There is a big question hanging over the technology sector: How long will your massive investments in AI infrastructure actually last?
Tech giants are spending hundreds of billions of dollars on artificial intelligence infrastructure — primarily, data centers and the chips that power them. It’s an investment that they say will set the stage for AI to reshape our economy, our jobs and even our personal relationships.
This year alone, technology companies should invest $400 billion in AI-related capital expenditures.
Some of this will almost certainly put recurring pressure on corporate balance sheets.
And for companies that depend on AI for their future, the question of how often they will have to upgrade or replace advanced chips is critical — especially as there is growing skepticism about whether AI will produce returns large or fast enough to recoup existing investments and cover future infrastructure costs.
This is fueling concerns around an AI bubble — concerns that the hype around AI and the spending on it are out of sync with its true value.
These concerns arise at a time when the actions of “Magnificent Seven” technology companies represent around 35% do valor do S&P 500raising questions about what an AI crash would mean for the economy.
“The extent to which all this construction is a bubble depends partly on the lifespan of these investments,” said Tim DeStefano, a research associate professor at Georgetown’s McDonough School of Business.
Life cycle two chips
It’s unclear how long high-end graphics processing units (GPUs), the chips most commonly used for AI training and processing, will remain useful.
Several technology experts told CNN who estimate that AI chips could be used to train large language models across 18 months and three years. But the chips could continue to be used for less intensive tasks for several years, they added.
By contrast, central processing units (CPUs) used in traditional non-AI data centers are typically replaced every five to seven years, experts said.
This is partly because training AI models exposes chips to significant stress and heat, wearing them out more quickly. About 9% of GPUs will fail over the course of a year, compared to about 5% of CPUssaid David Bader, a data science professor at the New Jersey Institute of Technology.
Subsequent generations of AI chips are also rapidly improving and becoming more efficient, meaning it may not be economical to continue running AI workloads on older chips even if they are working.
Different experts offer slightly different estimates. DeStefano said AI chips will likely break after about five to 10 years of use, but their economic lifespan is only about three to five years.
Meanwhile, Bader estimates that GPUs could be used to train AI models for 18 to 24 months. But he said older chips could still handle tasks like processing users’ AI queries, known as inference, for another five years, prolonging their value.
A Nvidiathe largest supplier of AI chips, claims its CUDA software system allows customers to upgrade the software of existing chipspotentially postponing the need to upgrade to the latest product.
Nvidia Chief Financial Officer Colette Kress said on the company’s latest earnings call last month that GPUs “shipped six years ago are still running at full capacity today” thanks to the CUDA system.
But whether the chips last two years or six, tech companies still face the same question: “Where will the revenue come from that will allow us to rebuild on this scale?”said Mihir Kshirsagar, director of the technology policy clinic at Princeton’s Center for Information Technology Policy.
What does this have to do with the AI bubble?
The faster chips degrade, the more pressure companies will feel to earn a return on AI to fund their replacement.
And the long-term demand for AI remains uncertain, especially in light of reports this year that indicate most companies implementing the technology have yet to see benefits to their bottom lines.
Enterprise customers will be the real revenue generators for AI companies, but these companies are still figuring out how to use the technology to generate revenue or reduce costs, DeStefano said.
“There is demand for generative AI from individual users… but that is not enough for these big AI companies to recoup their investment costs,” he said.
Michael Burry, the famous investor behind “The Big Short,” recently warned of an AI bubble. His argument is based, in part, on the prediction that technology companies are overestimating the useful life of their chip investments, which could end up weighing on their profits.
AI leaders are also starting to speak more openly about the issue.
Microsoft CEO Satya Nadella said in a podcast interview last month that the company has started spacing out its infrastructure investments so that chips in its data centers don’t become obsolete at the same time.
And OpenAI chief financial officer Sarah Friar raised the alarm last month when she said the company’s role as a pioneering maker of AI models depends on the durability of the most advanced chips: “three years, four years, five years or even longer.”
If that life cycle is shorter, she suggested the company may need the U.S. government to “guarantee” the debt it is taking on to finance its aggressive infrastructure commitments. (OpenAI quickly tried to walk back the comment, saying it was not seeking a government guarantee.)
In previous market bubbles, infrastructure built during the hype cycle that went dormant after the burst was still usable years later. Fiber-optic cables installed during the internet bubble of the late 1990s, for example, now provide the foundation for today’s internet.
But the — if it’s real — said Paul Kedrosky, managing partner at investment firm SK Ventures. He argued that AI data centers will not maintain the same usage potential over time without continued investment in new chips. And the ramifications could extend far beyond the balance sheets and share prices of tech giants.
“We’re not just building these data centers, (tech companies) are pushing to build electricity plants to support all of this,” Kshirsagar said. “If the economy doesn’t work, there will be some very important social issues.”
