What is interesting about AI is shifting. Away from large centralized data centers and towards AI chips embedded in devices at the grid edge. And away from the general and towards highly specific applications. The AI cameras in our driveway recently picked up a Eurasian lynx this summer, which exemplifies both shifts, as discussed in our latest video below.
Last year, the power consumption of large AI data centers and their reliance on various forms of gas generation seemed like the most interesting debates in energy. But this year, we are sensing many decision-makers getting tired of the debate.
The interconnect pipelines of major transmission operators and utilities have become so massive that seemingly no one believes the numbers anymore. Our own US load growth forecasts are, more soberly, at 1-3% pa.
At the same time, the most interesting part of AI, seems to have moved away from “AI in general” to “AI in particular”, across edge devices and real-world deployment, in the seismic industry, residential heating, for power transmission networks or the circular economy.
This video illustrates the point. We picked up a lynx in our driveway last week, via 2 x €180 AI cameras from Unifi, which is thus one of the best ‘investments’ we have made in our home.
There isn’t even a data center behind these AI cameras. The compute lives directly in the cameras. We estimate our 2 x AI cameras consume 100kWh of electricity per year, or +1% of our total household energy use.
It is the same in most of the autonomous vehicles and advanced robotics concepts that have been exciting us. The compute is mostly embedded within the device, and syncs up with a central data center only secondarily.
Ultimately, the compute from autonomous devices will likely even become larger than the compute within centralized data centers. If 2bn cars and autonomous robots, all contain 1kW of compute for full autonomy embedded in the device, that is 2TW of demand, and more than even the most bullish data center forecasts?!
Hence “the lynx effect” is our new name for highly specific AI deployments that lean more on AI chips being deployed at the grid edge, behind the meter, rather than in large, centralized data-centers.
Post-script. This week, the lynx returned and actually met our dog, live and in person, in an encounter that can only be described as ‘Fenton-esque’ !! :-/.