Recent weeks have been busy for the artificial intelligence industry, with a string of chipmakers and AI-linked groups unveiling deals often worth eyewatering sums.
Since the start of September, ChatGPT-maker OpenAI alone has revealed a $300 billion with cloud software titan Oracle, a $100 billion investment from AI-darling Nvidia, and strategic partnerships with Nvidia rivals AMD and Broadcom.
Other big-name U.S. tech firms who are shelling out massive amounts of cash on AI, known as hyperscalers, have also highlighted fresh investments in the second half.
While the announcements have underpinned hopes that a now multi-year boom in AI has more room to run, fears have begun to spring up around the circular nature of many of these transactions -- many of which would see semiconductor firms effectively backstop their own customers.
Some observers have likened the mounting spending promises to the dotcom craze of the late 1990s, a bubble that ultimately burst more than two decades ago.
"Announcements of further increases in an already sizable amount of AI capex spending have amplified concerns around the sustainability of AI investment," analysts at Goldman Sachs include Joseph Briggs wrote in a note to clients.
But they argued that the anticipated levels of expenditures are sustainable and supported by the "technological backdrop," even though the eventual winners of the AI arms race are less clear.
"First, AI applications are boosting productivity when deployed. Second, unlocking these productivity benefits requires significant computational power, especially since models are increasing in size much more quickly than computation and energy costs are falling," the analysts wrote.
They added that they are not worried about the total amount of AI investment, noting that, as a share of U.S. gross domestic product, AI spending stands at less than 1%. This is smaller than levels of between 2% to 5% in prior large technology cycles, they said.
A "solid" broader economy should bolster AI capex, albeit as long as companies anticipate the need to be the first-mover in breakthroughs of the technology and investments in compute capacity lead to the improvement of model performance and, potentially, the emergence of artificial generative intelligence, the analysts said.
source :
