Navitas Semiconductor: CEO Says Mobile Business Is "Essentially Gone" as Claros Acquisition Doubles AI Power SAM
Citi TMT Conference, September 8, 2026
Navitas Semiconductor CEO Chris Allexandre used his appearance at Citi's 2026 Global TMT Conference to declare the company's year-long pivot away from consumer electronics essentially complete, while laying out a four-stage roadmap for power semiconductor adoption in AI data centers that pushes the company's addressable market well beyond what investors have previously modeled.
Mobile Exit Nearly Complete, AI Infrastructure Now a Third of Revenue
Allexandre, who took over as CEO roughly a year ago and launched what he calls "Navitas 2.0," said the transformation from a mobile-focused gallium nitride supplier to a high-power semiconductor company serving AI data centers and grid infrastructure has largely run its course. Mobile represented as much as 85% of Navitas revenue at the start of the transition. By year-end, he said, it will be "insignificant," falling to low single digits as a percentage of sales, even though the legacy mobile business has continued to grow double digits per quarter. AI infrastructure, by contrast, is expected to exceed one-third of revenue by year-end. The company has also built backlog extending into 2027, something Allexandre noted "was not the case 18 months ago" when the business was still mobile-centric. He argued that investors have not fully priced in this compositional shift because reported revenue is roughly flat versus 18 months ago even as the underlying mix has changed dramatically, and suggested two or three more quarters of consistent double-digit growth should resolve that disconnect.
A Four-Stage Roadmap to 800-Volt Data Centers
The most substantive new framework Allexandre offered investors was a detailed, sequenced explanation of how data centers migrate from today's 50-volt architecture to an eventual 800-volt standard, a process he insists is gradual rather than a single "digital switch." The first inflection, already underway, involves replacing silicon with silicon carbide in AC-DC conversion as power density rises within existing 50-volt racks. The second stage, arriving in the first half of 2027, is the so-called "sidecar" architecture, where AC-DC and DC-DC power shelves move outside the compute rack into an adjacent enclosure running at 800 volts, while the rack itself retains its 50-volt busbar. The third stage, expected in the second half of 2027 into early 2028, makes the rack itself 800-volt native, pulling DC-DC conversion inside the compute rack and driving heavier GaN content. The fourth and most distant stage, 2028 and beyond, involves solid-state transformers replacing traditional grid transformers to deliver 800 volts directly to data centers, a shift Allexandre said is gated by grid reliability and security concerns rather than technology readiness. "Would 800 volt have happened if NVIDIA didn't drive the transformation? The answer is yes, but way later than it's happening now," Allexandre said, crediting hyperscalers and xPU vendors with accelerating a transition that would otherwise have taken much longer.
Claros Acquisition Nearly Doubles the Addressable Market
Perhaps the most important disclosure for investors was Allexandre's sizing of the incremental market opportunity created by Navitas' pending acquisition of Claros, a vertical power delivery and integrated voltage regulator (IVR) company, expected to close around October. Navitas has previously sized its GaN and SiC opportunity inside U.S. data centers and grid infrastructure at $3.5 billion by 2030, plus roughly $1 billion for JFET-based circuit protection, for a combined $4.5 billion serviceable market. Allexandre said the IVR and vertical power delivery opportunity addressed by Claros "is basically adding as much as that," effectively doubling the total addressable market as power conversion moves closer to the GPU or xPU itself. He was careful to explain the technical distinction underlying the deal, noting that vertical power delivery refers to physically relocating discrete MOSFETs, capacitors and inductors into a module mounted behind the circuit board, while integrated voltage regulation is a more fundamental shift toward putting control, FETs, drivers, capacitors and inductors onto a single monolithic piece of silicon. Claros' technology, he said, uses a "3.3-volt entry, sub-1 volt output, 40 amp" chip with no external capacitors or inductors, deployed in a scalable mesh array that can be multiplied to meet the current demands of different xPU architectures. Allexandre pointed to Analog Devices' acquisition of Empower as evidence that the broader power semiconductor industry is converging on the same conclusion, that MOSFET-based delivery cannot keep pace with the current requirements of next-generation AI processors. He was explicit that this shift represents a 2028-2029 revenue driver rather than a change to the near-term plan, noting that Claros' own CEO has cited 2028 as the ramp timeline for its lead xPU customer, and stressed that the acquisition "doesn't change our organic plan" for 2026 and 2027, which remains driven by the earlier SiC and GaN inflections.
Content Per Megawatt and the Power Wall
Allexandre offered granular modeling assumptions that investors can use to triangulate revenue potential, based on an industry deployment assumption of 220 to 300 gigawatts of data center capacity by 2030, a range he said is "getting closer to 300 by the day." Within that framework, GaN content inside the data center is modeled at roughly $15,000 per megawatt, replacing silicon in DC-DC conversion. Silicon carbide contributes another $10,000 per megawatt inside the data center for AC-DC conversion and JFET-based protection, plus a further $10,000 to $15,000 per megawatt in grid modernization applications running at higher voltages. That totals roughly $25,000 per megawatt today, which Allexandre acknowledged is "still a small portion" of total power semiconductor content per megawatt, which he estimated at $100,000 to $150,000 including silicon that will persist for years. He also introduced the framing of a "power wall" now constraining AI infrastructure performance, arguing that after compute and memory bottlenecks were addressed, power delivery and copper-to-fiber interconnects are the next limiting factors. "Many xPU companies are underclocking their processor because they can't get enough power delivered to the processor," he said, a comment that underscores why hyperscalers are now directly involved in power architecture decisions rather than leaving them entirely to merchant power suppliers such as Delta, Flex, LITE-ON, Vertiv and Eaton.
Fabless Strategy Defended Against IDM Competitors
Asked about competing against integrated device manufacturers, Allexandre pushed back firmly on the notion that owning fabs is necessary to compete in power semiconductors, pointing out that Navitas uses Global Foundries and previously TSMC for GaN, X-FAB for SiC, and that Claros works with Samsung on 14-nanometer process technology. "I know other companies... that have their own fab and that they are in the world of hurt right now, okay, with underutilization," he said, without naming names, arguing that fabless and fab-lite models offer more financial flexibility during periods of rapid technology change, provided the company retains proprietary process and device-level differentiation with its foundry partners.
Allexandre closed by reiterating that the biggest misconception among investors is treating the 800-volt transition as a single event tied to a specific NVIDIA product launch. "This is not winning the iPhone," he said, framing the opportunity instead as a multi-year sequence of overlapping technology transitions across AC-DC conversion, sidecar architectures, native 800-volt racks, vertical power delivery, and eventual grid modernization, each representing a distinct content-expansion opportunity that Navitas must compete for individually.