Analog Devices is in talks to acquire an AI power chip startup for roughly $1.5 billion, according to The Information. The deal would mark one of the largest plays yet by a legacy semiconductor giant to grab a foothold in the AI infrastructure layer that’s becoming as strategic as the GPUs themselves.
The Information reports the negotiations are active, though final terms haven’t been confirmed publicly. If it closes, Analog Devices, a Massachusetts-based chipmaker with deep roots in signal processing and power management, gets a fast lane into one of the hottest corners of the AI supply chain.
Why power chips suddenly matter
GPUs get all the headlines. But every Nvidia H100, B200, or custom accelerator sitting in a data center needs a specialized power delivery system to feed it stable, ultra-efficient voltage. That’s the job of AI power chips.
A few things changed the math here:
- Modern AI accelerators draw 700W to 1,200W per chip, several times what previous-generation server processors needed.
- Rack densities are climbing past 100kW, forcing data center operators to rethink power conversion from the wall to the silicon.
- Even a 1% efficiency gain at hyperscale translates into hundreds of millions in saved electricity and cooling costs.
Power chip startups have been quietly building IP around 48V-to-core conversion, integrated voltage regulators, and silicon-carbide and gallium-nitride designs that squeeze more watts out of less space. That’s the kind of capability $1.5 billion buys.
What stands out about this move
Analog Devices isn’t a pure-play AI company. It’s a $9 billion-revenue incumbent that competes with Texas Instruments and Infineon in analog and power. The fact that it’s willing to write a billion-dollar-plus check for an AI-specific power startup tells you where management thinks the growth is coming from.
This fits a broader pattern. Marvell, Broadcom, and Renesas have all been positioning around AI infrastructure, and Nvidia itself keeps absorbing networking and power-adjacent capabilities. The race isn’t just for the accelerator anymore. It’s for everything that surrounds it.
What practitioners should watch
If you’re building or buying AI infrastructure, three implications stand out:
- Pricing pressure on power components is going to ease, then tighten. Consolidation usually means fewer suppliers and stronger bargaining power for incumbents.
- Integration roadmaps will accelerate. Expect bundled power-and-compute reference designs from major chipmakers within 12 to 18 months.
- Smaller power startups become acquisition targets. Anyone with credible silicon-carbide or GaN-based AI server tech is now in play.
For anyone tracking the AI hardware stack, this is a signal worth marking. The center of gravity is shifting from “who makes the accelerator” to “who owns the full power-and-compute system around it.”
Full details at the original report from The Information.