sic

I've been finding the third generation semiconductor space increasingly interesting lately, so I organized my research to share with you all.
Let’s start with something many may not have noticed: AI doesn't just need GPUs; it needs power.
Nowadays, an AI data center rack's power has surged from 10 kilowatts to several hundred kilowatts; traditional AC power distribution can't handle it. NVIDIA is pushing for an 800V DC power distribution architecture, converting municipal power directly to 800V DC fed into racks, cutting out several layers of conversion, boosting efficiency by 5%, halving copper usage, and reducing maintenance costs by 70%.
The core components of 800VDC are Silicon Carbide (SiC) and Gallium Nitride (GaN).

This company NVTS, let's drop a jaw-dropping figure: annual revenue of $45.9 million, market cap of $6 billion, P/S ratio of 131x. In the power semiconductor sector, the norm is 4 to 6x, and this one is over 25x that. This might be the most absurdly valued power chip company I've analyzed.
But why did it pump 10x in a year? Because it hit the right direction: AI data centers are freaking energy hogs.
NVIDIA's GB200 rack consumes 120kW; this juice has to be converted from high-voltage AC from the grid to low-voltage DC for the GPUs. If they lose just 1% efficiency in the conversion, the whole data center burns through millions more in electricity costs.