Nvidia Reels After China’s
A.I. Breakthrough
The
tech industry has had an insatiable appetite for Nvidia’s chips over the last two
years. But the feast may be over sooner than many had expected.
Nvidia,
which soared to the top of the stock market by selling the computer chips fueling the world’s artificial intelligence boom, has been dealt
a tough reality check by a small Chinese company that showed it could do more with
less of what Nvidia makes.
On
Monday, shares of Nvidia plunged 17 percent after the company, called DeepSeek, showed that it could train a cutting-edge A.I. system
with a fraction of the Nvidia chips that had been used in the past by OpenAI, the maker of ChatGPT. The
company lost roughly $600 billion in market value, on what was its worst trading
day since the pandemic sell-off in March 2020.
DeepSeek’s release challenged a tech industry consensus
that in order to build bigger and better A.I. systems, companies would have to spend
billions and billions of dollars on new data centers.
At the center of those data centers
would be the one thing that, perhaps until now, no A.I project could do without:
a huge cache of Nvidia’s chips.
The
Silicon Valley company, by some estimates, controls 90 percent of the market for
specialized chips used to build A.I. systems. It has had a remarkable run since
OpenAI released ChatGPT in late
2022. Over the past two calendar years, Nvidia’s revenue has jumped more than 200
percent to $126 billion, while the total value of the company has rocketed 700 percent
as of Friday’s market close, peaking at $3.62 trillion in November.
But
DeepSeek’s apparent breakthrough has shown that the appetite
for Nvidia’s chips may not be as limitless as some had imagined just a week ago.
While Nvidia is still in an enviable position — there is little competition for
its A.I. chips — the companies that have been buying its technology could slow down
their spending.
“Before,
A.I. was bigger, better, faster. Bigger chips equal bigger A.I. capabilities,” said
Patrick Moorhead, chief executive of Moor Insights & Strategy, a tech and semiconductor
research firm. “But this was so quick it raises questions about how long that is
true for Nvidia and whether people will need as many of its chips in the future.”
The
DeepSeek release also dragged down shares of other semiconductor
companies, including Broadcom, Micron Technology and Taiwan Semiconductor Manufacturing
Inc.
DeepSeek’s arrival has crystallized a concern that
was already casting a shadow over Nvidia’s business. Late last year, A.I. leaders
began to warn that the improvements to chatbots were slowing down. They had previously
relied on a simple formula to deliver advances: culling as much data as possible
from the internet and pushing into large language models — the technology that powers
chatbots — on ever-bigger computers.
But
that concept, which is known in the industry as Scaling Laws, has begun to fall
out of favor because tech companies are running out of
data. That has led companies to begin to experiment with new techniques to keep
improving their systems. It has also fueled questions
from Nvidia investors about the repercussions for its business.
At
the CES technology trade show in January, Jensen Huang, Nvidia’s chief executive,
said that the new techniques were “driving enormous demand for Nvidia’s chips.”
He said that companies are using Nvidia’s chips to power models that do more reasoning
in data centers, which means there will be more demand
for its technology, not less.
In
a statement on Monday, Mylene Mangalindan, Nvidia’s spokeswoman,
said DeepSeek shows that those new techniques are working.
She added that “DeepSeek is an excellent A.I. advancement.”
But
investors have been watching to see if the change in the way A.I. is being made
would throttle Nvidia’s business, and DeepSeek’s success
speaks to uncertainty about whether it will be cheaper to build and deliver A.I.
systems.
“This
was something that we have been working toward,” said Daniel Newman, chief executive
of Futurum Group, a tech research firm. “Everyone has
been pursuing this goal because the costs of training A.I. was too high.”
Not
everyone is convinced that Nvidia’s rise is stalling. Its stock has been volatile
and tumbled in August by more than 10 percent over reports the company would delay
shipping its newest artificial intelligence chip. It later rebounded.
In
a note for investors on Monday, Stacy Rasgon, a semiconductor analysts at Bernstein Research, said DeepSeek spent more money to build its system than it claimed.
He added that being able to build more A.I. systems more efficiently should mean
more demand because more companies can afford to invest in them.
The
panic over the weekend, Mr. Rasgon said, “seems overblown.”
The
changes in Nvidia’s value is the latest testament to how
A.I. continues to upend the stock market. Last year, Microsoft replaced Apple as
the world’s most valuable company for its early push into A.I. In June, Nvidia surged
past both companies to claim the crown.
Now,
Apple is back in the lead after it released its own A.I. system called Apple Intelligence
for iPhones. But there are signs that it may not hold that position for long. This
month, it disabled one of the signature capabilities — aggregating and summarizing
news notifications — after customers and companies complained that its software
was misrepresenting news reports.
“There
hasn’t been a winner here,” Mr. Newman said. “The tools have been just OK. But if
everyone can make better models at lower costs, then people may start adopting A.I.”