- There are signs across AI models, chips, and new form factors that the market is getting frothy.
- OpenAI will need to generate big returns to justify its record funding round.
- Nvidia rival Cerebras filed to go public, but its prospectus raises a potential revenue weakness.
In years to come, Silicon Valley may view the events of the past few weeks as the moment the generative AI boom got too frothy.
Investors spent the summer wondering if top AI stocks could continue to justify soaring valuations in the face of absent returns from their massive AI spending. Now, signs have emerged that they're not yet done with generative AI mania.
As September rolled into October, some of the most significant investment areas in AI received fresh votes of confidence — be it in the chip space, the large language model arena, and devices.
But there could be reason to feel cautious about the latest developments.
The Nvidia rival with just one main customer
Cerebras, a chip startup, included an eyebrow-raising detail in its IPO prospectus. But first, here's a quick primer on the Sunnyvale-based company.
Founded in 2015, Cerebras aims to accelerate AI development by designing processors to train and deploy smarter LLMs. It believes these processors, among other offerings, can compete with the powerful chips sold by Nvidia.
That's no walk in the park. Nvidia has emerged as one of the biggest winners of the generative AI boom, having grown from a market capitalization of $364 billion at the start of 2023 to over $3 trillion.
As things stand, however, Cerebras' plan to take it on involves generating the vast majority of its revenue from just one customer: Abu Dhabi-based AI firm G42.
According to its IPO prospectus, Cerebras — backed by investment firms such as Altimeter, Benchmark, and Coatue — generated 83% of its $78.6 million revenue in 2023 and 87% of the $136.4 million revenue generated in the first half of 2024 from G42.
Founded in 2018, G42 has emerged as one of the Middle East's biggest AI opportunities. Abu Dhabi's sovereign wealth fund, Mubadala, owns a stake, as do Microsoft and the private equity firm Silver Lake. There appears to be reason to work with G42, then, but by Cerebras' admission, an "adverse" change in the relationship could harm the chip firm.
While Cerebras has sought to strengthen its customer base — it reached an agreement with oil giant Saudi Aramco in September — its revenue concentration will weigh on investors as it aims to raise $1 billion in its public listing at a $7 billion to $8 billion valuation, per Bloomberg.
OpenAI reaches dizzying new heights
Warning signals emerged elsewhere as OpenAI shook up the market for LLMs on Wednesday by closing the most valuable funding round in Silicon Valley history.
While the ChatGPT maker announced that it had raised $6.6 billion at a $157 billion valuation from a series of new and existing investors, it did so with a few notable asterisks.
For one, per the Financial Times, OpenAI has asked its investors not to back its rivals, which include Anthropic, Elon Musk's xAI, and Safe Superintelligence (SSI), the startup founded in June by OpenAI's former chief scientist, Ilya Sutskever.
While it has the right to do this, the ask is unusual in the world of startup investing, where venture capitalists often spread their bets. Gary Marcus, a professor who testified next to OpenAI's Sam Altman at a Congress hearing in 2023, described the move as OpenAI "running scared." David Sacks, general partner at Craft Ventures, called the ask a "shady" one. "The mask is really coming off," he wrote on X.
It is also worth taking a closer look at some of the investors in its latest round. As my colleague Darius Rafieyan noted earlier this week, some of the most "prolific bubble chasers" — investors with a track record of betting on companies close to their peak — now count themselves as OpenAI backers. Think SoftBank, the Japanese conglomerate that burned billions of dollars on WeWork, and Tiger Global, an investment fund that suffered brutal losses in recent years after going on huge startup betting sprees.
This has taken place against a backdrop of wider challenges. OpenAI no longer has a chief technology officer after Mira Murati quit the firm late last month, becoming the latest top-tier executive to depart.
The company has also been battling to reverse its lossmaking operations. According to a September report from The New York Times, OpenAI's monthly revenue grew to $300 million in August —but it has projected $5 billion in losses this year.
AI's new form factors get a Meta push
The other notable sign of market frothiness came as Mark Zuckerberg unveiled a bold new wearable device for the AI era: Meta's Orion.
The augmented reality glasses, introduced at the company's Connect conference in late September, put AI front and center by embedding the technology in a way that "can sense and understand the world around you," according to Meta.
While they have impressed — Nvidia CEO Jensen Huang gave a glowing endorsement of the wearables in a promotional video, and Meta's stock responded positively after the event — there are a few matters to consider beyond the hype for it.
The first is cost. According to The Verge's Alex Heath, who spoke to Meta executives, the cost of building a single Orion unit at present is a steep $10,000.
Even if that cost can be brought down — something Orion lead Rahul Prasad told The Verge would be a focus of the device's next development phase — Meta will need to contend with the fact that appetite for these wearables hasn't exactly impressed in the past.
Meta rival Snapchat took a go at introducing glasses to the market in the mid-2010s but failed to make a splash after incurring a $40 million loss on them. It is taking another crack with AI-embedded eyewear, announcing a partnership last month with OpenAI to embed its tech.
Other form factors for AI that have emerged recently, such as the Humane AI pin that launched in April and the pocketable rabbit R1 introduced in January, have largely flopped, too.
Where does this leave us? In short, a lossmaking startup must justify its $157 billion valuation. A chip firm worth up to $8 billion must persuade investors not to worry about its undiversified revenue streams. Meta will need to convince consumers to switch to another way of interacting with AI entirely.
All of these things may come to pass. If so, this will be seen as a juncture where AI goes on to truly revolutionize the industries it touches — and repay the faith of investors. But if history is anything to go by, the high-stakes tech world is rarely plain sailing.