TI 800 VDC architecture targets AI data centres
Texas Instruments has outlined a full power-delivery approach for next-generation AI data centres, built around NVIDIA’s emerging 800 VDC rack and facility design. The move is aimed at one of the more awkward problems in AI infrastructure right now: getting ever larger amounts of power into compute trays and GPU clusters without wasting too much of it as heat, copper and conversion overhead.
Why TI 800 VDC architecture matters
The basic pitch is straightforward enough. Rather than keep stacking conversion stages on top of a legacy AC and low-voltage DC chain, TI says its design cuts the route from 800 V to processor core rails down to two main stages. In practice, that means an 800 V-to-6 V isolated bus converter followed by a 6 V-to-sub-1 V multiphase buck stage close to the load. That matters because AI racks are heading into power territory where older 48 V and 54 V distribution schemes start to look increasingly cumbersome.
As previously reported by eeNews Europe when Nvidia pushed an 800 V power-distribution consortium for AI datacentres, the industry’s interest in higher-voltage DC distribution is largely about reducing current, cabling bulk, losses and service complexity as rack power rises towards the megawatt class.
TI 800 VDC architecture and the reference designs
TI’s announcement is less about a single chip than about filling out the blocks needed around an 800 VDC system. The company says the portfolio includes an 800 V hot-swap controller for rail protection, an 800 V-to-6 V DC/DC bus converter using integrated GaN power stages, and a 6 V-to-<1 V multiphase buck stage for advanced GPU cores. TI is also talking up a 30 kW 800 V AC/DC power supply, 800 V capacitor bank units based on supercapacitor cells, and an 800 V-to-12 V converter for tray-level power conversion.
The headline figure in the release is 97.6% peak efficiency for the 800 V-to-6 V converter, alongside power density above 2,000 W/in3. As ever with this sort of announcement, those are best read as reference-design numbers under stated conditions rather than a blanket promise for every deployment. Still, the direction of travel is clear enough: fewer conversion steps, more density, and less tolerance for dead weight in the power chain.
What it means for AI data centres
NVIDIA has been positioning 800 VDC as a practical route away from the multiple AC/DC and DC/DC stages that dominate current data-centre power trees. TI’s role here is to show that the semiconductor and power-conversion pieces needed for that transition are getting more concrete. The harder part, of course, is not drawing the architecture but getting operators, power-system vendors and rack builders to adopt it at scale.
So this is not a sudden revolution. It is another sign that the supply chain around AI infrastructure is trying to industrialise around much higher rack power, tighter thermal budgets and far less patience for inefficient legacy distribution. TI 800 VDC architecture is part of that broader shift, and it gives the company a clearer story in AI data centres than simply saying it sells power parts into the market.
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
