University of Oxford spinout Densix is developing a power converter with output two orders of magnitude higher than current technology. It may prove critical to meeting the needs of AI data centres.
University of Oxford spinout Densix is betting the future of electrification and AI will not just be decided in data centres and substations but inside an unassuming metal box: the power converter.
Backed by Oxford Science Enterprises and built on IP from the University of Oxford’s Department of Engineering Science, Densix is developing a new class of power converters that can deliver much higher power in the same footprint as today’s devices without the need for brand new materials.
The startup’s patented design pushes power density to new limits and targets sectors at the centre of the AI build-out: data centres and the electricity grid.
Re-engineering old technology
Today’s power converters and transformers are based on a concept that has been around for 100 years. It a very mature technology that has been optimised for durability. Power converters are everywhere: in laptop chargers, street-level substations and building basements.
Densix’s technology doesn’t replace these with a new material; instead, it re-engineers how they are designed.
CEO Anders Hakfelt gives the example of a typical laptop charger. It takes in around 220V AC and outputs 20V DC, with a power capacity of about 100 watts (0.1kW). Densix’s design takes the same size of charger but can deliver 10,000 watts, a roughly 100 times increase in power density.
“Now, the cleverness here is that it is not like a superconductor or is chilled with liquid nitrogen or anything like that. It’s a very clever design and very clever thermal management,” says Hakfelt.
In March, the spinout raised $4m in seed funding, led by Oxford Science Enterprises, an investment company that backs University of Oxford spinouts. UK spinout investor Parkwalk Advisors and the University of Oxford Innovation Fund also participated in the funding round.
The company is targeting data centres initially. The constraints that data centres face are mounting rapidly with the growth of AI. It is taking increasing amounts of energy to power the processors in server racks for AI inference and the training of large language models.
Power consumption about to rocket
Nvidia, the maker of GPU’s widely used in AI computing, plans to deploy processors in server racks later this year that use four times as much compute power than its top-end GPUs use today.
“What they’re rolling out now that will be deployed later this year are racks that use 600kW. So, all of a sudden, in the same footprint, you need to bring in a vast amount of power, and that creates all sorts of problems,” says Hakfelt.
Issues data centres and power systems could face without greater power density and efficiency include overloaded systems and overheated cables.
“Being able to take power from the grid, convert it really efficiently and put it to the server, if you will, at the right voltage level, in the right place, is really valuable,” says Hakfelt.
The company also sees power grid operators as first customers of its technology. The electricity grid is undergoing transformative change with the electrification of the economy. As more renewable energy is integrated onto the grid and fast-charging electric vehicles are increasingly deployed, regional distribution networks will become more vulnerable to overload.
The cascade of power failures in the electricity grid that caused blackouts in Spain, Portugal and France in April last year gave an early warning sign of the vulnerabilities in regional power systems.
While increasing amounts of renewable energy comes online and transportation systems move towards electrification, infrastructure upgrades are not keeping up.
“If you want to increase the capacity of an office block in central London or wherever, often you have an old school transformer sitting in a room in a basement doing the work in a limited space.
“All of a sudden, you want to connect rapid charging points for 10 EVs – you don’t have the capacity unless you can upgrade in the same footprint but deliver more power in the same place,” says Hakfelt.
With its new seed funding, the company plans to set up a lab where it can develop its products and recruit a team that can support its growth.
Hakfelt emphasises that the company’s products are designed for the systems of the future rather than those of today. It will seek to be part of the conversation on how to develop standards for infrastructure that will be integral to the transformation of an AI- and electrification-driven economy.
“Our approach is to develop for the future not for the past, and that means our designs are geared towards the standards of tomorrow. Those standards, in some cases, are still being shaped. Which means we need to be part of those working groups.”
Read our latest Energy sector report
Oil and gas company investments in startups sees uptick
Last quarter capped off a year that saw venture investments by oil and gas companies start to recover from what was a nearly 50% collapse in dollars invested between 2022 and 2024.
