How To Invest In Groq

By Jozef Soja, Frank Downing

Research Analyst

Jozef Soja joined ARK in January 2023. As an Analyst, he focuses on artificial intelligence, cloud computing, and software-as-a-service in support of ARK’s Next Generation Internet strategy.

Director of Research, Next Generation Internet

As Director of Research, Frank guides the research of the Next Generation Internet team. Frank's research focuses on cloud computing and cryptoassets.

Self-Directed Investors have historically had limited access to private companies like Groq before they go public. All US investors can download the SoFi or Titan app and, with as little as $500, invest in a portfolio of private and public companies through the ARK Venture Fund.

ARK's investment strategy centers on identifying companies at the forefront of disruptive innovation, and we believe Groq represents a compelling opportunity at a pivotal inflection point in artificial intelligence (AI) infrastructure. Our research suggests that as AI scales from experimental deployments to real-time, consumer-facing applications, demand for compute will continue to shift from training to inference. In 2025, inference surpassed training in global AI compute revenue for the first time,¹ a structural transition that our research suggests will favor architectures that are purpose-built for speed and efficiency over general-purpose graphics processing units (GPUs) optimized for batch training workloads. 

Founded in 2016 by Jonathan Ross, one of the architects of Google's original Tensor Processing Unit (TPU), Groq designs Language Processing Units (LPUs)—application-specific integrated circuits engineered from the ground up for large language model (LLM) inference. Whereas GPU-based systems rely on High Bandwidth Memory (HBM) and complex scheduling to manage variable workloads, Groq's deterministic processor architecture uses on-die Static Random-Access Memory (SRAM) and a compiler-driven execution model refined iteratively over more than six years of development. The result is a fundamentally different cost and performance curve: independent benchmarks show Groq delivering approximately 300 tokens per second on Llama 2 70B, roughly ten times the throughput of Nvidia H100 GPUs on a per-user basis.² Energy efficiency follows the same logic, with Groq consuming an estimated one to three joules per token compared to ten to thirty for GPU-based inference.³ In our view, those are not incremental improvements but ones that represent the kind of order-of-magnitude cost declines that historically have unlocked exponential adoption across technology platforms. 

The competitive dynamics reinforce our conviction. As smaller, more efficient models proliferate for many use cases, the hosting layer becomes increasingly differentiated by cost per token, latency, and developer experience. Groq is well positioned on the first two dimensions, and its GroqCloud application programming interface (API) platform—now serving more than three million developers—demonstrates growing traction on the third. Enterprise adoption includes companies like Dropbox, Chevron, Volkswagen, and Robinhood, alongside strategic partnerships with Meta as an official Llama API provider and Hugging Face as an inference partner. Saudi Arabia's HUMAIN initiative also committed $1.5 billion to Groq-powered infrastructure in early 2025, further validating demand for dedicated inference capacity at scale.⁴ Those are signals that Groq's technology is meeting real production requirements, not merely winning benchmarks. 

Moreover, in December 2025, Nvidia agreed to a non-exclusive licensing deal valued at approximately $20 billion for Groq's inference technology, a transaction that we believe underscores the strategic significance of Groq's architecture within the broader AI compute ecosystem. Nvidia subsequently incorporated the Groq 3 LPU into its Vera Rubin platform, citing 35 times more inference throughput per megawatt compared to HBM-based GPUs.⁵ Groq continues to operate independently under CEO Simon Edwards, and based on current trajectory, the company has line of sight to a billion-dollar annualized cloud revenue run rate. At ARK Venture Fund's entry valuation, execution against this target likely would position Groq to exceed our hurdle rate of return. 

We recognize meaningful risks. Groq's post-licensing operational structure must demonstrate that it can sustain independent innovation and commercial momentum with a significantly smaller team. Application-specific integrated circuit architectures carry inherent inflexibility relative to GPUs, and Groq must continue to advance its compiler to maintain performance advantages as model architectures evolve. Competition from Cerebras and other inference-optimized chip designers will intensify as the market opportunity becomes more widely recognized. 

In our view, however, the asymmetric upside is substantial. ARK's research estimates that AI could drive $7 trillion in annual software investment and $1.5 trillion in annual infrastructure spending⁶ as knowledge worker productivity transforms through 2030. Within that landscape, the inference layer is where marginal investment is flowing, and where AI equity value is accruing most rapidly.⁷ Groq's architecture, developer ecosystem, and validated technology position it to potentially capture a meaningful share of that structural shift. We believe Groq is precisely the kind of long-term, innovation-driven opportunity that ARK Venture Fund was designed to pursue. 

Important Information  

We offer our view of Groq because it is part of the total ARK Venture Fund portfolio. To see the most updated portfolio, please visit here. Holdings subject to change. Not a recommendation to buy, sell, or hold any specific security. 

BEFORE INVESTING YOU SHOULD CAREFULLY CONSIDER THE FUND'S INVESTMENT OBJECTIVES, RISKS, CHARGES AND EXPENSES. THIS AND OTHER INFORMATION IS IN THE PROSPECTUS, A COPY OF WHICH MAY BE OBTAINED HERE. PLEASE READ THE PROSPECTUS CAREFULLY BEFORE YOU INVEST. 

There is no assurance that the Fund will meet its investment objective. The value of your investment in the Fund, as well as the amount of return you receive on your investment in the Fund, may fluctuate significantly. You may lose part or all of your investment in the Fund or your investment may not perform as well as other similar investments. Therefore, you should consider carefully the risks at the bottom of this page before investing in the Fund. 

ARK Investment Management LLC is the investment adviser Go the ARK Venture Fund. 

Foreside Fund Services, LLC, distributor.