Today, everyone is talking about AI and how it is changing the world.
And now, they’re mentioning robotics/deep tech/etc as the next wave of innovation. But an area of rapid innovation growth that I believe is overlooked today is the energy space.
One of my biggest bets in the near future is how fast and exciting the energy space will become, specifically as it relates to AI, data centers, and general purpose use.
In this post, I’m collecting some of my thoughts together and diving into the fundamentals of data centers and energy infrastructure, which I think is pretty cool.
Diving into it, it seems the rapid data center growth and electricity demand in recent years finally caused people to re-evaluate the USA’s aging grid infrastructure. Trivially, an aging grid infrastructure combined with an increase in energy demand doesn't align well.
Throw data centers into the mix and we have an even more interesting problem.
So why exactly are data centers bad for the grid?
Yes, data centers use a lot of electricity, but it’s because they have highly variable uses of electricity that make for a challenging grid stabilization problem.
Specifically, data center electricity usage varies throughout the day, because people use the internet/AI/etc throughout different times of the day. This may cause sharp spikes in both increased and decreased data center electricity demand.
The US grid maintains a voltage of 110-120V (60 Hz). Increases and decreases in electricity demand cause this number to fluctuate on power lines. Rapid fluctuation causes electricity supply to not meet demand (or vice versa), causing grid instability and outages.
Thus, when electricity demands increase, electricity suppliers need to generate more electricity through a form of generation (coal, gas, nuclear, renewables, etc.). When electricity demands decrease rapidly, there needs to be an instant response to maintain 60 Hz on the grid, typically by storing excess electricity in batteries, etc. Or else there is instability. This gentle balance of stabilization makes it a difficult challenge.
Data center composition
Another challenge of data centers is that they need constant power reliability, and since the grid isn’t necessarily that reliable, data centers also rely on backup generators.
This is part of why hyperscalers are building their own power generation sources, such that they are not dependent on local generation resources.
Additionally, local resources are actually a key part of data centers, primarily because they need to be built in strategic locations, specifically near reliable electricity generation sources.
The data center stack includes:
IT infra (chips, racks, etc)
Cooling infra (air cooling, water cooling, direct to chip, immersion, etc)
Power infra (power lines, cables, backup generators, etc)
One of the challenges data centers run into is deciding what stack of the above 3 they want. Do you buy the best-in-class options across all 3? Yes, all 3 are of highest quality and you're not vendor-locked, but this lack of cohesion may not be the most effective way of reducing energy consumption. Or do you buy all 3 from the same vendor, hoping that cohesion works well for you, but risk vendor lock-in?
Cohesion of the data center stack plays a role in efficient energy consumption and helps drive data center costs down.
So why does any of this matter?
Yes, software is democratized.
Yes, robotics/deep tech will be one of the next technological revolutions.
But what enables that revolution?
Data centers and energy.
There are companies building across every layer of this stack from chips to cooling to power generation to grid stabilization, and I think most people in tech are underestimating how interesting/important this space is about to become.
