The Deal: $1.5 Billion for the Power Behind AI Chips
Chipmaker Analog Devices is in advanced talks to buy startup Empower Semiconductor for about $1.5 billion, in a deal that reflects demand for technology that can manage the intense energy needs of AI chips. Empower Semiconductor, founded in 2014, has spent a decade developing power management integrated circuits that solve one of the most technically demanding problems in modern electronics: delivering the precise, stable, high-current power that AI accelerators require at the nanosecond timescales that GPU and NPU architectures demand. The company's technology sits at the intersection of two trends that are simultaneously driving semiconductor M&A activity: the exponential growth in AI compute density, which is creating power delivery challenges that conventional power management architectures cannot address, and the intensifying competition among established analogue semiconductor companies to secure positions in the AI infrastructure supply chain before the market structure consolidates around a handful of dominant suppliers.
The $1.5 billion price tag — for a 12-year-old startup with fewer publicly disclosed revenue figures than its acquirer would typically prefer — reflects the strategic premium that Analog Devices is willing to pay for capabilities it cannot build fast enough organically. The acquisition would expand Analog Devices' portfolio of chips that support AI, whose complex workloads require massive amounts of energy. Power management may be the least glamorous component category in the AI infrastructure stack, but it is arguably the most physically consequential. A GPU cluster that receives poorly regulated power — with voltage droops, overcurrent events, or insufficient current delivery speed — does not merely perform suboptimally. It fails, generating the kind of operational reliability problems that hyperscalers cannot tolerate in data centres running at utilisation rates above 90%. Clean, fast, precise power delivery is the silent prerequisite for the AI compute performance that the market is building around.
Why Power Management Is the AI Infrastructure Bottleneck Nobody Talks About
The energy requirements of modern AI accelerators have created a power delivery challenge that is physically different in kind from anything the electronics industry has previously had to solve in volume. Nvidia's H100 GPU draws up to 700 watts in peak compute operation. The B200 draws up to 1,000 watts. GB200 NVLink systems, which combine multiple B200 chips with network switching, can draw 10 kilowatts from a single rack unit. At these power levels, the physics of power delivery become engineering constraints: the current required at low voltage (typically 0.7-1.0V for modern logic chips) is enormous, and delivering it from a data centre power distribution system to a chip package without unacceptable resistive loss requires power management circuitry that is faster, more efficient, and physically closer to the chip than anything the previous generation of high-performance computing required. The industry term for this problem is "power integrity," and solving it for AI-scale compute is one of the most commercially valuable engineering capabilities in the semiconductor supply chain.
Empower Semiconductor's specific contribution to this problem is a switched-capacitor power converter architecture that delivers higher efficiency and faster transient response than conventional inductor-based voltage regulator designs. The technical advantage matters because AI chip workloads are characterised by extreme power transients — the current demand can swing from near-zero to full rated load in microseconds when a new batch of inference requests arrives. A power delivery system that cannot respond fast enough creates voltage excursions that force the chip's internal protection circuits to throttle performance, directly reducing the compute throughput that data centre operators are paying for. Solving the transient response problem at the power density required for modern AI accelerators is Empower's core technical claim, and it is the capability that Analog Devices has decided is worth $1.5 billion to acquire rather than develop.
Semiconductor M&A in 2026: The Race to Own the AI Infrastructure Stack
The Analog Devices-Empower deal is one transaction in a consolidation wave that is systematically restructuring the analogue and mixed-signal semiconductor landscape around AI infrastructure requirements. The pattern is consistent across multiple deals: established analogue semiconductor companies — whose core competencies in precision measurement, signal conversion, and power management were built for industrial, automotive, and communications applications — are acquiring startups with AI-specific implementations of those same capabilities, gaining both technology and the engineering talent needed to serve hyperscaler customers whose procurement processes, technical requirements, and qualification timelines are different from traditional industrial buyers.
U.S. national labs are looking beyond traditional chip suppliers as the AI race reshapes high-performance computing, exploring new semiconductor players for supercomputers as AI workloads demand different architectures, higher memory bandwidth, and faster, more energy-efficient processing. The government procurement signal reinforces the commercial trend: across both the hyperscaler and government high-performance computing markets, the architectural requirements of AI are forcing a re-evaluation of established supplier relationships that have been stable for decades. Companies that have supplied power management, signal conditioning, and precision measurement components to defence contractors and research institutions are suddenly competing — and sometimes losing — to startups with AI-optimised architectures built from the ground up for GPU-scale power delivery. The Analog Devices acquisition of Empower is a direct response to this competitive threat: better to acquire the capability at a $1.5 billion premium than to watch a competitor absorb it and use it to displace Analog Devices from the AI data centre supply chain it is trying to enter.
The Broader Supply Chain Implication: Power as the AI Infrastructure Constraint
Positioning the power management supply chain as a strategic asset — rather than a commodity procurement category — is a recognition that the physical constraints on AI infrastructure scaling are increasingly concentrated in energy delivery rather than compute silicon. Nvidia can manufacture more GPUs. TSMC can add wafer capacity. Memory manufacturers can expand HBM output. But the power infrastructure required to run those components — from the grid connection to the data centre, through the facility's power distribution system, down to the chip-level voltage regulation — is subject to constraints at every stage that are not solved by semiconductor investment alone. The Analog Devices-Empower deal signals that the semiconductor industry has identified power management as the layer of the AI infrastructure stack where strategic M&A can secure durable competitive position — and that the window for acquiring power management capability at startup valuations is closing as the strategic value of the category becomes visible to every potential acquirer simultaneously.