On May 23rd, according to multiple internal sources, Groq, a company focused on developing specialized server chips and AI software, has communicated its intention to investors to raise around $300 million in funding. The company has enlisted Morgan Stanley to assist in fundraising. If successful, Groq's total funding would double, further narrowing the gap with competitors like the startup Cerebras.
Jonathan Ross, co-inventor of Google's specialized TPU (Tensor Processing Unit) chips, founded Groq approximately eight years ago. The company aims to develop a cost-effective and fast product to replace Nvidia's chips, the benchmark in the AI industry. Media reports indicate that nearly 20 AI chip developers, including Groq and Cerebras, have collectively raised over $5.5 billion. However, given the industry's high complexity and costs, success remains challenging for most companies.
One source mentioned that Groq aims to secure investor commitments by the end of this month. Previously, Groq raised over $367 million from investors such as Tiger Global Management, D1 Capital Partners, and Lee Fixel's Addition, valuing the company at over $1 billion. Although Groq has not finalized the valuation for this funding round, a manager of a Special Purpose Vehicle (SPV) involved in the round disclosed to potential investors an expected valuation exceeding $30 billion.
When seeking substantial funding or facing unfavorable financing terms, startups often engage investment banks for private placements. Bankers sometimes use this opportunity to explore the possibility of attracting acquisition interest from other companies.
Groq's chips are primarily designed for AI inference tasks, supporting deployed AI programs on servers rather than aiding companies like OpenAI in training new models. Groq's chips can already run various open-source models, including Meta's Llama 3, Google's Gemma, and Mistral's Mixtral, with developers having free access. According to insiders, the service is currently free, but Groq plans to start charging from June.
Groq also plans to drive revenue growth by selling servers equipped with Groq chips to large clients like government agencies and financial firms, as well as deploying them in clients' data centers.
Quora's AI assistant app Poe mentioned in a post on the X platform that users can access and run the Llama 3 model on servers with Groq chips. This enables developers to test their applications on different servers to compare speeds based on different chips.
Additionally, Groq is collaborating with Aramco's tech subsidiary, Aramco Digital, to establish an advanced AI computing center in Saudi Arabia.
Groq's fundraising activities have been ongoing for months, during which a dispute arose with one of its key early investors, Chamath Palihapitiya of Social Capital. Reportedly, Palihapitiya dismissed board members Jay Zaveri and investor Ravi Tanuku in March due to organizing an SPV investment in Groq without Palihapitiya's full consent.
The ousted representatives from Social Capital claimed unfair treatment. Zaveri has left the board, with Social Capital temporarily replaced by CFO Steven Trieu. Additionally, former Amazon executive Raju Gulabani left Groq's board earlier this year, as seen on LinkedIn. In February, Groq announced the addition of Harvard professor Youngme Moon to its board.
Palihapitiya stated on the X platform last year, "Groq may be the most impactful AI company you've never heard of, but they've been toiling away on building the world's fastest AI hardware for years." He mentioned investing in Groq before Ross officially founded the company. Despite increased funding, Groq still faces significant challenges, aiming to persuade developers to move away from using familiar Nvidia chips and their accompanying software, Cuda. Over the years, major companies like Amazon, Google, and Microsoft have attempted to develop new AI server chips to reduce reliance on Nvidia for their cloud customers, with limited success.
Meanwhile, some of Nvidia's competitors are also facing challenges. Reports suggest that the UK AI chip company Graphcore, despite raising hundreds of millions of dollars, is considering the possibility of a sale after suffering significant losses.
