Nvidia Corp. failed to live up to investor hopes with its latest results on Wednesday, delivering an underwhelming forecast and news of production snags with its much-awaited Blackwell chips.
The company’s quarterly report — the most anticipated part of the tech industry’s earnings season — met or beat analysts’ estimates on nearly every measure. But Nvidia investors have grown accustomed to blowout quarters, and the latest numbers didn’t qualify.
Moreover, Nvidia’s next big cash cow — the new Blackwell processor lineup — has proven more challenging to manufacture than anticipated. The product is the next generation of the company’s dominant artificial intelligence processor, and fears of delays contributed to a stock decline of 2.4% in premarket trading Thursday, trimming previous losses. The shares had more than doubled this year through Wednesday’s close, following a gain of 239% in 2023.
“It was up against lofty and unsustainable expectations,” Bloomberg Intelligence analysts Kunjan Sobhani and Oscar Hernandez Tejada said in a note.
Third-quarter revenue will be about $32.5 billion, the company said. Though analysts had predicted $31.9 billion on average, estimates ranged as high as $37.9 billion.
The disappointing outlook threatens to tamp down an AI frenzy that has transformed Nvidia into the world’s second-most-valuable company. The chipmaker is the key beneficiary of a race to upgrade data centers to handle AI software, and its sales forecasts have become a barometer for that spending boom.
Heading into the announcement, there was concern that Nvidia was having problems with its new Blackwell design. The company acknowledged that there were issues with production, saying that it was making changes to improve its manufacturing yield — the number of functioning chips that come out of factories. At the same time, the company said it expects to bring in “several billion dollars” of revenue in the fourth quarter from the product.
Supplies will be plentiful after manufacturing gains momentum, Chief Executive Officer Jensen Huang said later during a Bloomberg Television interview. “We’re going to have lots and lots of supply, and we will be able to ramp,” he said.
Nvidia is coming off a string of quarters that shattered Wall Street expectations — even as analysts continued to raise estimates. But the amount of upside has been trending down.
Most of Nvidia’s growth also has come from a small group of customers. About 40% of Nvidia’s revenue stems from large data-center operators — companies like Alphabet Inc.’s Google and Meta Platforms Inc. — which are pouring tens of billions of dollars into AI infrastructure.
Though Meta and others have increased their capital expenditure budgets this earnings season, there’s been concern that the amount of infrastructure being put in place exceeds current requirements. That could lead to a bubble. But Nvidia’s Huang has maintained that this is only the beginning of a new era for technology and the economy.
Expectations were lofty. Nvidia has been the best performing stock in the S&P 500 Index this year, eclipsing gains by all other semiconductor companies. At a market value of more than $3 trillion, Nvidia is worth roughly the same amount as the next 10 largest chip firms combined.
Nvidia made its name by selling video-game cards, but is now best known for so-called AI accelerators. These chips, derived from its graphics processors, are used to develop artificial intelligence software by bombarding it with information.
The process, known as training, makes AI models better at recognizing and responding to real-world inputs. Nvidia’s components are also used in systems that then run the software, a stage known as inference, and help power services such as OpenAI’s ChatGPT.
Last quarter’s results topped Wall Street projections, and the Santa Clara, California-based company’s board approved an additional $50 billion in stock buybacks.
Nvidia’s revenue more than doubled to $30 billion in the fiscal second quarter, which ended July 28. Excluding certain items, profit was 68 cents a share. Analysts had predicted sales of about $28.9 billion and earnings of 64 cents a share.
Nvidia got a jump on other chipmakers because its technology was well-suited to the needs of AI. But rivals are trying to catch up. Advanced Micro Devices Inc. is now its closest competitor, with Intel Corp. — once the world’s biggest chipmaker — trailing further behind. Their combined revenue from the market is only about 5% of Nvidia’s total.
Nvidia’s data-center division — now by far its largest source of sales — generated $26.3 billion of revenue last quarter. Gaming chips provided $2.9 billion. Analysts had given targets of $25.1 billion for the data-center unit and $2.79 billion for gaming.
Blackwell is expected to generate a fresh wave of growth when it rolls out in the coming months. Analysts have downplayed concerns about delays, noting that the company still enjoys huge demand for its current generation of products. That could help Nvidia cope with any delays without a big financial hit.
In describing its challenges with Blackwell, Nvidia said it had to change a mask production step to improve its yield. A mask is the template used to burn the circuit pattern into materials deposited on a disk of silicon.
Production of Blackwell is set to ramp up in the fourth quarter and continue into the next fiscal year, Nvidia said.
During a post-results conference call, analysts sought more details on the amount of revenue that the new Blackwell chips would deliver and when. Huang and Chief Financial Officer Colette Kress stuck to their promise of billions of dollars in the fourth quarter, refusing to elaborate further.
The stock extended its declines as the call went on and answers weren’t provided.
In his typical fashion, Huang made high-level predictions on the future of the computing industry, arguing that a trillion dollars of equipment will be needed to replace outmoded gear in the world’s data centers. That replacement process is just beginning, he said.
AI is taking over computer search, helping companies speed up their business processes, and needed by countries to secure data, he said.
“It’s affecting how every layer of computing is done,” Huang said.