Nvidia CEO Jensen Huang delivers a keynote address during the Nvidia GTC Artificial Intelligence Conference at SAP Center on March 18, 2024 in San Jose, California.
Justin Sullivan | Getty Images
Nvidia reported its fourth-straight quarter of triple-digit revenue growth on Wednesday, sailing past estimates on the top and bottom line while also issuing a forecast that topped Wall Street expectations. The company even bolstered its buyback program with a plan to repurchase $50 billion in shares.
But the stock dropped 7% in extended trading.
That’s life for Nvidia, which has ridden the artificial intelligence boom to a $3 trillion market cap, soaring almost nine-fold since the end of 2022 and surpassing every public company other than Apple in valuation. (It topped Apple for a stretch in June.)
In addition to reporting 122% annual revenue growth on Wednesday to over $30 billion, Nvidia said sales in the current period will jump about 80% to roughly $32.5 billion. Analysts were expecting close to $32 billion.
However, Stacy Rasgon, an analyst at Bernstein, told CNBC before the report came out that “buyside whispers” were closer to $33 billion to $34 billion, meaning Nvidia would have to dramatically surpass analyst estimates in its guidance in order to see a pop.
Rasgon, who recommends buying shares of the chipmaker, said there are no indications that demand is waning for Nvidia’s graphics processing units (GPUs), the core infrastructure for developing and running AI models.
“There’s still a ton of demand,” Rasgon said on CNBC’s “Closing Bell.” “They’re still shipping everything that they can sell.”
Nvidia said it expects to ship “several billion dollars” worth of Blackwell revenue in the fiscal third quarter, which ends in October. Blackwell is the company’s latest generation of technology, following Hopper. There had been some concerns that Blackwell would be delayed, but CFO Colette Kress said on the call with analysts that “supply and availability have improved.”
Still, “demand for Blackwell platforms is well above supply, and we expect this to continue into next year,” Kress said.
Other than missing the “whisper” numbers, some investors may be looking at Nvidia’s gross margin, which slipped a bit in the quarter to 75.1% from 78.4% in the prior period. That’s up from 43.5% two years ago and 70.1% in the fiscal second quarter of last year.
For the full year, the company said it expects its gross margin to be in the “mid-70% range.” Analysts were expecting full-year margin of 76.4%, according to StreetAccount.
‘Getting returns right away’
On the earnings call, analysts asked Nvidia executives about customers and whether they’re making money on their investment. Following the company’s prior report, Kress gave investors data points showing that a cloud provider could make $5 over four years selling access to $1 of Nvidia chips.
This time, Nvidia took a different approach. CEO Jensen Huang said on Wednesday’s call that Nvidia’s technology will be taking work away from traditional processors, like those made by Intel or AMD. He also said generative AI would start to do more coding, that companies like Meta can use Nvidia chips for recommender systems, and that nations are starting to buy more chips.
“The people who are investing in Nvidia infrastructure are getting returns on it right away,” Huang said.
Huang also said that next-generation AI models would require “10, 20, 40 times” more computing power, echoing comments recently made by former Google CEO Eric Schmidt.
The logo of Nvidia Corporation is seen during the annual Computex computer exhibition in Taipei, Taiwan.
Tyrone Siu | Reuters
“The frontier models are growing in quite substantial scale,” Huang said.
He said Nvidia’s main customers are vying to be first to produce new AI advancements.
“The first person to the next plateau gets to introduce a revolutionary level of AI,” Huang said. “The second person who gets there is incrementally better or about the same.”
But buying into Nvidia at these levels is a bet that the company can continue to outperform very high expectations and requires a willingness to accept the kind of stock volatility generally reserved for much smaller companies.
After reaching a record in June, Nvidia proceeded to lose almost 30% of its value over the next seven weeks, shedding roughly $800 billion in market cap. It’s since recovered most of those losses.
In the past two years, the stock has moved 5% or more in a single day on 50 separate occasions. For Microsoft, that’s happened only six times, which is one more than for Apple. At Meta, it’s happened 21 times. Tesla fans, however, can relate. Shares of the electric automaker have moved at least 5% on more than 70 trading days over that stretch.
One reason for Nvidia’s increased volatility is that it relies on a small group of customers — including those mentioned above — for an outsized amount of its revenue. Top execs at Alphabet and Meta both acknowledged recently that they could be overspending in their AI buildout, but said the risk of underinvesting was too great for them to not be aggressive.
For a third time since taking office in January, President Donald Trumpplans toextend a deadline that would require China’s ByteDance to divest TikTok’s U.S. business.
“President Trump will sign an additional Executive Order this week to keep TikTok up and running,” White House Press Secretary Karoline Leavitt said in a statement. “As he has said many times, President Trump does not want TikTok to go dark. This extension will last 90 days, which the Administration will spend working to ensure this deal is closed so that the American people can continue to use TikTok with the assurance that their data is safe and secure.”
ByteDance was nearing the deadline of June 19, to sell TikTok’s U.S. operations in order to satisfy a national security law that the Supreme Court upheld just a few days before Trump’s second presidential inauguration. Under the law, app store operators like Apple and Google and internet service providers would be penalized for supporting TikTok.
ByteDance originally faced a Jan. 19 deadline to comply with the national security law, but Trump signed an executive order when he first took office that pushed the deadline to April 5. Trump extended the deadline for the second time a day before that April mark.
Trump told NBC News in May that he would extend the TikTok deadline again if no deal was reached, and he reiterated his plans on Thursday.
Prior to Trump signing the first executive order, TikTok briefly went offline in the U.S. for a day, only to return after the president’s announcement. Apple and Google also removed TikTok from the Apple App Store and Google Play during TikTok’s initial U.S. shut down, but then reinstated the app to their respective app stores in February.
Multiple parties including Oracle, AppLovin, and Billionaire Frank McCourt’s Project Liberty consortium have expressed interest in buying TikTok’s U.S. operations. It’s unclear whether the Chinese government would approve a deal.
— CNBC’s Kevin Breuninger contributed to this report
Amazon Web Services is set to announce an update to its Graviton4 chip that includes 600 gigabytes per second of network bandwidth, what the company calls the highest offering in the public cloud.
Ali Saidi, a distinguished engineer at AWS, likened the speed to a machine reading 100 music CDs a second.
Graviton4, a central processing unit, or CPU, is one of many chip products that come from Amazon’s Annapurna Labs in Austin, Texas. The chip is a win for the company’s custom strategy and putting it up against traditional semiconductor players like Intel and AMD.
At AWS’s re:Invent 2024 conference last December, the company announced Project Rainier – an AI supercomputer built for startup Anthropic. AWS has put $8 billion into backing Anthropic.
AWS Senior Director for Customer and Project Engineering Gadi Hutt said Amazon is looking to reduce AI training costs and provide an alternative to Nvidia’s expensive graphics processing units, or GPUs.
Anthropic’s Claude Opus 4 AI model is trained on Trainium2 GPUs, according to AWS, and Project Rainier is powered by over half a million of the chips – an order that would have traditionally gone to Nvidia.
Read more CNBC tech news
Hutt said that while Nvidia’s Blackwell is a higher-performing chip than Trainium2, the AWS chip offers better cost performance.
“Trainium3 is coming up this year, and it’s doubling the performance of Trainium2, and it’s going to save energy by an additional 50%,” he said.
The demand for these chips is already outpacing supply, according to Rami Sinno, director of engineering at AWS’ Annapurna Labs.
“Our supply is very, very large, but every single service that we build has a customer attached to it,” he said.
With Graviton4’s upgrade on the horizon and Project Rainier’s Trainium chips, Amazon is demonstrating its broader ambition to control the entire AI infrastructure stack, from networking to training to inference.
And as more major AI models like Claude 4 prove they can train successfully on non-Nvidia hardware, the question isn’t whether AWS can compete with the chip giant — it’s how much market share it can take.
The release schedule for the Graviton4 update will be provided by the end of June, according to an AWS spokesperson.
The U.S. banking giant told CNBC on Tuesday that it’s planning to launch a so-called deposit token on Coinbase’s public blockchain Base, which is built on top of the Ethereum network. Each deposit token is meant to serve as a digital representation of a commercial bank deposit.
JPMD will offer clients round-the-clock settlement as well as the ability to pay interest to holders. It is a so-called “permissioned token,” meaning it is only available to JPMorgan’s institutional clients — unlike many stablecoins, which are publicly available.
“We see institutions using JPMD for onchain digital asset settlement solutions as well as for making cross-border business-to-business transactions,” Naveen Mallela, global co-head of Kinexys, J.P. Morgan’s blockchain unit, told CNBC Tuesday.
“Given the fact that deposit tokens would eventually be interest bearing as well, this would provide better fungibility with existing deposit products that institutions currently use,” he added.
Deposit token vs. stablecoin
JPMorgan said the benefit of launching a deposit token over a stablecoin is that it gives institutional clients a way to move money around faster and easier while still having a close connection with traditional banking systems.
A stablecoin is a type of digital token that’s designed to be pegged 1:1 to the value of a fiat currency at all times. The most popular stablecoins are Tether’s USDT and Circle’s USDC. The entire stablecoin market is worth approximately $262 billion, according to data from CoinGecko.
In the U.S., stablecoins remain broadly unregulated — although this is likely to change soon. The Senate is set to vote Tuesday on the GENIUS Act, legislation that would introduce formal regulation for such tokens.
Elsewhere, the European Union regulates stablecoins under its Markets in Crypto-Assets Regulation, or MiCA, while the U.K. has also laid out plans to regulate the crypto industry. Britain’s Financial Conduct Authority is currently consulting on proposals to require stablecoin issuers to ensure their tokens maintain their value against a given asset.
Read more CNBC tech news
JPMorgan’s digital asset chief told CNBC that the bank chose Coinbase as its blockchain partner since the crypto exchange is already a long-standing client and a leader in the crypto space.
JPMD has had “preliminary interest from large institutional players who want more native onchain cash solutions from pre-eminent and reputed financial institutions,” Mallela added.
Speculation had been building around JPMorgan’s new crypto offering after a trademark application filed by the bank for “JPMD” was made public Monday.
The trademark outlined a broad range of crypto services under the JPMD name, including trading, exchange, transfer and payment services for digital assets.
Various crypto media outlets had speculated whether the bank was about to launch its own stablecoin. However, JPMorgan says that, while its token may share some similarities with a stablecoin, it’s ultimately a different kind of product.