AI chipmaker Cerebras is trying to be the first major venture-backed tech company to go public in the U.S. since April and to capitalize on investors’ insatiable demand for Nvidia, now valued at $3.3 trillion.
While its position in artificial intelligence infrastructure represents a major tail wind, Cerebras has challenges — most notably a hefty reliance on a single Middle Eastern customer — that may prove too weighty to overcome in the company’s attempt to ride the Nvidia wave. Valued at $4 billion in 2021, Cerebras is reportedly seeking to roughly double that in its IPO.
“There’s too much hair on this deal,” David Golden, a startup investor at Revolution Ventures who led tech investment banking at JPMorgan Chase from 2000 to 2006, said in an interview this week. “This would never have gotten through our underwriting committee.”
Cerebras launched in 2016 and three years later unveiled its first processor. The company, headquartered in Sunnyvale, California, claims its current chip is faster and more efficient than Nvidia’s graphics processing unit, or GPU, for training large language models.
In 2023, Cerebras’ sales more than tripled to $78.7 million. In the first half of 2024, revenue climbed to $136.4 million, and growth appears poised to ramp up significantly, as Cerebras says in its prospectus that it’s signed agreements to sell $1.43 billion worth of systems and services, with prepayment expected before March 2025.
But the most glaring red flag in Cerebras’ filing relates to customer concentration. One company based in Abu Dhabi, United Arab Emirates, accounted for 87% of revenue in the first half of the year. The customer, G42, is backed by Microsoft, and it’s entirely responsible for the $1.43 billion purchase commitment.
Cerebras doesn’t list any other clients in its prospectus, but it does name a few on its website, including AstraZeneca, GlaxoSmithKline and the Mayo Clinic. Cerebras says in the filing that, in expanding its customer base, the company plans to “aggressively pursue opportunities in relevant sectors such as healthcare, pharmaceutical, biotechnology” and other areas “where our AI acceleration capabilities can address critical computational bottlenecks.”
In addition to its reliance on G42 for business, Cerebras counts the company as an investor, and it’s seeking clearance from the Treasury Department’s Committee on Foreign Investment in the U.S., or CFIUS, to give the Middle Eastern firm a bigger position. G42 has agreed to purchase a $335 million stake by April that, at current levels, would make it the largest owner. G42 can pick up $500 million more in Cerebras shares if it commits to spend $5 billion on the company’s computing clusters.
CFIUS has the authority to review foreign investments in U.S. companies for potential national security concerns. Cerebras said in its filing that it doesn’t believe CFIUS has “jurisdiction over G42’s purchase of our non-voting securities” but added that “there is no guarantee that CFIUS will approve” it. Reuters on Tuesday reported that Cerebras was likely to delay its initial public offering and call off its roadshow, scheduled to start next week, due to a national security review. Reuters cited people familiar with the matter.
U.S. lawmakers have expressed unease about G42’s historic ties to China, through both past investments and customer relationships. G42 said in February that it had sold its stakes in Chinese companies after Rep. Mike Gallagher, R-Wis., chairman of the Select Committee on the Chinese Communist Party, wrote a letter of concern to Commerce Secretary Gina Raimondo about what he called G42’s “extensive business relationships with Chinese military companies, state-owned entities and the PRC intelligence services.”
G42 didn’t respond to a request for comment.
Shunned by top banks
Even if it achieves CFIUS approval, Cerebras has a lot to overcome in trying to sell this deal to investors following a long stretch of suppressed valuations for smaller tech companies and a shortage of IPOs since the end of 2021.
Adding to Cerebras’ list of potential roadblocks is the fact that none of the primary tech investment banks are involved.
Goldman Sachs and Morgan Stanley have long dominated IPO underwriting in tech, with JPMorgan Chase also battling to get in the mix. They’re all absent from the Cerebras deal, and sources with knowledge of the process, who asked not to be named because the talks are private, said they stayed away in part due to the risks associated with customer concentration and foreign investment.
The deal is being led by Citigroup and Barclays, which are both large global banks but not the ones that get leadership positions on top tech IPOs.
Representatives from Citigroup, Goldman Sachs, JPMorgan and Morgan Stanley declined to comment. Barclays didn’t respond to a request for comment.
Cerebras’ auditor is BDO, which isn’t one of the so-called Big Four accounting firms. For the other three venture-backed IPOs this year, the accountants were KPMG (Reddit and Rubrik) and PwC (Astera Labs), which are two of the Big Four, along with Deloitte and Ernst & Young.
BDO declined to comment.
There’s also Cerebras CEO Andrew Feldman, who pleaded guilty in 2007 to one count of circumventing accounting controls when he was vice president of marketing at a public company called Riverstone Networks a few years earlier.
“What else could you have added to this to make it really difficult?” Revolution’s Golden said.
A Cerebras spokesperson declined to comment for this story.
The major Wall Street banks, for their part, are finding other ways to play in the burgeoning AI infrastructure market. Last week, Goldman Sachs, JPMorgan and Morgan Stanley were among a roster of banks that participated in issuing a $4 billion revolving line of credit to OpenAI. And on Friday, Nvidia GPU provider CoreWeave announced the close of a $650 million credit facility that was led by the top three tech banks.
Peter Thiel, president and founder of Clarium Capital Management LLC, speaks during the Bitcoin 2022 conference in Miami, Florida, on Thursday, April 7, 2022.
Eva Marie Uzcategui | Bloomberg | Getty Images
For Cerebras, there’s still a path to an IPO, given the sheer excitement around AI chips and the dearth of investable opportunities in the market.
Also, Nvidia is trading near a record. Mizuho Securities estimates that Nvidia controls 95% of the market for AI training and inference chips used for models like OpenAI’s GPT-4. Venture capitalist Peter Thiel said at the All-In Summit last month that Nvidia is the only company in the space that’s making money.
“Nvidia is making over 100% of the profits,” Thiel said in an on-stage interview at the event in Los Angeles. “Everybody else is collectively losing money.”
Cerebras is still in the money-losing column, reporting a second-quarter net loss of almost $51 million. However, excluding stock-based compensation, the company is close to breakeven on an operating basis.
Retail investor Jim Fitch, a retired homebuilder in Florida, is among those excited about the opportunity to get in early. Fitch, who said he sold out of his Nvidia stock years ago, told CNBC that the benefits outweigh the risks. He noted that Feldman, Cerebras’ co-founder and CEO, sold his prior company, SeaMicro, to Nvidia rival Advanced Micro Devices for more than $300 million over a decade ago.
Fitch is drawn to the promise of Cerebras’ technology, particularly its WSE-3 chip, which the company calls “the fastest AI processor on Earth,” packed with 4 trillion transistors.
Sanjay Beri, chief executive officer and founder of Netskope Inc., listens during a Bloomberg West television interview in San Francisco, California.
David Paul Morris | Bloomberg | Getty Images
Cloud security platform Netskope will go public on the Nasdaq under the ticker symbol “NTSK,” the company said in an initial public offering filing Friday.
The Santa Clara, California-based company said annual recurring revenue grew 33% to $707 million, while revenues jumped 31% to about $328 million in the first half of the year.
But Netskope isn’t profitable yet. The company recorded a $170 million net loss during the first half of the year. That narrowed from a $207 million loss a year ago.
Netskope joins an increasing number of technology companies adding momentum to the surge in IPO activity after high inflation and interest rates effectively killed the market.
So far this year, design software firm Figma more than tripled in its New York Stock Exchange debut, while crypto firm Circle soared 168% in its first trading day. CoreWeave has also popped since its IPO, while trading app eToro surged 29% in its May debut.
Read more CNBC tech news
Netskope’s offering also coincides with a busy period for cybersecurity deals.
Founded in 2012, Netskope made a name for itself in its early years in the cloud access security broker space. The company lists Palo Alto Networks, Cisco, Zscaler, Broadcom and Fortinet as its major competitors.
Netskope’s biggest backers include Accel, Lightspeed Ventures and Iconiq, which recently benefited from Figma’s stellar debut.
Morgan Stanley and JPMorgan are leading the offering. Netskope listed 13 other Wall Street banks as underwriters.
Meta CEO Mark Zuckerberg makes a keynote speech at the Meta Connect annual event at the company’s headquarters in Menlo Park, Calif., on Sept. 25, 2024.
Manuel Orbegozo | Reuters
Meta is planning to use its annual Connect conference next month to announce a deeper push into smart glasses, including the launch of the company’s first consumer-ready glasses with a display, CNBC has learned.
That’s one of the two new devices Meta is planning to unveil at the event, according to people familiar with the matter. The company will also launch its first wristband that will allow users to control the glasses with hand gestures, the people said.
Connect is a two-day conference for developers focused on virtual reality, AR and the metaverse. It was originally called Oculus Connect and obtained its current moniker after Facebook changed its parent company name to Meta in 2021.
The glasses are internally codenamed Hypernova and will include a small digital display in the right lens of the device, said the people, who asked not to be named because the details are confidential.
The device is expected to cost about $800 and will be sold in partnership with EssilorLuxottica, the people said. CNBC reported in October that Meta was working with Luxottica on consumer glasses with a display.
Meta declined to comment. Luxottica, which is based in France and Italy, didn’t respond to a request for comment.
Meta began selling smart glasses with Luxottica in 2021 when the two companies released the first-generation Ray-Ban Stories, which allowed users to take photos or videos using simple voice commands. The partnership has since expanded, and last year included the addition of advanced AI features that made the second generation of the product an unexpected hit with early adopters.
Luxottica owns a number of glasses brands, including Ray-Ban, and licenses many others like Prada. It’s unclear what brand Luxottica will use for the glasses with AR, but a Meta job listing posted this week said the company is looking for a technical program manager for its “Wearables organization,” which “is responsible for the Ray-Ban AR glasses and other wearable hardware.”
In June, CNBC reported that Meta and Luxottica plan to release Prada-branded smart glasses. Prada glasses are known for having thick frames and arms, which could make them a suitable option for the Hypernova device, one of the people said.
Last year, Meta CEO Mark Zuckerberg used Connect to showcase the company’s experimental Orion AR glasses.
The Orion features AR capabilities on both lenses, capable of blending 3D digital visuals into the physical world, but the device served only as a prototype to show the public what could be possible with AR glasses. Still, Orion built some positive momentum for Meta, which since late 2020 has endured nearly $70 billion in losses from its Reality Labs unit that’s in charge of building hardware devices.
With Hypernova, Meta will finally be offering glasses with a display to consumers, but the company is setting low expectations for sales, some of the sources said. That’s because the device requires more components than its voice-only predecessors, and will be slightly heavier and thicker, the people said.
Meta and Ray-Ban have sold 2 million pairs of their second-generation glasses since 2023, Luxottica CEO Francesco Milleri said in February. In July, Luxottica said that revenue from sales of the smart glasses had more than tripled year over year.
As part of an extension agreement between Meta and Luxottica announced in September, Meta obtained a stake of about 3% in the glasses company according to Bloomberg. Meta also gets exclusive rights to Luxottica’s brands for its smart glasses technology for a number of years, a person familiar with the matter told CNBC in June.
Although Hypernova will feature a display, those visual features are expected to be limited, people familiar with the matter said. They said the color display will offer about a 20 degree field of view — meaning it will appear in a small window in a fixed position — and will be used primarily to relay simple bits of information, such as incoming text messages.
Andrew Bosworth, Meta’s technology chief, said earlier this month that there are advantages to having just one display rather than two, including a lower price.
“Monocular displays have a lot going for them,” Bosworth said in an Instagram video. “They’re affordable, they’re lighter, and you don’t have disparity correction, so they’re structurally quite a bit easier.”
‘Interact with an AI assistant’
Other details of Meta’s forthcoming glasses were disclosed in a July letter from the U.S. Customs and Border Patrol to a lawyer representing Meta. While the letter redacted the name of the company and the product, a person with knowledge of the matter confirmed that it was in reference to Meta’s Hypernova glasses.
“This model will enable the user to take and share photos and videos, make phone calls and video calls, send and receive messages, listen to audio playback and interact with an AI assistant in different forms and methods, including voice, display, and manual interactions,” according to the letter, dated July 23.
The letter from CBP was part of routine communication between companies and the U.S. government when determining the country of origin for a consumer product. It refers to the product as “New Smart Glasses,” and says the device will feature “a lens display function that allows the user to interface with visual content arising from the Smart Features, and components providing image data retrieval, processing, and rendering capabilities.”
CBP didn’t provide a comment for this story.
The Hypernova glasses will also come paired with a wristband that will use technology built by Meta’s CTRL Labs, said people familiar with the matter. CTRL Labs, which Meta acquired in 2019, specializes in building neural technology that could allow users to control computing devices using gestures in their arms.
The wristband is expected to be a key input component for the company’s future release of full AR glasses, so getting data now with Hypernova could improve future versions of the wristband, the people said. Instead of using camerasensors to track body movements, as with Apple’s Vision Pro headset, Meta’s wristband uses so-called sEMG sensortechnology, which reads and interprets the electrical signals from hand movements.
One of the challenges Meta has faced with the wristband involves how people choose to wear it, a person familiar with the product’s development said. If the device is too loose, it won’t be able to read the user’s electrical signals as intended, which could impact its performance, the person said. Also, the wristband has run into issues in testing related to which arm it’s worn on, how it works on men versus women and how it functions on people who wear long sleeves.
The CTRL Labs team published a paper in Nature in July about its wristband, and Meta wrote about it in a blog post. In the paper, the Meta team detailed its use of machine learning technology to make the wristband work with as many people as possible. The additional data collected by the upcoming device should improve those capabilities for future Meta smart glasses.
“We successfully prototyped an sEMG wristband with Orion, our first pair of true augmented reality (AR) glasses, but that was just the beginning,” Meta wrote in the post. “Our teams have developed advanced machine learning models that are able to transform neural signals controlling muscles at the wrist into commands that drive people’s interactions with the glasses, eliminating the need for traditional—and more cumbersome—forms of input.”
Bloomberg reported the wristband component in January.
Meta has recently started reaching out to developers to begin testing both Hypernova and the accompanying wristband, people familiar with the matter said. The company wants to court third-party developers, particularly those who specialize in generative AI, to build experimental apps that Meta can showcase to drum up excitement for the smart glasses, the people said.
In addition to Hypernova and the wristband, Meta will also announce a third-generation of its voice-only smart glasses with Luxottica at Connect, one person said.
That device was also referenced by CBP in its July letter, referring to it as “The Next Generation Smart Glasses.” The glasses will include “components that provide capacitive touch functionality, allowing users to interact with the Smart Glasses through touch gestures,” the letter said.
Google CEO Sundar Pichai gestures to the crowd during Google’s annual I/O developers conference in Mountain View, California on May 20, 2025.
Camille Cohen | Afp | Getty Images
Alphabet shares rose on a Friday report that Apple is in early discussions to use Google’s Gemini AI models for an updated version of the iPhone-maker’s Siri assistant.
The company’s shares rose more than 3% on the Bloomberg report, which said Apple recently inquired of Google about the potential for the search giant to build a custom AI model that would power a new Siri that could launch next year. Google’s flagship AI models Gemini have consistently been atop key benchmarks for artificial intelligence advancements while Apple has struggled to define its own AI strategy.
The reported talks come as Google faces potential risk to its lucrative search deals with Apple. This month, a U.S. judge is expected to rule on the penalties for Google’s alleged search monopoly, in which the Department of Justice recommending eliminating exclusionary agreements with third parties. For Google, that refers to its search position on Apple’s iPhone and Samsung devices — deals that cost the company billions of dollars a year in payouts.
The Android maker has said its Gemini models will become the default assistant on Android phones. Google this year has showed Gemini doing capabilities that go beyond Siri’s capabilities, such as summarizing videos.
Craig Federighi, who oversees Apple’s operating systems, said at last year’s developer conference that the iPhone maker would like to add other AI models for specific purposes into its Apple Intelligence framework. Federighi specifically mentioned Google, whose Gemini can now hold conversations with users and handle input that comes from photos, videos, voice or text. Apple is also exploring partnerships with Anthropic and OpenAI as it tried to renew its AI roadmap, according to a June Bloomberg report.
Documents revealed during Google’s remedy trial showed executives from Apple were involved in the negotiations over using Google’s Gemini for a potential search option.