Nvidia Challenger Groq Enlists Morgan Stanley to Raise $300 Million
The Takeaway
How much does it cost to beat Nvidia? Groq is one of the startups trying to find out. It has tapped Morgan Stanley to help raise $300 million to develop specialized chips to run AI models.
Groq, an Nvidia rival that develops specialized server chips and software for artificial intelligence, has told investors it wants to raise about $300 million, according to two people who spoke to the company’s leaders. It has tapped Morgan Stanley to help with the fundraising, according to one of the people and a third person. The new round would roughly double its total funding, bringing it closer to startup rivals such as Cerebras in terms of total cash raised.
Started nearly eight years ago by Jonathan Ross, one of the inventors of Google’s specialized chip called the tensor processing unit, Groq has positioned itself as a cheaper and faster alternative to Nvidia chips, which have long been the gold standard for AI developers. Nearly 20 AI chip developers, including Groq and Cerebras, have collectively raised more than $5.5 billion, according to The Information’s AI Chip Database. But success has been elusive for the vast majority of them, underscoring the difficulty and expense of the task.
Groq wants to land an offer from investors by the end of month, according to one of the people. Groq previously raised more than $367 million from investors including Tiger Global Management, D1 Capital and Lee Fixel’s Addition at a valuation of more than $1 billion. It has still not set a valuation for the current round. One manager of a special purpose vehicle aiming to participate in the new round has told potential investors in the SPV that Groq would be worth more than $3 billion, according to a person who saw the pitch.
Startups sometimes hire investment banks for private fundraisings if the amount they’re seeking is large or if they are finding it hard to raise on the terms they want, and bankers sometimes use the opportunity to see whether they could solicit acquisition offers from other companies.
Groq designs chips for AI inference, or semiconductors for servers that run AI that has already been developed, as opposed to chips that help companies like OpenAI train new models. The company runs a set of open source models on its chips, including Meta’s Llama 3, Google’s Gemma and Mistral’s Mixtral, which developers can access. The company is offering the service for free but plans to charge for it starting in June, according to a person involved with the company.
Groq also plans to generate revenue from larger customers, such as government agencies and financial firms, who want to buy servers with Groq chips and put them in their own data centers, the company said recently.
Poe, Quora's AI assistant app, which allows users to access several AI models, said in a post on X that users can access and run the Llama 3 model on servers with Groq chips. This allows developers to test their apps on different servers to compare speed based on chips. Groq is also working with Aramco Digital, the technology arm of Saudi Arabian petrochemical giant Saudi Aramco, on creating an AI computing center in the country.
Groq’s fundraising process, which started months ago, overlapped with a dust-up at one of its biggest and earliest backers, Chamath Palihapitiya’s Social Capital. In March, Palihapitiya fired partner Jay Zaveri, who was on Groq’s board at the time, and investor Ravi Tanuku after they arranged a special purpose vehicle to invest in Groq without the full go ahead from Palihapitiya, according to Bloomberg.
Representatives for the former Social Capital investors have said that they were unjustly terminated. Zaveri left the board and Social Capital has replaced him with the VC firm’s chief financial officer Steven Trieu for the interim, according to the person involved with the company. Former Amazon executive Raju Gulabani also left the Groq board at the start of the year, according to his LinkedIn. In February, Groq said it added Harvard professor Youngme Moon to its board of directors.
Palihapitiya said on X last year that “Groq is probably the most consequential AI company you haven’t heard of—but they’ve been working hard for years to build the fastest hardware for AI.” He wrote an investment check to Ross before Ross had even incorporated a business, he has said.
Even with more money, Groq will face an uphill battle as it aims to get developers to switch off of Nvidia’s chips and the accompanying software that they’re familiar with, known as Cuda. Much larger companies, including Amazon, Google, and Microsoft, have been trying for several years to develop an AI server chip to wean their cloud customers off Nvidia’s, but they’ve made limited progress.
Some Nvidia challengers have faced problems. U.K.-based Graphcore, founded by chip veterans, raised hundreds of millions of dollars but has explored a sale after heavy losses, according to The Telegraph.