Every leap in AI now rests on a physical bill: the chips, memory, interconnects, and power packed inside the data centre. Against this backdrop, Yole Group releases its latest report, Data Centre Semiconductor Trends 2026, offering a comprehensive analysis of the semiconductor market, ecosystem, supply chain, and technologies shaping the data centre semiconductor landscape.
An acceleration that keeps outpacing the industry
AI has become the semiconductor industry’s single growth engine, and it is compounding faster than any prior cycle. Each Yole Group edition since 2024 has captured a further step change in demand, with the market set to roughly double in 2026 alone. Yole Group’s analysts now size it at $1.5 trillion by 2031, the point at which semiconductor value starts growing faster than the data centre capex that drives it, itself heading toward roughly $1.8 trillion by the end of the decade.
“AI has turned the data centre into a full-system problem. Performance is now sold by the rack, not the chip, and the winners will be decided by how memory, advanced packaging, optics, and power come together,” said Eric Mounier, Chief Analyst at Yole Group.
Four segments redrawing the map
Yole Group’s new edition tracks how the semiconductor content of an AI data centre is shifting, segment by segment:
- Logic is still the largest slice of the market. GPUs dominate today, but the custom AI ASICs that hyperscalers co-design are the fastest-growing challenge to that lead
- Memory is the swing factor of 2026: surging HBM demand and a steep rise in server DRAM prices have made it the fastest-growing segment of all
- Optical is moving to the centre of the design. As co-packaged optics reach the market in 2026, interconnect is becoming as critical to AI performance as compute itself
- Power is where a new layer of value is forming, as racks move to higher-voltage architectures to feed ever-denser AI systems
From chip to rack
Underpinning this is a shift from the chip to the rack. AI performance is now sold at the full-rack level, in systems such as NVIDIA’s NVL576, AMD Helios, AWS NL72, and Google Ironwood, and the real bottleneck at scale has become advanced-packaging capacity, alongside a fast-moving HBM roadmap.
Concentration, regionalisation, and a widening power gap
Concentration and regionalisation are reshaping the supply chain. No single company covers the whole chain, though NVIDIA has grown into a supply chain powerhouse, capturing ~16% of all semiconductor revenue and locking in priority capacity at TSMC, SK hynix, and packaging partners. Regional paths diverge: Europe backs sovereignty with a €200 billion InvestAI plan, the United States is defined by NVIDIA’s dominance, and China’s build-out stays gated by SMIC under US export controls.
But power is fast becoming the industry’s real ceiling. With AI racks approaching 1MW by decade’s end, higher-voltage delivery (800V, SiC and GaN) and direct liquid cooling are becoming standard, and grid access is now as much a constraint as the chips themselves.
With AI a durable, long-term growth driver, the technology, supply chain, and competitive dynamics behind this market have never mattered more to chipmakers, memory, and packaging suppliers, hyperscalers, ODMs, and investors. Data Centre Semiconductor Trends 2026 provides the market intelligence and technology insight to navigate it and spot the next opportunities.
