MPS expands AI power capacity while 800 V awaits qualification
Illustrative image: Unsplash contributor on Pexels
Monolithic Power Systems has a capacity problem to solve well before 800 V becomes a substantial business. Quarterly revenue reached $1.2 billion, including $506.7 million from enterprise data, up 164.6% year on year and management says that growth is coming from platforms already being built.
Its 800 V work is further back in the process. MPS is sampling GaN solutions and working with SiC, but management does not expect meaningful revenue before late 2027 and regards 2028 as more likely. That timetable is a useful corrective to the impression that every AI power investment is already being driven by 800 V distribution.
The demand arriving now comes from established power architectures, high-speed memory, accelerator boards and the surrounding conversion stages. MPS also reported its first orders for 48 V automotive products and a ramp in robotics sensing and motor control, so the capacity plan is wider than one data-centre voltage transition.
Capacity in a fabless business
MPS says its supply network can support more than $6 billion of annual revenue and is targeting $8 billion in the second half of 2027, followed by $10 billion. Those are revenue-support estimates rather than wafer-start figures and for a fabless supplier they depend on capacity across manufacturing partners, packaging and test as well as the inventory needed to keep products moving.
Expanding that network is a qualification job as much as a purchasing exercise. A new process or package route can change electrical behaviour, parasitics and reliability, so extra volume has to arrive through an approved manufacturing flow. Long production cycles then force inventory decisions before customer demand is certain.
GaN has a reliability hurdle to clear
Management’s comments on GaN were more cautious than the usual infrastructure launch language. It described a reliability target below one part per million failures and the need to build substantial shipment experience, a demanding requirement when customers are contemplating large fleets of high-power systems. The target is not evidence that the company has already achieved that field performance.
MPS is developing both GaN and SiC options, leaving the choice to the converter design and customer requirements. Distribution at 800 V does not dictate a single switch technology, because topology, device voltage rating, switching frequency, isolation and fault handling all influence the result. Any efficiency or size advantage has to survive the complete converter comparison.
Testing 800 V equipment in its own building should help MPS accumulate operating experience, though it cannot substitute for customer qualification across different loads and installations. Management’s revenue timetable allows for that work rather than assuming a sampling announcement will turn quickly into volume shipments.
Current demand sets the investment pace
Enterprise-data revenue also rose 33.2% quarter on quarter as existing customers increased purchases. Serving that business while preparing a later voltage transition gives the capacity expansion two different timescales, with current orders funding the business while future architectures still require qualification.
The risk is committing too early to a product mix that changes. Automotive 48 V orders and robotics programmes broaden MPS’s demand base, but they do not make every manufacturing route interchangeable. The capacity targets will be more informative as the company discloses how much is qualified, available and used.
Sources
Photo: Abolfazl Pahlavan / Pexels. Illustrative image. https://www.pexels.com/photo/electronic-circuit-board-with-various-components-33813265/ | https://www.pexels.com/license/
