Unlock Editor's Digest free of charge
FT editor Roula Khalaf selects her favourite stories on this weekly newsletter.
Nvidia and its foremost supplier Taiwan Semiconductor Manufacturing Company are facing production issues with the highly anticipated next generation of its strongest AI chips, which may lead to delays in shipments planned for this yr.
Nvidia's cutting-edge designs, which use a brand new manufacturing process from TSMC, have caused complications with certain models of the upcoming Blackwell family of information center chips as they prepare for mass production, in response to people acquainted with the situation.
Amid a broader sell-off in stock markets on Monday, Nvidia shares fell as much as 15 percent in early trading in New York, while TSMC closed 10 percent lower on the Taiwanese stock exchange after The information first reported on the possible delays. By 12 noon in New York, Nvidia's decline had eased to six percent.
Customers like Microsoft, Google, Meta and Amazon, in addition to AI startups like OpenAI, are lining up to purchase Nvidia's latest chips to construct the subsequent generation of their AI systems.
Major technology firms invest tens of billions of dollars in AI infrastructure every quarter and have announced plans to extend their investments further in the approaching months. Some analysts at the moment are predicting that a trillion dollars will likely be invested in data centers over the subsequent five years to advance AI.
However, concerns concerning the sustainability of the AI ​​boom have grown on Wall Street in recent weeks. Nvidia's market value has fallen by around $750 billion because it briefly became the world's most beneficial company in mid-June.
Hedge fund Elliott Management told investors in a recent letter seen by the Financial Times that it believes Nvidia and other major technology stocks are in a “bubble situation” and that AI is “overvalued and lots of applications aren’t yet mature.”
When Nvidia CEO Jensen Huang introduced the brand new chips in March, he said Blackwell could be twice as powerful at training AI models as its predecessor, Hopper.
In May, he told investors and analysts on a quarterly earnings call that Nvidia would generate “a variety of revenue from Blackwell this yr,” and said last week that the corporate had begun shipping engineering samples.
However, an individual acquainted with the manufacturing process said there have been “difficulties” in getting the Blackwell chip toward mass production. “These are related to the interposer,” the person said. The interposer is a layer that connects the varied chips which might be packaged together within the complex chips needed for AI applications.
Nvidia declined to comment but reiterated that “Blackwell sampling has begun and production is heading in the right direction to ramp as much as volume production within the second half of 2024.” Demand for the prevailing Hopper chips stays “very strong,” Nvidia added.
TSMC didn’t reply to a request for comment.
The problems underscore the immense engineering challenges of packing the facility required by the newest AI chips into limited space, and are prone to further exacerbate the capability bottleneck in advanced packaging, the ultimate stage in producing essentially the most sophisticated chips.
TSMC, the world's largest chipmaker and Nvidia's only manufacturing partner, has struggled over the past yr to extend the capability of its most advanced production technology quickly enough to fulfill demand for AI chips.
When the corporate announced its second-quarter results last month, Chief Executive Officer CC Wei said the corporate wouldn’t give you the option to balance supply and demand by the top of this yr as originally planned. TSMC now hopes to attain this “sometime in 2025 or 2026,” Wei said.
Mark Li, a semiconductor analyst at Bernstein, said Nvidia would likely have to make a small design change to repair the issue.
Analysts at BNP Paribas said that such issues would normally take two to a few months to resolve, but they don’t expect the delay to vary “the medium to long-term picture for Nvidia or the adoption of AI,” even though it might be positive news for Nvidia's biggest rival, AMD.
Citi analysts said in a note to clients that the delay could reduce Nvidia's data center revenue by as much as 15 percent within the quarter through January, even though it could boost revenue in the next period.
