WEEKLY ROUNDUP Hello, Reader. Lemons and blue jeans might seem to have as little in common as a top sirloin and artificial intelligence. But they all share one common trait: a nearly unquenchable thirst. Each of them owes its existence to spectacular volumes of water. Growing just one lemon consumes about 13 gallons, while producing one 16-ounce steak requires about 475 gallons of water. And according to Levi Strauss & Co. (LEVI), producing a single pair of its iconic 501 jeans consumes nearly 1,000 gallons of water. AI's water consumption is not as easy to calculate as a top sirloin's, but it is enormous… and increasing rapidly. Although AI does not directly consume water, the data centers that power AI technologies do. Most of that H2O finds its way into the cooling systems that prevent data center server racks and components from overheating. The centers also use water indirectly from their electricity use. On average, an Alphabet Inc. (GOOGL) data center consumes 450,000 gallons of water a day. But some of the newest centers consume 10 times that volume, as do some of the newest semiconductor fabs. Taiwan Semiconductor Manufacturing Co. Ltd. (TSM) is building a state-of-the-art chip-fab in Arizona that will guzzle 4.75 million gallons of water a day – enough to supply about 1.5 million average homes. And the nebulous force of "the internet" cannot operate without millions of physical servers humming along 24 hours a day inside the world's data centers. Thus, these servers rely on and consume prodigious volumes of electricity and water. And most folks don't realize how water-intensive GenAI programs like ChatGPT can be. Shaolei Ren, an associate professor of electrical and computer engineering at University of California, Riverside, conducted research to calculate that impact. He and his team observed… GPT-3 needs to "drink" (i.e., consume) a 500ml bottle of water for roughly 10-50 responses, depending on when and where it is deployed. These numbers may increase for the newly-launched GPT-4 that reportedly has a substantially larger model size. The tech industry's large and growing water consumption has not yet attracted the same level of attention – or concern – as the industry's growing energy consumption. But as tech companies expand their data center footprints worldwide, water consumption will become an increasingly challenging issue for them, and an increasingly expensive one. But despite the early signs of water challenges, none of the major tech companies are hitting the "pause" button on their data center expansion plans, which means their collective water consumption is on the verge of erupting like "Old Faithful." Google's water consumption, for example, jumped 17% in 2023, compared to the year before. Synergy Research Group predicts worldwide hyperscale data center capacity will almost triple within the next six years. The expanding number of new facilities will account for some of that growth, but the technical evolution from CPUs (central processing units) to GPUs (graphics processing units), like the ones Nvidia Corp. (NVDA) produces, will deliver most of that capacity boost. GPU servers have become the industry standard for data centers. They deliver vastly superior performance for AI-related processing than the legacy CPU servers that most existing data centers use. Inconveniently, GPUs consume about four times more electric power than CPUs, which means they also generate significantly more heat than CPUs. Furthermore, as AI technologies become more widely adopted, they will require ever more data centers with ever more powerful processors. Consequently, data center cooling systems must become more robust. Water-cooled systems are the obvious choice. Air cooling can do the job in some cases, but because water's thermal capacity is 3,000 times higher than air's, it can absorb more heat than air, using less energy and at lower cost. To be clear, the data center industry will not cause any immediate harm to the global water cycle. The industry's collective water consumption has not yet risen to alarming levels. But the more stressed the world's water supplies become, the greater the likelihood that water costs will rise for the data center industry. This relationship is just starting to create compelling opportunities for investors. In fact, I recently recommended an emerging leader in the water-handling business for the oil & gas industry to my Fry's Investment Report subscribers. I believe this company is in an excellent position to capitalize on a powerful opportunity that will continue to gain momentum. To learn more about this recommendation, join me at Fry's Investment Report today. Now, let's look at what we covered here at Smart Money this past week… |
No comments:
Post a Comment