The bottleneck nobody is pricing
Analog Devices announced on Tuesday that it will acquire Empower Semiconductor for about $1.5 billion in cash - the same day the company is set to report Q2 fiscal 2026 earnings. The timing isn't accidental. It is a signal about where ADI thinks the next phase of AI infrastructure demand is heading. And the answer has nothing to do with compute.
The market is still obsessed with who makes the fastest GPU. But the real bottleneck in AI infrastructure is shifting - from silicon performance to power delivery. Every new generation of AI accelerator draws more current, faster, in tighter bursts. Nvidia's latest chips demand 1,000-plus watts per die; entire server racks are demanding more than 100 to 1,000 kilowatts per rack. You can design the most advanced processor in the world, but if the voltage collapses the moment it tries to draw a spike of current, the chip throttles. The compute is useless.
That is what Empower Semiconductor does: it solves the problem of getting enough clean power to AI chips, fast enough, in small enough packages, that the processor never has to wait. They are not on everyone's radar. Now they should be.
Vertical power delivery is the architecture shift no one is tracking
Empower's core technology is something called integrated voltage regulation - IVRs - combined with silicon capacitor technology. Put plainly: instead of running power from a board-level regulator across a copper trace to the chip (which introduces resistance, delay, and voltage droop under load), Empower's regulators sit directly beneath the AI processor in the stack. Power arrives vertically, in millimeters instead of centimeters.
For a $1,500-watt GPU drawing 8,000-plus amperes in transient bursts, that distance matters enormously. It means the voltage stays stable when it needs to. It means higher efficiency. It means smaller board footprint and fewer discrete components - which matters when server manufacturers are packing eight or more GPUs into a single rack.
Empower's Crescendo vertical power delivery platform, showcased at APEC 2026 earlier this year, was explicitly designed for this kilowatt-class AI and HPC workload. The company's CEO Tim Phillips - who founded Empower in 2014 - has been public about the thesis: power design now has to start at the chip level. The architecture is inverted.
This is a market structure transition I've been watching. The industry spent years optimizing for compute density and memory bandwidth. The next constraint is electrical. As chips get more powerful, the power delivery network becomes the thing that holds them back - unless you redesign the entire stack.
Why ADI, and why now
Analog Devices is a $202 billion company with $3.16 billion in revenue last quarter - up 30% year-over-year. Data center revenue alone grew approximately 50% in fiscal 2025, driven by AI server deployments. ADI already makes power management chips, precision monitoring components, and system-level analog solutions. But Empower's IVR architecture is a different animal. It is not an incremental improvement to ADI's existing portfolio. It is an entirely new layer of power delivery that lives inside the chip stack.
Empower raised over $140 million in its last funding round and remained closely held. At $1.5 billion, the acquisition is roughly 0.75% of ADI's market cap - small enough that it doesn't distort the balance sheet, large enough to signal strategic conviction. The deal is expected to close in late 2026.
What this tells me is that ADI's management has identified the power delivery bottleneck as the next growth vector in the AI stack - and they are unwilling to wait for an internal team to catch up. Empower's technology is already in development for AI processors that don't ship yet. Buying the company gives ADI first-mover positioning before this architecture becomes table stakes.
The competition is already moving
Here is the part that makes the timing urgent. ADI is not the only player converging on this problem.
Marvell introduced its Package Integrated Voltage Regulator (PIVR) solutions in mid-2025, targeting the same on-package power delivery space. Infineon has been collaborating with Nvidia on next-generation AI server rack power delivery architecture. Multiple smaller startups are working on GaN-based vertical power solutions. The integrated voltage regulator market itself was valued at $6.26 billion in 2025 and is projected to grow at over 6% annually through 2030.
The window to establish architecture leadership is narrow. Once a power delivery standard becomes embedded in a hyperscaler's server design - or co-optimized into a foundry's packaging flow - switching costs become enormous. That is why this acquisition is not about today's revenue. It is about owning the reference architecture for the next generation of AI systems.
However - the earnings question
There is a counter-signal I need to address. Tomorrow - May 20 - ADI reports Q2 fiscal 2026 earnings. Analyst consensus expects revenue near $3.5 billion and EPS of roughly $2.90. The Q1 beat was already a strong one: $3.16 billion in revenue with broad-based growth across all end markets, and operating cash flow that surprised to the upside.
The risk here is straightforward. If Q2 shows deceleration in data center AI growth - even a modest slowdown from that 50% fiscal-year pace - the market may question whether the $1.5 billion Empower bet is premature. Power delivery demand is derivative of AI chip demand. If hyperscaler capex softens, or if Nvidia and AMD face a cycle pause, Empower's addressable market shrinks with it.
I do not believe the long-term thesis is at risk. AI server power density will only increase. But the near-term setup requires ADI's own growth story to hold. The acquisition announcement timed alongside earnings is a classic move: use a strategic catalyst to frame the quarter. Whether the underlying numbers support the narrative is the question.
Where the capital goes
The debate is not whether power delivery becomes the next constraint layer in AI infrastructure. It is whether ADI can execute on the integration fast enough to matter.
I believe ADI is on the right side of the transition from compute-first to power-first thinking. The Empower acquisition gives them architecture depth that Marvell and Infineon do not yet match in the vertical IVR space. But the return profile depends on two things: tomorrow's earnings holding up, and the successful integration of Empower's technology into ADI's data center product roadmap over the next 18 to 24 months.
For me, this is a "small allocation, watch closely" setup. ADI at $202 billion is no longer the obscure analog company it was two years ago. The 30% revenue growth and data center momentum are real, but they are also increasingly reflected in the multiple. The Empower deal adds a long-term optionality layer - the kind of architectural play that compounds over years, not quarters.
If ADI's Q2 data center growth stays above 40%, and if the company can demonstrate a credible path to shipping Empower-derived IVR products by 2027, I would be comfortable increasing allocation. If data center growth decelerates below 25%, or if the integration timeline slips into 2028 and beyond, the opportunity cost begins to look less favorable compared to alternatives deeper in the AI supply chain.
The AI infrastructure transition is not over. It is just moving to a different layer. Power delivery is that layer now. ADI seems to understand that before most of the market does.