Elon Musk has reiterated this statement on numerous occasions: "Before we run out of electricity, we might run out of transformers."
Recently, at the Bosch ConnectedWorld conference, he brought it up again, ranking the current scarcities as follows: chips, transformers, and electricity.
AI causing shortages in chips and power resources has been talked about extensively, just like everyone else.
Like many of my friends, the focus this time has shifted to transformers. How did these devices suddenly make it onto Musk's "scarcity list"?
Strictly speaking, not all transformers made it onto the list, but rather, it's the step-down transformers.
Back in our science class days, we learned that to minimize losses, electricity is transmitted at high voltages of several hundred kilovolts through the grid. To bring it down to the standard household voltage of a few tens of volts, it needs to go through several steps of reduction, and this is where step-down transformers come into play.
Following Musk's logic, the shortage of transformers is due to the increasing demand for electricity in the future, mainly driven by AI and new energy sources.
When it comes to AI, Musk has turned to puns, joking that Transformers (referring to the underlying architecture of most large models) are causing a shortage of transformers (the electrical devices).
After all, following his logic, AI's heavy electricity usage in data centers could lead to exponential growth in power consumption as computational demands increase.
That's just one aspect; what many may not realize is that CPUs and GPUs also require a significant number of voltage regulators.
This is because low voltage and high current have long been the mainstream power consumption solution for data centers.
So, here's the deal: For quite a long time, processors have typically required an input voltage of around 1 volt. However, over the past couple of years, as chip fabrication processes have advanced, the voltage requirements for many processors have dropped below 1 volt.
This fussiness over voltage in processors is all about energy consumption in the entire data center.
For an individual processor, its power consumption (P) can be represented by the following formula:
In this context, 'C' stands for a constant related to the manufacturing process, 'f' represents frequency, which ideally should be high, while power consumption 'P' should be kept low. Thus, the only option is to adjust the voltage 'V'.
However, reducing voltage isn't a simple task, especially considering the unique requirements of data centers. They need to maintain continuous power to avoid any potential data loss.
Typically, data centers initially step down the voltage to 12 volts using a voltage regulator, allowing them to switch to backup batteries during power outages. Later, they use specific converters to further reduce the voltage to 1 volt.
There are two options for adding a 12-volt intermediate voltage:
One approach is to equip each server with a 12-volt voltage regulator.
Another approach is centralized voltage reduction, where a more robust transformer is employed to initially lower the voltage to 12 volts before distributing it to multiple servers.
Wow, talk about perfect timing! The past couple of years have seen a huge explosion in AI, coinciding with a period of rapid expansion.
Take Alibaba Cloud's data center in Hebei, for example. In just the past two years, they've already scaled up to 300,000 servers, with plans to expand that to a whopping 1 million servers in the near future.
When faced with such a situation, there are only two ways to go about it. Either you stock up on more transformers using the first method, or you opt for the second method, which involves switching to transformers with a relatively higher voltage (many data centers are considering replacing 12 volts with 48 volts).
Whichever method you choose, it's evident that it will significantly strain the transformers.
Furthermore, Musk also mentioned the power consumption of electric vehicles. The China Association of Automobile Manufacturers conducted a study indicating that by around 2030, electric vehicles in China alone would consume approximately 140 billion kWh of electricity annually. This surge in electricity usage will inevitably lead to a considerable increase in the demand for step-down transformers. However, we won't dwell on this aspect any further.
Transformers indeed don't have the same high technical barriers as chips do, and currently, the most challenging ones to manufacture are the extra-high-voltage transformers, ranging from several hundred to even thousands of kilovolts.
Well, let's break it down. Transformers, theoretically speaking, aren't that difficult to manufacture. But despite that, they still make it onto Musk's scarcity list because, well, they're genuinely scarce in the U.S.
The entire American market heavily relies on imports for transformers. In fact, over the past couple of years, the U.S. has skyrocketed to become the largest importer of transformers globally.
Before AI and new energy sources became all the rage, their transformer imports were sailing smoothly. So, there was never really any thought put into establishing a local supply chain for transformers. Even crucial materials like electrical steel for transformers aren't produced much.
There's an even more awkward situation brewing: about seventy percent of the transformers in the American power grid are due for retirement. Despite this, to sanction China, the US has imposed an import ban on transformers. Consequently, it's still impossible to purchase new ones, leaving the existing ones to soldier on in extended service.
In the past, importing transformers to the US used to take just a few weeks at most. However, last year, in some areas, the wait time stretched to over a year, with prices skyrocketing by more than fourfold.
Due to the supply chain struggles and a shortage of labor force, setting up local transformer production lines in the short term is nearly impossible.
Even for a leading US company, expanding its transformer production line comes with an estimated timeframe of up to 4 years.
Even the Economic Council is fed up with the situation, issuing a report urging the US government to invoke the Defense Production Act and exert official pressure to push manufacturers to establish transformer production chains.
In contrast, the "scarce challenge" of transformers isn't as severe domestically.
Because whether it's research or production lines, things have been going on uninterrupted in China. We've always been the world's top exporter of transformers. Maybe the shortage in the US this time could even become an opportunity for us to further expand.
So, let's not worry too much about Musk's scarcity rankings.
Lastly, after water, electricity, transformers, and chips, what do you think will be the next to make the list?
