- UK startup wants to bring compute even closer by embedding micro data centers inside lampposts
- Self-destructing Nvidia chips will ensure sensitive information remains private
- Some challenges need to be addressed before this becomes a reality
UK startup Conflow Power Group Limited (CPG) has proposed a major shakeup to the world’s data centers by embedding micro units directly within urban street infrastructure such as lampposts, rather than concentrating compute inside the hyperscale facilities as we know today.
Besides dispersing compute across cheaper and more manageable micro locations, the plan also centers around local solar energy generation and battery backup systems to tackle one of the biggest criticisms data center campuses face – sustainability and environmental impacts.
Under the new proposals, CPG aims to bring AI compute closer to users and devices, which would reduce latency and ease pressure on national telecoms infrastructure.
Future data centers could be located just outside your front door
The BBC reports $2,000 Nvidia AI accelerators could be used instead of high-end and flagship GPUs like H100 and B200 systems, which cost tens of thousands of dollars per unit.
Using self-destructing chips is also a noteworthy addition to the scheme, with Nvidia including firmware locking, encryption and other anti-tampering protections that can effectively disable hardware if compromised, relocated or accessed by unauthorized methods or people.
Such anti-tampering technologies already exist for export compliance, and are generally seen when AI accelerators are sold into restricted or edge deployments to ensure maximum safety.
Tens of thousands of micro data centers could spread compute throughout cities, with each one handling localized AI workloads that could span applications like traffic monitoring, CCTV, autonomous vehicle coordination, telecoms, environmental sensing and more.
Though the idea of embedding data centers into lampposts might be a novel one, bringing AI to the edge is a trend that’s accelerating as AI workloads become more sensitive to bandwidth and latency. Putting the compute physically closer to where it’s being applied would also likely reduce costs associated with data transfers.
The energy implications are also compelling, with hyperscale data centers facing constraints from delayed grid connections and unsustainable supplies. Spreading out power consumption geographically certainly addresses the current bottlenecks, with some of the biggest campuses consuming as much electricity as small towns.
Lampposts in particular are appealing due to their existing electrical connections, their dense distribution in urban environments like towns and cities, and the fact that many are also already connected to fiber networks.
Many have already been multipurposed into small 5G cells, traffic camera mounts, Wi-Fi access points and electric vehicle chargers.
Interestingly, there are also significant geopolitical benefits to positioning micro data centers across existing street infrastructure networks. Increasing sovereign compute concerns in Europe and the UK are causing governments to drive forward schemes that support more local processing.
For these to work at scale, CPG would have to address the many challenges that come with embedding compute within lampposts, though. Existing infrastructure may need upgrading to ensure protection against the weather and vandalism. There are also concerns around thermal management, with large-scale data centers slated for their excessive water consumption for cooling.
While upgrading existing infrastructure may not be economically viable, building multipurpose networks in the future could allow a single element, such as a lamppost, to serve far more functions than we’d previously imagined.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.
https://cdn.mos.cms.futurecdn.net/ZMxP3gbMxCajQNH9F5W7jN-1920-80.jpg
Source link
