The first of these is the so-called “geopolitical risk” in Taiwan (that is, the potential for an invasion from China, which U.S. experts have predicted could happen by 2027). We would also note that the recent earthquake, which in addition to the tragic loss of life and infrastructure damage, has caused Taiwan Semiconductor Manufacturing Co (TPE: 2330) to temporarily halt production and evacuate parts of its plant. As we highlighted in our article on NVIDIA back in February, “NVIDIA is not immune to this risk. NVIDIA’s fabless production model outsources the majority of manufacturing to other semiconductor giants such as Taiwan Semiconductor Manufacturing Co (TPE: 2330) and has other Taiwan-based companies in its supply chain, meaning that it, too, would be heavily impacted if conflict arose in the region.”

Another risk that many market pundits and professional investors (particularly so-called value investors) will cite is the company’s valuation. For some perspective, James Stack of InvesTech Research recently noted that Nvidia constitutes 5.1% of the S&P 500 Index, surpassing the total 3.9% of the energy sector. However, the ten largest energy companies generate annual revenue that is eighteen times greater than Nvidia’s revenue from the last 12 months.

There is also an incredible amount of capital required to support these AI endeavours. While in the past we often heard investors promote businesses as being “capital-light”, particularly in sectors such as software, social media, technology and franchising, it does not appear that the huge capital costs of AI are being discussed much. Nor are there any indications of what the return on these investments might be. We have recently seen enormous government grants to the likes of Taiwan Semiconductor and Intel (NASDAQ: INTC) and continued multi-billion dollar investments by Microsoft (NASDAQ: MSFT). These alone should be the focus of prospective investors. Yet OpenAI chief executive Sam Altman was reportedly in discussions with a group of investors for an AI chip project that may require raising up to US$7 trillion–equating to the combined gross domestic product (GDP) of Germany and the United Kingdom, or nearly 30% of the mighty U.S. economy!

The energy demand of AI technologies is a critical aspect that needs careful consideration too. As AI continues to advance, its appetite for electricity is growing at an unprecedented rate. For instance, the operation of AI data centres could account for a significant portion of the U.S. power supply by the end of the decade. In a report in January, the International Energy Agency estimated that the average ChatGPT request requires 2.9 watt-hours of energy, equivalent to running a 60-watt light bulb for nearly three minutes and nearly 10 times that of a conventional “Google” search. In total, Forbes estimates that ChatGPT consumes over half a million kilowatts of electricity each day, equating to nearly 180,000 U.S. households. (On another note, a single ChatGPT conversation also uses about 500 ml of water, equivalent to one plastic bottle worth). Further, according to McKinsey, one solitary Nvidia graphics processing unit (GPU) is estimated to use the same amount of energy as the average U.S. home–an amount that is only expected to increase with the newer version of Nvidia’s semiconductor chips.

To service the expected demand growth in AI usage, the IEA estimates that power demand is expected to grow by at least 10 times in just the next three years. This incredible power usage is beginning to gain some coverage, with Arm Holdings (NASDAQ: ARM) chief executive Rene Haas the latest technology executive to cite AI’s ‘insatiable’ need for energy. Haas estimates that by the end of the decade, AI data centres could consume a whopping 20%-25% of the total U.S. power supply

Demand for electricity in the U.S. has been largely stable for nearly 30 years, but the rise in cloud computing and AI is causing a surge in demand, and the country’s sixty-year-old grid is struggling to keep up. The Washington Post recently ran a headline entitled: “Vast swaths of the United States are at risk of running short of power as electricity-hungry data centres and clean-technology factories proliferate around the country, leaving utilities and regulators grasping for credible plans to expand the nation’s creaking power grid.” It describes the surge in demand for industrial power in Georgia, where the projection is that demand over the next decade will be a staggering 17 times what it is today.

What’s more, this is happening at a time when utilities are pulling back on investment, as described by investing great Warren Buffett in this year’s Berkshire Hathaway letter to Shareholders. Typically, investment amongst utility companies is constrained because of the fixed returns set by regulators and the companies’ high dividend payout ratios. Berkshire Hathaway Energy (BHE) was traditionally an exception, making substantial investments to expand its service area and drive down the cost of energy production. However, an increasingly unfavourable regulatory environment and a greater threat from bushfires as a result of more convective storms is reducing utilities’ willingness and ability to invest. Buffett succinctly states that “when the dust settles, America’s power needs and the consequent capital expenditure will be staggering.”

In the U.S. in particular (but also in many other parts of the world), this extraordinary growth in power demand comes amidst ageing electricity infrastructure and an uncertain (and partly unproven) transition toward renewable energy. Bridging the gap between the current energy supply and the growing demand for AI technologies will present unique opportunities for thoughtful investors who can identify and support innovative solutions to modernise and expand the electricity infrastructure.