AI Data Center Expansion Squeezes Power Chip Supply, Auto Electronics May Suffer
nashnova research
Surging AI data-center demand is stretching power-semiconductor lead times to 30 weeks at some major suppliers, as capacity is re-prioritized toward data centers and industrial use, leaving automotive and consumer electronics at risk of being crowded out.
How did power chips become the new bottleneck?
The shortage story has moved past GPUs and memory. It is now spreading down the supply chain to power-management chips — the less glamorous devices that regulate voltage and deliver current.
This means → AI data centers don't just consume compute silicon; they are pulling the entire power-delivery ecosystem tight.
Lead times at some major IDMs — integrated device manufacturers that both design and fabricate chips — have reached 30 weeks, while mature-node foundry capacity is tightening in parallel.
Why does the data-center power architecture need an overhaul?
Infineon warns that the current 54 V rack-level power architecture is hitting a capacity ceiling and recommends a shift to a centralized 800 V DC architecture.
In plain terms = today each server rack steps voltage down on its own, like every apartment having its own transformer. An 800 V bus delivers high-voltage power to the building first, then distributes — cutting losses and raising efficiency.
The upgrade requires silicon carbide (SiC — a semiconductor material that handles high voltage and heat), gallium nitride (GaN — a high-frequency, high-efficiency power material), and conventional silicon devices working in combination, multiplying chip demand.
What opportunity does Texas Instruments see?
TI notes that the power-management system in a single AI server contains more than 1,000 analog and power-related chips.
This means → every additional AI server deployed consumes several times the power-chip count of a conventional server, giving TI entry points across voltage regulators, high-voltage converters, and GaN products.
TI also acknowledges, however, that supply constraints across the broader analog-chip market are intensifying — opportunity and bottleneck are arriving together.
Whose capacity gets squeezed out?
When major international chipmakers allocate capacity, industrial applications and power infrastructure are now listed as top priority.
This reflects a zero-sum trade-off: total capacity cannot expand quickly in the short term, so the higher-margin, faster-growing data-center and industrial segments get served first — automotive and consumer electronics are pushed further back in line.
Taiwanese suppliers may pick up overflow orders, but only if they can secure enough foundry capacity — and average utilization at 8-inch wafer fabs has already climbed back to nearly 90%, leaving little slack.
Has a real shortage arrived yet?
Analysts are split. One camp argues that a severe shortage requires sustained AI demand growth plus a simultaneous auto-and-industrial recovery — conditions not yet met.
The other camp points to multiple research firms projecting that power-semiconductor demand growth will outpace supply growth for years to come.
In plain terms = the market is "tight but not yet in outright shortage." If both demand engines fire at once, shortages and pricing pressure will ripple further downstream across more end markets.
市场有风险，内容仅供研究参考，不构成投资建议。