Grok and Groq are two different companies. Grok, with a k, is the AI chatbot made by xAI, the Elon Musk company that SpaceX now owns. Groq, with a q, is an AI inference company founded in 2016 that runs open models on its own chips and sells access to developers through GroqCloud. In the Grok vs Groq mix-up, the two share a sound, not an owner.
The confusion is understandable, because Groq has been in the news all year. In December 2025 Nvidia licensed Groq’s chip technology and hired its founder in a deal reported at about $20 billion. Groq kept its cloud business and has since raised $1 billion at a lower valuation. This guide explains what Groq is, who owns it now, whether it is free and whether you can buy its stock. It also shows how to tell which of the two names you actually meant.
The Key Takeaways
- Not related: Grok is xAI’s chatbot, launched on 3 November 2023. Groq is a separate inference company founded in 2016 by former Google TPU engineer Jonathan Ross.
- The Nvidia deal: Nvidia licensed Groq’s LPU technology and hired Ross in December 2025. Its annual report books $14.4 billion of goodwill and $2.5 billion of technology for it, without buying the company.
- Who runs Groq now: Groq LLC is led by CEO Adam Winter and chaired by Alex Davis of Disruptive. Its August 2026 Series A valued it at $3.5 billion, down from $6.9 billion in 2025.
- Free to try: GroqCloud has a free plan. On gpt-oss-120b it allows 1,000 requests a day, and no payment method is needed until you upgrade.
- No public stock: Groq is private, so there is no Groq ticker. Nvidia, which licensed the technology, is the public company closest to it.
Grok vs Groq: The Short Answer
Grok is something you chat with, and Groq is something developers run models on. Grok is a closed AI model and chatbot that only xAI serves. Groq is infrastructure: it runs other companies’ open models, such as OpenAI’s gpt-oss and Alibaba’s Qwen, very fast, and charges developers per token. Groq does not run Grok, and xAI has no disclosed stake in Groq.
| Grok (with a k) | Groq (with a q) | |
|---|---|---|
| What it is | AI chatbot and model family | AI inference cloud, built on its own LPU chip |
| Made by | xAI, now branded SpaceXSI | Groq LLC |
| Owned by | SpaceX, since 2 February 2026 | Private investors, led by Disruptive |
| Started | Announced 3 November 2023 | Founded 2016 |
| Current flagship | Grok 4.7 | No model of its own; serves gpt-oss, Qwen, Llama and others |
| Who uses it | Anyone, on grok.com, X and the Grok apps | Developers, through an API key |
| Free option | Free tier with limits | Free API plan with rate limits |
| Public stock | Through SpaceX (SPCX) | None, Groq is private |
The one-letter difference matters more than it looks. Search for “groq ai” and you land on a developer console. Search for “grok ai” and you land on a chatbot. The rest of this guide is mostly about Groq, because it is the company most people know least about.
What Is Groq AI?
Groq is a US company that runs AI models fast for other businesses. It was founded in 2016 by Jonathan Ross, one of the engineers who created Google’s Tensor Processing Unit, the chip Google built for machine learning. Groq designed its own chip for running AI models, called the LPU, and it sells access to that hardware as a cloud service called GroqCloud.
The LPU: A Chip Built Only for Inference
Groq’s LPU, short for Language Processing Unit, is a processor built for inference, the step where a trained model answers a prompt. Training a model and running it are different jobs. Nvidia’s GPUs do both. Groq built the LPU to do only the second one, and to do it with very low delay between tokens.
Groq’s own explainer names four design choices: software first, an “assembly line” layout that moves data between units like a conveyor belt, deterministic timing, and memory on the chip itself. The last one is the big one. Groq says its on-chip SRAM has memory bandwidth “upwards of 80 terabytes/second”, against about eight terabytes a second for the separate memory a GPU uses. Those are Groq’s figures for its first-generation LPU, not an independent test. Nvidia’s page for the newer Groq 3 chip lists 150 terabytes a second of SRAM bandwidth per chip.
What GroqCloud Runs
GroqCloud runs open models from other labs, not a model of its own. As of 24 October 2026, the models open to every developer include OpenAI’s gpt-oss-120b and gpt-oss-20b, Alibaba’s Qwen 3.8 27B (a preview model) and Whisper for speech to text. Meta’s Llama 3.1 8B and Llama 3.3 70B are still listed, but only for enterprise customers. Grok is not on the list. Developers call these models through an API that Groq describes as mostly compatible with OpenAI’s client libraries. Groq also runs a simple chat demo at chat.groq.com, which needs a sign-in, but the product is the API.
Groq says more than six million developers build on GroqCloud across 13 data centers in North America, Europe, the Middle East and Asia-Pacific. Those numbers come from Groq’s August 2026 funding announcement.
Groq and Nvidia: What Happened in December 2025
Nvidia did not buy Groq.
On 24 December 2025, Groq announced a non-exclusive licensing agreement that gave Nvidia its inference technology. Groq’s founder Jonathan Ross, president Sunny Madra and other team members moved to Nvidia. Nvidia CEO Jensen Huang told employees, in an email CNBC obtained, that “we are not acquiring Groq as a company.” Groq said it would stay independent and that GroqCloud would keep running without interruption.
What Nvidia Paid
The deal was reported at about $20 billion, Nvidia’s largest ever. The firmer numbers are in Nvidia’s annual report for fiscal 2026. It says Nvidia licensed Groq’s language processing unit technology and hired certain employees, and that “no customer contracts, existing products, or equity interests were purchased.” Nvidia recorded $14.4 billion of goodwill, mostly for the workforce and future development, and a $2.5 billion technology asset. It paid $13.0 billion at closing, with $4 billion more due within a year.
Nvidia has already turned the license into a product. In March 2026 it announced the Nvidia Groq 3 LPX, a rack that packs 256 Groq 3 chips. On 24 August 2026, CNBC reported that the rack was in full production, with the neocloud Nebius as the first host. Groq said the same day that it would be among the first companies to deploy it, working with Dell.
The Lawsuit and the DOJ Probe
The Groq deal is now being tested in court and by regulators. In early October 2026, two former Groq engineers filed a proposed class action in Delaware’s Court of Chancery against Groq’s directors and officers, including Jonathan Ross, Tom’s Hardware reported. They allege that the deal kept billions from other shareholders. Groq called the suit “meritless”. Separately, The Register reported in September 2026, citing The New York Times, that the US Department of Justice had opened an antitrust probe into the deal.
No ruling has been reported in either matter.
Who Owns Groq Now?
Groq is owned by private investors, and Elon Musk is not one of them. Since its June 2026 funding round, the company, now called Groq LLC, has been chaired by Alex Davis, founder of the investment firm Disruptive, and led by CEO Adam Winter. Simon Edwards became CEO briefly after the Nvidia deal in December 2025. Winter now runs the company with a new team that includes chief operating officer Alan Rice, who previously worked at xAI.
Groq raised $650 million in June 2026, led by Disruptive and Infinitum. In August 2026 it announced a $350 million Series A that valued it at $3.5 billion, with Nvidia planning to take part. That is about half the $6.9 billion valuation Groq reached in September 2025, before the Nvidia deal took its founder and many of its engineers.
Groq has also joined the NVIDIA Cloud Partner program, so its data centers will run Nvidia hardware alongside its own LPUs.
Is Groq Free? GroqCloud’s Free Plan
Yes, Groq has a free plan for developers. You sign up at console.groq.com, create an API key, and call the models within the free rate limits. No payment method is needed until you move to the paid Developer plan, which asks for a credit card, a US bank account or a SEPA debit account. The free limits below are from Groq’s rate-limits page, checked on 24 October 2026.
| Model | Requests per minute | Requests per day | Tokens per minute | Tokens per day |
|---|---|---|---|---|
| gpt-oss-120b | 30 | 1,000 | 8,000 | 200,000 |
| gpt-oss-20b | 30 | 1,000 | 8,000 | 200,000 |
| Qwen 3.8 27B | 30 | 1,000 | 8,000 | 200,000 |
| Whisper large v3 | 20 | 2,000 | n/a | n/a |
Groq calls these a high-level summary and says your own account page shows the exact limits. On the paid plan, gpt-oss-120b costs $0.15 per million input tokens and $0.60 per million output tokens. Groq is one of several ways to use capable models for nothing, which our guide to the best free AI models compares. If you are deciding between a raw API key and a monthly plan, our explainer on API keys versus subscriptions covers what each one buys.
How to Get a Groq API Key
A Groq API key takes about a minute to create, and the free plan applies as soon as you have one. These steps follow Groq’s quickstart guide:
- Go to console.groq.com and sign in or create an account.
- Open the API Keys page at console.groq.com/keys and create a new key.
- Copy the key and store it as an environment variable, for example
export GROQ_API_KEY=your-keyin Terminal, so it never sits in your code. - Call a model such as gpt-oss-120b through Groq’s Python or JavaScript library, or any OpenAI-compatible client.
Can You Buy Groq Stock?
No. Groq is a private company and has no public stock or ticker symbol. Its most recent price is the $3.5 billion valuation from the August 2026 Series A. If you want public-market exposure to the LPU, the nearest company is Nvidia, which licensed Groq’s technology and now sells it inside the Groq 3 LPX rack. That is a bet on Nvidia, though, not on Groq.
Grok, by contrast, does trade, indirectly. xAI became a wholly owned subsidiary of SpaceX on 2 February 2026, according to SpaceX’s quarterly report. SpaceX is listed on Nasdaq as SPCX.
What Is Grok?
Grok is the AI chatbot and model family from xAI, the company Elon Musk founded in 2023. xAI announced Grok on 3 November 2023 and described it as “an AI modeled after the Hitchhiker’s Guide to the Galaxy.” SpaceX completed its acquisition of xAI on 2 February 2026. In October 2026 the company switched its branding to SpaceXSI, after Musk posted that “SpaceX is a super intelligence company”.
The current flagship is Grok 4.7, which xAI’s developer docs call its model “for code and everything else”. You use Grok on grok.com, in the Grok apps and inside X. For the full picture, including models, ownership and safety, see our guide to what Grok AI is, and for the plans see Grok pricing.
Why the Names Collide: Groq Came First
Groq had the name seven years before Grok. When xAI launched its chatbot in November 2023, Groq wrote an open letter to Musk on its blog. The post, dated 29 November 2023, opened: “Did you know that when you announced the new xAI chatbot, you used our name?” It added, “Plus, we own the trademark.” Groq suggested Musk rename the bot Slartibartfast, after a character from The Hitchhiker’s Guide to the Galaxy.
The letter was playful, and xAI kept the name anyway. Groq has since taken the post off its site, but an archived copy of Groq’s letter is still online. The mix-up is one of the most common naming errors in AI, alongside ChatGBT and Gemeni, which our list of the most misspelled AI names ranks by searches.
Grok vs Groq: Which One Did You Mean?
You meant Grok if you want to chat with an AI, ask about posts on X, or generate images. You meant Groq if you are a developer looking for fast, cheap API access to open models, or if you read about Nvidia’s chip deals.
The words around the name usually settle it. Elon Musk, X, SuperGrok, the X Subscription, Grok Imagine and grok.com all point to Grok. LPU, GroqCloud, API key, tokens per second, Nvidia, Jonathan Ross and console.groq.com all point to Groq.
Using Grok Without Picking a Side
Most people who type either name want the chatbot, and then they need to decide whether Grok is worth another subscription. You don’t have to pick a side. In Fello AI, Grok sits in the same Mac, iPhone and iPad app as ChatGPT, Claude and Gemini. Ask Grok and another model the same question, keep the better answer, and pay for one subscription instead of two.
The Verdict
Grok and Groq are not rivals. They are not even in the same business. Grok is a chatbot you talk to. Groq is a cloud that developers rent to run other people’s models fast. The more interesting story is Groq’s own. Nvidia paid billions for its chip and its people without buying the company. The Groq that remains has bet everything on its cloud, while a lawsuit and a federal probe play out. If you came here looking for the chatbot, you want Grok. If you came for fast open-model inference, Groq’s free plan is the easiest place to start.
FAQ
Are Grok and Groq the same company?
No. Grok is the AI chatbot from xAI, which SpaceX owns and which now brands itself SpaceXSI. Groq is an unrelated AI inference company founded in 2016. It runs open models on its own LPU chips and sells access through GroqCloud.
Is Groq owned by Elon Musk?
No. Groq is owned by private investors and chaired by Alex Davis of Disruptive. Elon Musk controls Grok through SpaceX, which bought xAI in February 2026. He has no role at Groq.
Did Nvidia buy Groq?
No. In December 2025 Nvidia signed a non-exclusive license for Groq’s LPU technology and hired its founder Jonathan Ross and other staff. Nvidia’s annual report says it bought no customer contracts, products or equity. Groq kept running GroqCloud as an independent company.
Which came first, Groq or Grok?
Groq came first. It was founded in 2016, and xAI announced Grok on 3 November 2023. Groq answered with an open letter to Musk on 29 November 2023, which said “we own the trademark.”
Does Groq run Grok?
No. Groq’s model list, checked on 24 October 2026, includes gpt-oss, Qwen, Llama and Whisper but not Grok. Grok is a closed model that only xAI serves, through grok.com, X, the Grok apps and the xAI API.
