Nvidia Challenger Groq Discusses $6 Billion Valuation
The Takeaway
• AI chip startup Groq is in talks to raise money at nearly double its valuation
• The discussions follow a string of deals with Saudi Arabia, Finland
• The company has fallen short of its first quarter target deploying chips to data centers
Groq, an Nvidia challenger that develops chips to power large language models, has talked to investors about raising between $300 million to $500 million at a $6 billion post-investment valuation, according to two people with direct knowledge of the fundraising effort.
The deal discussions come as the company told investors its revenue will more than quintuple this year. If completed, the deal would double the startup’s post-investment valuation of $2.8 billion from an equity financing a year ago.
The company is seeking the cash after recently inking a chip deal with Saudi Arabia. In February, Groq said it had a $1.5 billion “commitment” from Saudi Arabia to expand the startup’s sale of chips to the country, though it didn’t say whether the commitment represented binding purchase orders. Groq this week also announced it was working on a data center in Finland to house its chips.
As a result of its contracts in Saudi Arabia, Groq has told investors that it expects to generate around $500 million in revenue this year, up from $90 million last year.
Groq primarily generates revenue from selling a cloud service for companies to run AI on its chips, similar to the way companies purchase access to OpenAI models or to AI models from Amazon Web Services. Groq also sells its chip systems and data center operating services to other companies, such as telecommunications company Bell Canada. Groq says nearly 2 million developers and teams have used its services.
The San Jose, Calif.-based Groq is one of a couple dozen startups trying to wrest a slice of the AI semiconductor market from Nvidia, which on Wednesday became the first company to reach a $4 trillion market capitalization. Groq CEO Jonathan Ross, one of the inventors of Google’s Tensor Processing Unit, which powers Google’s AI, is trying to develop a cheaper, faster and more energy-efficient alternative to Nvidia chips.
Groq’s chips aim to power AI that has already been developed, known as inference, as opposed to helping developers train new models, which typically requires a large concentration of chips in one place, tied together with costly networking gear. Groq says customers use its chips to run applications powered by open-source models such as Meta’s Llama 4, Mistral’s Mixtral and Google’s Gemma. Chip manufacturer GlobalFoundries makes Groq’s chips.
As of earlier this month, Groq had gotten about 70,000 of its chips online, according to someone close to the company. This is at least 30% short of a goal for the first quarter it had set last year. Groq chips are significantly less powerful than Nvidia’s Hopper or Blackwell AI chips.
Groq has raised more than $1 billion in equity from investors including fund giant Blackrock, the venture arms of Cisco and Samsung, as well as D1 Capital, Lee Fixel’s Addition and Tiger Global Management.
Groq and other chip startups face an uphill battle convincing AI developers to switch from Nvidia. Increasingly, they have sold chips to customers in the Middle East, where Nvidia chips are not as widely available. For instance, Palo Alto, Calif.-based AI chip startup SambaNova Systems has been supplying chip systems and software to Saudi petroleum giant Saudi Aramco, which is developing a large language model called Metabrain.
The high capital costs of supplying semiconductors has led these Nvidia challengers to seek new sources of capital, including debt. The 24 startups in The Information’s AI Chip Database have raised more than $7 billion from investors.
Cerebras, which sells chips specialized for training AI, last year filed to go public. But those plans stalled after a regulatory review of its ties to United Arab Emirates’ tech company G42, which is responsible for 90% of its revenue. In May, the CEO said during a conference that the company still aspires to go public this year.
Last year, D-Matrix, which is developing an AI-focused chip to run LLMs more effectively, set out to raise $250 million and has so far closed about $120 million of that round. It’s trying to raise another $180 million, targeting a total round size of $300 million, according to a person with direct knowledge of the matter.