When AI shifted from experiment to industry, data centres became the new factories. They power the training runs, host the inference services and consume electricity at a scale that keeps CFOs awake. Vertical Semiconductor, an MIT spinout, is betting on a small but radical fix: change how electricity reaches AI chips, not the chips themselves. That shift could cut cost, heat and waste right where it hurts most.
The Spark
Cynthia Liao remembers when the problem became impossible to ignore. In an MIT lab, she and her co-founders kept hitting the same wall. GPUs were getting faster and more efficient, but the power delivery modules and voltage converters around them still worked the old way. The result: racks that demanded more cooling, space and energy for the same output.
Vertical’s answer lies in material science. By using vertical gallium nitride (GaN) transistors to convert and route power directly at the server level, the company aims to deliver massive efficiency gains. The change sounds microscopic, but such improvements can shift entire industrial margins.
The Early Backing
Investors have noticed. An $11 million seed round led by Playground Global gave the team room to build prototypes and move toward commercial samples. The pitch is simple in theory but complex in execution. If Vertical’s chips can cut heat and shrink power conversion footprints, hyperscale data centres will see lower operational costs.
It’s perfect timing. Nvidia and cloud giants are rethinking hardware around AI workloads, creating space for infrastructure upgrades that make computation more efficient.
Promise and Pressure
There’s a clear playbook and a clear risk. The playbook: identify a high-value bottleneck, apply novel materials engineering, and target the hyperscaler market through licensing or direct supply. The risk: scaling GaN device fabrication, proving 24/7 reliability, and integrating into existing server ecosystems.
GaN is not new; it powers some of today’s best chargers but moving it into vertical transistor layouts for data centres is uncharted. Competitors like Infineon and other GaN startups are also in the race. Vertical’s edge is structural innovation and academic pedigree, but commercial traction still has to be earned.
Execution on the Ground
Right now, the focus is practical:
Prototype testing with system architects to validate thermal and reliability metrics under production loads.
Partnerships with manufacturers who can translate wafer runs into volume production.
The seed money covers the first goal. The second will need far more capital or industrial alliances. Playground’s involvement signals that patient, hardware-savvy investors are in for the long haul.
Why It Matters
The world’s data centre boom driven by cloud adoption and digital services faces the same power and cooling challenges as Silicon Valley. Lower-loss power conversion could dramatically cut operating expenses and ease cooling loads in dense urban centres. If Vertical’s tech proves itself, it won’t just help hyperscalers, it will benefit local colocation providers that live and die by energy efficiency.
The Road Ahead
Verticals can’t rush. Moving from lab demo to production-grade hardware demands time, validation and trust. The next 12 months will decide whether the startup’s prototypes hold up under real-world stress. Success will lead to a Series A and deeper fab partnerships. Failure will be a reminder of how tough lab-to-fab transitions can be.
The Takeaway
Vertical Semiconductor represents the kind of hidden infrastructure innovation that quietly shapes massive industries. A small shift at the transistor level could ripple into global cost and energy savings. If the startup delivers, the payoff will be huge. If not, it will still leave behind lessons about how deep-tech ideas evolve from research to real impact.
