Is Memphis The Next Silicon Delta?
The Hype Machine and AI - Caveat Emptor
Elon Musk has decided that Memphis will be the home of his xAI company’s new supercomputer. According to the South African billionaire, it will be the world's most powerful (read: fastest) supercomputer. There’s no surprise there. After all, men like Musk seldom set out to build the second or third largest anything. Just check out the size of his “rocket.” The idea of being the home of the largest supercomputer in the world, understandably, created an excitement in the Bluff City’s business leadership that was palpable.
Few details about the project deemed the “xAI's Gigafactory of Compute” are available. Nonetheless, Ted Townsend, the president and CEO of the Greater Memphis Chamber of Commerce, was in full cheerleader mode. Townsend claimed that Musk's proposed investment would be the largest in the city’s history and make Memphis home to the “power necessary to place humans on the surface of Mars.”
For those of us given to geekiness, these AI computers are wonders. The Hewlett Packard Enterprise’s Frontier (HPEF) - located in Oak Ridge, TN, by the way)) - is the current holder of the title “world’s fastest supercomputer1” and was faster than the next seven most powerful supercomputers in the world combined when it came online. HPEF was the first supercomputer to break the exaflop barrier. An exaflop is not a shoe you wear when you visit the beach in the summer. An exaflop is the same as a quintillion but cooler sounding, and a quintillion is a ONE, followed by 18 ZEROS. The HPEF in Oak Ridge can process 1.1 exaflops of calculations per second.
Like so much that Elon does, the devil is in the details, and neither he nor the leadership in Memphis appear to have consulted the Evil One regarding the Ts and Cs of this project yet. It is unclear whether the supercomputer Musk wants to place in America’s largest town will be the most powerful once its switch is flipped sometime in 2025. It will need to do more than outperform the supercomputer in Oak Ridge to reach that goal. Aurora, a supercomputer under construction just outside of Chicago, will be able to perform 2 billion billion (technically, two quintillion) operations in a single second when it is fully operational.
If Elon wants to put a supercomputer in Memphis, rest assured that our leaders will find a way to make that happen. Having them find a way to make it happen that reallocates a bit of Elon’s wealth to help the homeless and poverty-stricken would be excellent, but that is likely a bridge too far. Nonetheless, before we all sign up to march in Elon’s parade - and rest assured it is Elon’s, and you will march to his beat - maybe we should look at the narrative surrounding AI companies to date.
Investigative journalist Julia Angwin’s guest essay in the New York Times, titled “Press Pause on the Silicon Valley Hype Machine,” is just the cold shower Mr. Townsend and the rich looking to get richer need before they sign on to work for Mr. Musk. Angwin’s article is an expose of how the Silicon Valley elite have hyped AI (i.e., alluded to it doing things that no one has any idea whether it will ever do) to attract investors to their companies and extort favorable terms out of local governments. She points out that it was just over a year ago that several big names in AI were suggesting that there be “a six-month pause” in the development of larger systems using AI” based on their fear “that the systems would become too powerful.” With a straight face these “experts” suggested that to continue a pace with the frantic pursuit of AI might result in humanity losing “control of our civilization.” Nothing in our exploration of AI so far suggests this is possible. On the other hand, history is awash with oligarchs and plutocracies that have set civilizations back centuries.
As Angwin points out, there was no pause - not six months, six weeks, six days, or even six minutes. Time, after all, is money. And civilization remains, for better or worse, firmly in the hands of humans. The alarm these AI gurus sounded was nothing more than the kind of hype that takes place in front of strip clubs on Bourbon Street in the French Quarter. Enticing, but this side of a few Hurricanes, it’s not based on facts in evidence.
Nonetheless, the response to the alarm these “experts” sounded was as predictable as the story that repeats each night in The Big Easy. For days, the media imagined all the ways AI might destroy humanity. It was a license to report science fiction and conspiracy as news, which has become more the norm than an exception - given that the world is far more boring than a good story built on a kernel of truth.
The story was created to draw attention and build excitement around the projects in which each of these “experts” has a stake. After all, who doesn’t want a piece of world domination? This is how Silicon Valley works. It is a storyteller’s paradise where a good story can turn your company into a unicorn and you into a billionaire. Never mind that, according to Angwin and others, AI can barely lay down a quality piece of prose or a usable line of code. The truth is that AI is simply math (but NOT simple math) expressed in computer code, which is self-learning. It takes an amazing amount of power to run these systems - exaflops of it. But there is nothing spooky here (or remotely conscious or sentient). AI algorithms - the same tool that hooks you on cat videos or clips from war movies - use the modeling given to them by their creators as a base for answering questions or manipulating the behavior of social media users, then learn from how you (and millions of others) respond. It incorporates its successes and failures into its “learning” (math equation) and then integrates this data into its calculations the next time it faces a similar challenge.
And as “magic” as all this seems, it is still, at best, mediocre in terms of the quality of its output. Angwin offers several examples of AI failing to live up to its hype. OpenAI claimed that ChatGPT-4 “aced the uniform bar exam.” For the folks at OpenAI, scoring in the 48th percentile is an acceptable outcome for someone taking the bar. I suppose this is correct if that individual never expects to practice law. Cognition AI released an “AI software engineer” called Devin, who was challenged by a human software developer, Carl Brown, to a “code off.” Carl completed a coding task in 36 minutes, which took Devin over six hours to finish. Professional writers find that generative AI writes about as well as it codes. Angwin’s own experiences suggest that while “AI models can often prepare a decent first draft… I must spend almost as much time correcting and revising its output as it would have taken me to do the work myself.”
Could the models get smarter? Sure. Will businesses be willing to use mediocre AI to replace human beings capable of far better performance? You betcha. Meanwhile, some questions are not being asked about AI that should be. For example, what does it cost in terms of resources and opportunity cost? Supercomputers - where AI lives, moves, and has its modest being - use a lot of energy and must be continuously cooled. This fact is very exciting for MLGW, which has publicly stated that the xAI supercomputer Musk intends to build in Memphis will consume $90 million of electricity annually. Specifically, xAI will need up to 150 megawatts of electricity to run its facility — enough energy to power 100,000 homes.
The utility company doesn’t highlight that this electricity has to “come” from somewhere, and right now, for MLGW, that means nuclear, natural gas, and/or coal. MLGW also failed to discuss how xAI’s needs would impact the “energy burden” of Memphis households. The energy burden is the percentage of a family’s income that goes into paying their utility bill. The national average is three percent, but in Memphis, the average is 27 percent. Then there is the water - 1.3 million gallons of water per day - that will be required to cool all that exaflopping the supercomputer will be doing. According to MLGW, the plan is to build the city’s first greywater system, which takes wastewater and re-treats it for industrial use, to accommodate this need. Unlike electricity, however, water is not “made” by humans. It comes from nature, and nature has been stingier with it of late, creating battles between states over water access in the United States.
These types of tangible costs are easier to calculate than opportunity costs. Opportunity cost is the value of the next best alternative foregone when a decision is made to pursue a particular course of action. For example, if I spend $1,000 on a vacation instead of investing it in stocks, the opportunity cost is the potential return that could have been earned from the investment. If, by investing, I had made $500 (unlikely, but for the sake of an example, hang with me), my vacation would have cost me $1500. Some of our best and brightest brains are working on AI today. There is a cost to having them work on AI that will power chatbots (one of xAI’s purposes) as opposed to focusing on climate change, especially given that the carbon footprint required to make AI possible might exacerbate the latter.
Make no mistake: AI is here to stay, but it is not yet and may never be what the Silicon Valley hype machine says it is now or could be in the future. Jeremy Grantham, an investor famous for predicting the dot-com crash in 2000 and the financial crisis in 2008, believes that AI is a bubble within a bubble that will start leaking air at some point, leading to a lot of pain for many people. Those people will not include Elon Musk and the other members of the billionaire club in Silicon Valley. Should we welcome Elon Musk to Memphis? Sure! Nobody does technology without critical thought better than we Americans. Let’s get our slice and hope we can find a chair to sit in when the music stops.
Thanks for reading Through The Looking Glass! Subscribe for free to receive new posts and support my work.
Based on the Top500 list of May 2022