trillion. Two years ago, these numbers would have been dismissed as speculative.

Why This Cycle Is Different

Previous semiconductor supercycles rode consumer waves: PC buildouts, smartphone penetration, automotive electrification. All discretionary at the margin. The current driver is not.

Chips account for more than 95% of a leading AI server rack's content value and more than 50% of the total capital expenditures required for building and operating an AI data center, according to a June 1, 2026 report by SIA and Deloitte. When you buy an AI rack, you are buying semiconductors—everything else is cost multiplier on the chip.

A single state-of-the-art AI server rack contains more than 4,500 packaged chips. Not just GPUs. Memory, networking silicon, analog controllers, power management, transceivers, foundational logic. The demand spreads across node generations rather than concentrating in a single leading-edge product. That pressure lands simultaneously on TSMC's advanced nodes and second-tier foundries' mature lines.

Annual revenue from semiconductors deployed in global AI data centers could exceed $1.2 trillion by 2028, according to SIA and Deloitte. That represents a nearly tenfold increase over the last four years and surpasses total global semiconductor sales from 2025, across all end uses, by more than 50%. One end-market segment projected to exceed last year's total industry output. Even a 50% miss reshapes the sector.

The Demand Stack Behind the Numbers

Two engines run in parallel. First, AI datacenter buildout. The four largest hyperscalers — Amazon, Microsoft, Google, and Meta — are expected to approach $600 billion in capital expenditure in 2026, and Goldman Sachs projects total hyperscaler capex from 2025 through 2027 will reach $1.15 trillion, more than double the $477 billion spent from 2022 through 2024. Much of this capital is already committed. Construction sites do not pause chip orders because spot prices move.

Second, consumer and industrial recovery from the supply constraints of 2023 and 2024. Year-to-year sales in April were up in the Americas (115.8%), Asia Pacific/All Other (114.9%), China (78.6%), Europe (54.7%), and Japan (15.6%). Regional breadth rules out a narrow AI story. Smartphones and automotive are moving real volume.

The combination matters. Unlike prior cycles where AI demand eventually plateaus or capacity catches up, consumer recovery fills in behind AI, sustaining utilization across the node spectrum. Global semiconductor sales were $298.5 billion in Q1 2026, a 25% increase compared to Q4 2025. That sequential jump at scale signals something structural.

Who This Pressures

Foundries with leading-edge capacity control the negotiation. TSMC's advanced nodes remain the chokepoint for AI accelerators through 2027.

The harder position hits fabless design firms without locked wafer agreements. Spot capacity at leading nodes is gone. A leading AI chip manufacturer has secured approximately 800,000 wafers for its main chip in 2026, producing about 20 chips per wafer, suggesting roughly 16 million chips in total — that capacity was locked years back. Fabless companies that did not secure long-term agreements in 2023 or 2024 now chase residual allocation at premium prices.

For infrastructure operators and CIOs, the takeaway is straightforward: expect a price floor on hardware. AI infrastructure spending is real, but so are power limits, financing costs, supply constraints, utilization questions, and uncertainty over how quickly enterprises convert experimentation into profitable production workloads. Demand is durable; trajectory is not. Budget around the floor.

To meet global demand for AI, SIA and Deloitte estimate government and industry will invest more than $4 trillion in new data center infrastructure through 2028, with $2.8 trillion of this spent on semiconductors. Procurement decisions this quarter are locking supply relationships that shape infrastructure costs through 2029.

A Note on the YoY Figure

The 93.9% year-over-year number needs context. Monthly sales figures are compiled by the World Semiconductor Trade Statistics organization and represent a three-month moving average, which smooths volatility but also reflects a rolling February–April window. Spring 2025 was depressed; the comparison is soft. The absolute level — $110.5 billion monthly — carries more weight than the percentage. The direction and the streak matter most.

What to Watch

1. TSMC and Samsung gross margins, Q2 2026 earnings. Margin expansion confirms demand is outpacing capacity additions. Compression signals either spot availability is improving or customers are resisting price increases—either resets vendor leverage.

2. Mature-node utilization at second-tier foundries through Q3. If power management, analog, and networking demand fills non-leading-edge capacity, the bifurcation between AI-node haves and commodity-node have-nots weakens. Track GlobalFoundries and SMIC commentary.

3. Equity stakes and long-term wafer agreements. The structural tell for sustained supply anxiety is when design houses move from purchase orders to equity investments in foundry capacity. Any announcement in that direction confirms operators should model higher-for-longer component costs.

4. Hyperscaler custom silicon ramp timelines. The biggest structural question for chip suppliers is whether hyperscalers will eventually build enough of their own silicon to reduce third-party GPU demand. Watch Google TPU, Amazon Trainium, and Microsoft accelerator deployment fractions in quarterly disclosures. Acceleration there is the one plausible demand brake on a two-to-three-year horizon.

5. The $1.5 trillion full-year 2026 projection. At $110.5 billion in April and a 14-month growth streak, the annualized run rate supports the WSTS forecast. A break in the streak in May or June forces reassessment of the structural argument.