This was one MeetUp you didn’t want to miss.  Most of the 30+ attendees said it was our best one ever—and we have had some great ones.  It was so good because we talked a lot about the context and the big issues of the next 20 years.  

Our speakers were Daveed Sidhu, a certified AI professional with years of experience, and James Frasher, Innovation Manager at SMUD, who has been examining all the ways AI could help and giving some of those a try.

Daveed began by giving us a detailed overview available in his slides.  They are packed with information and are a great background source.  His main point was that AI or machine learning are not all that new.  They have been evolving for decades.  What is new is the broader recognition of the potential and perils that AI can create across all we do.  

The upcoming challenge for the grid is electrification of the transportation sector and the buildings sector (to achieve decarbonization), not to mention all the power requirements of data centers to run AI software.  Daveed put the challenge this way:  US total electricity consumption is currently 4 Terawatt-hours per year and that could grow to 12 Terawatt-hours by 2045.  How are we going to cope with that?  But the crucial reality is that the power system – generators and power — are only running at about a 40% capacity factor (4 Twh of consumption served by 1300 GW of generation that has a theoretical ability to run 8,760 hours per year to make 11.5 Twh of electricity). 

So there is a lot of inefficiency in the current system.  That inefficiency has been the result of power demand varying dramatically over the day and over the seasons and the focus of utilities on serving that demand with enough supply, rather than considering demand to be something that could be managed to better match the generation available.  That may be a simple way of looking at, but it set the context.  Each part of the grid could be managed to make better use of it.  That is not just a matter of software.  To do its job, the software needs data and control points.  So there must be a lot more devices installed to monitor uses and supplies, plus new devices to get control over power flows so lines do not become overloaded.  Sounds like a big job.

James Frasher explained what SMUD is doing to take steps to understand what the best things are SMUD could do to meet these challenges. They are targeting 300-1200 MW of “virtual power plants” or combinations of controllable loads and customer-side generators in their Zero Carbon Plan for 2030.  They are looking at fleets of electric school buses as battery storage they could use (“Vehicle-to-Grid” or V2G), since the buses are used only a few hours a week, for example.  They are increasing the incentive for customers to install battery storage to as much as $10,000.   But he is also concerned about cybersecurity–that massive amounts of monitoring and control points could create huge vulnerabilities and opportunities for mischief.  Think about the ransomware attack on car dealerships today, but amplified to the whole power infrastructure.  

What about the power demand of all the data centers running AI models?  We have seen many prior forecasts of tremendous growth needed to run cryptocurrency platforms that did not turn out to be so serious.  Will the same apply to AI?  Will more efficient chips be used.  Will computing get away from microchips to something else.  All fascinating to ponder.

Everybody likes to peer over the horizon and see what the future may hold.  It captures the imagination.  This meetup was a prime example of that and the probable reason the attendees liked it so much.  It also means we should do more of this.  Please keep an eye on our schedule of MeetUps coming your way.