During the Berkshire Hathaway annual meeting last weekend, Warren Buffett made a lot of headlines on AI. By comparing AI to nuclear weapons and sounding the alarm on deepfakes by talking about how he was almost fooled by a deepfake of his own likeness, his commentary would put him squarely into the “AI doomer” camp.
As I shared in my recently published Q1 performance letter, I listen to Berkshire annual meeting Q&A’s mostly as a meditative exercise, not for specific market or industry analysis. Yet, if you listen closely to commentaries shared by people other than Buffett, it’s clear that the Berkshire empire will be a key player powering AI’s growth in the next decade. Here is what Buffett’s successor and chair of Berkshire Hathaway Energy (BHE), Greg Abel, said when answering a question about utilities regulations in Utah:
“...if we look at the demand that’s in place for MidAmerican Iowa Utility over the next, say, into the mid 2030s associated with AI and the data centers, that demand doubles in that short period of time, and it took 100 years plus to get where we are today, and now it’s going to double.
And that will require substantial amounts of capital from MidAmerican and its shareholders.”
Since MidAmerican Energy is a subsidiary of BHE and BHE is of course a major part of Berkshire, those “shareholders” whom Abel mentioned are Berkshire shareholders. While the investment world hem and haw about what Buffett will do with his $189 billion cash pile, it’s rather obvious that a good chunk of that pile will be poured into expanding MidAmerican Energy’s capacity – do in the next 10 years, what it did in the last 100 years.
The unlikely intersection between MidAmerican, Iowa, and AI supercomputers deserves more unpacking. GPT-4 was trained in the Azure supercomputer data center west of Des Moines, the capital of Iowa. If we take the personification of AI to the comical extreme, you can say ChatGPT is an “Iowan.”
So what’s so special about Iowa in the context of AI? And what role will MidAmerican play in fueling AI advancement in the next 10 years?
Networks, Land, Power, Taxes
Iowa has been ground zero for building advanced data centers for at least a decade and a half. Its appeal to tech companies is founded on a four-legged stool of networks, land, power, and taxes.
Land is simple enough. Data centers need a lot of flat land. Iowa, amidst miles of cornfield, has plenty of flat land. But so do many other states in the Midwest, including Buffett’s home state of Nebraska.
Network is also straightforward in that there is a consistent overlap between states’ with lots of railroads and strong networking cable infrastructure. In the early 1900’s, Iowa had more than 10,000 miles of railroad tracks. Much of that is gone, due to the consolidation and collapse of many railroad companies over the subsequent decades. However, the remnants of that construction remain and became the foundation for laying miles of fiber optic cables to connect data centers. Though again, many other midwestern states also have dense railroad networks, so Iowa is not particularly special in this regard.
Power and taxes are what distinguishes Iowa as one of the most ideal places in America to build AI supercomputers.
Iowa state and municipal governments have had a long history of using various tax incentives to lure tech companies. In 2007, when Google was contemplating whether it should build its next big data center in Council Bluffs, Iowa, the state legislature approved millions of dollars worth of sales and use tax exemptions on things like servers, power generators, and cooling systems – all the things that go inside the shell of a data center. The local Council Bluffs government also approved a property tax exemption worth $33 million over a 20-year period to sweeten the deal. To incentivize Microsoft to expand its data center footprint in West Des Moines, where it eventually trained GPT-4, a combination of tax incentives worth $91.3 million was offered by the state and local governments. In a town of 68,000 people, Microsoft now has five data centers in West Des Moines. Chances are, GPT-5 (or whatever Sam Altman feels like calling it) will be trained there as well.
But everybody and her grandmother knows AI is a power hungry beast. So where do the Iowa-born AI’s get their power? MidAmerican Energy.
Long before ChatGPT reached 100% name recognition, Microsoft, Google, as well as Meta, another major data center builder, have been working with MidAmerican to gain access to power. This is partly because MidAmerican is the largest player in town, serving two-thirds of the state’s energy needs, including prime data center locations like Des Moines and Council Bluffs. More important, MidAmerican has been a leader in producing wind power, which is especially important for data center operators.
Most major data centers have sustainability goals, because of their large energy footprint. Being able to run data centers that are 100% wind-powered is both important for the surrounding environment and critical to their cost-effectiveness. By working with MidAmerican, data centers built by big tech companies are able to achieve their sustainability objectives without compromising their ambitions. It is a competitive advantage and source of pride for Abel, a point he was quick to share when asked during last weekend’s shareholder meeting about BHE’s lack of renewable energy production in Nevada:
“We have great examples in Iowa where at times, 100% of our energy comes from wind. And we’re thrilled with that.”
Genie Is Out of the Bottle
Through a fascinating combination of land, railroads, taxes, wind, as well as rivers (specifically the Raccoon and Des Moines rivers) to provide cooling for data centers, the futures of MidAmerican Energy, Iowa, and generative AI are now closely intertwined.
Despite his skepticism, even Buffett recognizes that the “AI genie” is now out of the bottle and none of us can put it back. And powering the next stage of AI’s growth, at least in America, will most likely be left to Greg Abel’s able hands and capital allocation skills.
I’m not saying you should buy Berkshire stock to gain exposure to AI; there are probably better ways to do that. But if you are an AI optimist, like I am, then ironically we owe a dose of gratitude to Buffett’s masterpiece for having amassed so much cash ready for Abel to deploy, so MidAmerican Energy as a shot at doing in the next 10 years, what it did in the last 100 years.
All the new generative AI applications, many of which will be “Iowan” too, will need all the Berkshire energy they can get.