The Iran war exposed contradictions. Gulf funds are helping underwrite US efforts to win the artificial intelligence race, but now the US is starting a dispute that could destabilize those investments. Some estimates predict that Middle Eastern countries will make long-term commitments of $2 trillion to the AI boom, but that funding now appears precarious. At the same time, rising energy costs threaten to make data centers much more expensive to operate. But the aftershocks of the conflict are unlikely to completely extinguish the AI boom, and the market is unlikely to bifurcate, with so-called hyperscalers such as Alphabet Inc., Amazon.com Inc. and Microsoft Corp. most exposed to changes in their financial positions, while emerging AI labs such as OpenAI and Anthropic PBC are more isolated.
Investors have long treated AI’s mega-success as a single, monolithic story, when in reality there are two distinct components: an incredibly expensive infrastructure business and a cheaper software business. Among the architects of the latter component, Anthropic has been doing pretty well lately, with annual revenue more than doubling in the past three months to $19 billion, compared to OpenAI’s revenue of about $25 billion. Consumers, financial and life sciences business customers, and governments all pay for subscriptions and access. Unlike previous hype cycles regarding the Metaverse and cryptocurrencies, the momentum appears to be sustainable.
For all the concerns about OpenAI’s high cash burn rate, AI Labs also benefits from fixed enterprise contracts. Customers are unlikely to cancel these due to geopolitical uncertainty. Rather, they are likely to maintain them in the hope of making the organization efficient enough to weather any economic instability that comes its way.
Although AI software makers need data centers to run their businesses, they are not directly exposed to rising energy costs like server farm owners. To make money, OpenAI and Anthropic need to run existing AI models to answer queries from paying customers. This is a process known as inference. But training a new frontier model is much more energy-intensive, requiring thousands of AI chips (graphics processing units, or GPUs, made by Nvidia Corp.) to be used continuously over weeks or months.
Of course, the daily cost of inference increases over time, especially for a company like OpenAI, which has 900 million weekly users. However, the energy load is much lower, more distributed, and easier to manage than training next-generation models. The Institute can afford to delay this while it focuses on encouraging companies to incorporate existing technology into their workflows.
Hyperscalers such as Amazon, Google, Microsoft, Metaplatforms and Oracle are at greater risk given how dependent their $1.15 trillion businesses are on cheap, reliable energy, especially natural gas. It is the primary single energy source for U.S. data centers, providing about 40% of their electricity, according to the International Energy Agency, but is a problem amid soaring prices due to the Iran war.
The chip supply chain is similarly at risk. Taiwan Semiconductor Manufacturing Co. (TSMC), which makes nearly all of the high-end chips designed by Nvidia, also relies on the Middle East for about a third of its fuel, with the island nation getting most of its helium from Qatar. This gas is critical in semiconductor manufacturing because of its unique ability to cool and protect silicon wafers during manufacturing. Qatar Energy’s helium production at Ras Laffan Industrial City was halted last week following an Iranian drone attack. The broader implication is that it could take several months for chip production to fully recover.
So Nvidia is probably the most at risk. The world’s most valuable publicly traded company, with a market capitalization of more than $4 trillion, derives most of its revenue from selling chips to hyperscalers. Anything that slows down the construction of a vast new server farm will hurt the company’s orders.
While Alphabet and Amazon at least have recurring cloud subscriptions that serve as a financial cushion, Nvidia has no such revenue stream. The company only sells chips, but faces a double whammy: question marks over its recent mega-deal with the Middle East, as well as difficulties manufacturing in Taiwan. In November, the U.S. government approved Nvidia’s sale of 70,000 cutting-edge chips to the United Arab Emirates and Saudi Arabia, but the deal now appears more uncertain. (NVIDIA declined to comment.)
Energy and cash from the Gulf region are helping fuel the AI boom. No matter how strong application revenue growth is, the longer the war drags on, the more fragile the outlook for the underlying infrastructure looks.
