Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
Delays to Nvidia’s next-generation artificial intelligence processor will not derail the chipmaker’s plans to produce a new version of its flagship product every year, chief executive Jensen Huang told the Financial Times.
A production issue affecting Blackwell, Nvidia’s hotly anticipated new chip platform, “doesn’t matter” to its plan to accelerate its release schedule from every two years to a “one-year rhythm”, Huang said, after announcing the ambitious new schedule last October.
Nvidia’s accelerated pace of innovation is seen as key to maintaining its dominance of AI chips, in a market estimated to be worth hundreds of billions of dollars over the coming years.
Earlier this month, last-minute complications forced Nvidia and its partner Taiwan Semiconductor Manufacturing Company to make changes to how Blackwell was produced. That forced Nvidia to push back the full release of the chip towards the end of this year.
The delays contributed to a 2.8 per cent drop in Nvidia’s share price in pre-market trading on Thursday, as investors worried about slower growth and narrower profit margins after the chipmaker rode the AI boom to a more than $3tn valuation.
The US tech giant’s ability to hit ambitious production and earnings targets has taken on an outsized importance in the US stock market, with a blistering rally in the company’s shares driving a quarter of the year-to-date gains on the S&P 500.
Despite the increasing complexity of manufacturing its sophisticated chips, Huang said that Blackwell’s successors were still on track for annual upgrades. The company’s engineers were working on its next two generations, he said. “They will finish on their one-year rhythm.”
Nvidia is stepping up the pace at which it improves its products as it looks to stay ahead of competitors such as AMD and Intel in the market for graphics processing units.
The likes of Meta and OpenAI rely on Nvidia’s GPUs to train their large language models, the AI systems behind chatbots such as ChatGPT, and demand has far outstripped supply of its most powerful chips for almost two years.
Huang was peppered with questions by analysts on Wednesday about how Nvidia’s customers were making money from their huge investments in AI infrastructure, as investors debate how long Big Tech can continue to spend tens of billions of dollars each quarter on data centre equipment.
Huang told the FT that Nvidia itself was seeing significant productivity improvements from using AI to accelerate data processing, and to help design and build its products.
“We can’t design [our chips] without generative AI,” he said. “We couldn’t design Blackwell, we can’t write our software without AI any more.”
He said that Nvidia’s 20,000 engineers were “really kind of like 60,000” thanks to the boost from AI.
“I guess Nvidia is the world’s smallest large company,” he said. “We’re able to do things that can’t reasonably be done using handwritten software . . . Nvidia is not a large labour company.”
Huang told analysts on Wednesday that Blackwell’s production problems were primarily related to the “mask” — the template that imprints a chip’s design on to a silicon wafer, through a process called photolithography. The adjustments were needed to improve Blackwell’s manufacturing yield, which is vital to Nvidia’s profit margins.
“The change to the mask is complete,” Huang said. “There were no functional changes necessary.” Blackwell would ramp up to mass production and start shipping to customers during Nvidia’s fiscal fourth quarter, which ends in January 2025, he added.
Asked about the impact on Nvidia’s profit margins, which the company told investors would fall slightly over the coming months, finance chief Colette Kress said the company was “going to see some slight difference possibly in [the fourth quarter], again with our transitions and the different cost structures that we have on our new product introductions”.
Some Wall Street analysts have already begun to trim their estimates for Nvidia’s gross margins as a result, contributing to Thursday’s stock-price decline.
Matthew Ramsay, analyst at TD Cowen, said in a note to clients on Thursday that the challenges Nvidia and its rivals faced in attempting to emulate the iPhone’s annual release cycle were formidable.
“The entire server industry is ramping one of the most complex products in computing history,” Ramsay said. “This is bound to create stress in design and supply chains, and we continue to take the view that a slight shift in production timing is neither thesis-altering nor reflective of a change in broader demand for AI compute solutions overall.”
Additional reporting by Michael Acton
Read the full article here