News
DealsCorvex agrees USD 103 million take-or-pay deal to perform AI inference with Mid-Atlantic GPUsSovereignNeysa scales AI infrastructure in India with AMD Instinct MI350X GPUs on TytanInferenceMicrosoft Foundry adds Baseten, DeepSeek AI inference (MSFT:NASDAQ)ClustersDataOne and Rowan University sign engineering co-op program agreement for DataOne's Vineland AI factoryClustersLook inside Cal Poly's $3 million AI Factory (Video)DealsOpenAI Taps AMD Chips for 457 MW of AI Capacity in Six-Year TensorWave DealClustersCal Poly launches an AI data center on campus. Take a look insideInferenceArista takes aim at the next AI network battleground: distributed inferenceDealsOpenAI Signs 457 MW AI Data Center Deal as AMD Expands Its Challenge to NvidiaClustersBehind Dell AI Factory鈥檚 NVIDIA & AMD Enterprise ExpansionDealsCorvex agrees USD 103 million take-or-pay deal to perform AI inference with Mid-Atlantic GPUsSovereignNeysa scales AI infrastructure in India with AMD Instinct MI350X GPUs on TytanInferenceMicrosoft Foundry adds Baseten, DeepSeek AI inference (MSFT:NASDAQ)ClustersDataOne and Rowan University sign engineering co-op program agreement for DataOne's Vineland AI factoryClustersLook inside Cal Poly's $3 million AI Factory (Video)DealsOpenAI Taps AMD Chips for 457 MW of AI Capacity in Six-Year TensorWave DealClustersCal Poly launches an AI data center on campus. Take a look insideInferenceArista takes aim at the next AI network battleground: distributed inferenceDealsOpenAI Signs 457 MW AI Data Center Deal as AMD Expands Its Challenge to NvidiaClustersBehind Dell AI Factory鈥檚 NVIDIA & AMD Enterprise Expansion
EveryNeocloud
die shot of the LPU

Groq

Chip designer and inference cloud built on its own Language Processing Unit

Photo: Avivweinstein, CC BY-SA 4.0, via Wikimedia Commons

Groq is an American AI company that designs its own accelerator chip, the Language Processing Unit, and rents out inference capacity through GroqCloud. Founded in 2016 by ex-Google engineers led by Jonathan Ross, it moved from selling silicon to running a cloud business after a reported US$20 billion deal with Nvidia in December 2025.

Share on WhatsApp

  • Founded 2016 by ex-Google engineers led by Jonathan Ross
  • Designs its own inference chip, the Language Processing Unit
  • Samsung's Taylor, Texas plant took its first ever order from Groq
  • US$1.5 billion Saudi commitment announced February 2025
  • Nvidia deal reported at about US$20 billion in December 2025

What it is

Groq is a privately held American firm that designs a purpose-built AI accelerator and sells access to it as a cloud service. The chip, called the Language Processing Unit, is an application-specific integrated circuit aimed at inference, meaning the running of trained models rather than their training. Typical jobs include large language models, image sorting and predictive analysis. Alongside the silicon, the company supplies the hardware and software needed to speed up inference. Its GroqCloud platform lets developers call the Groq API and rent time on the chips. Head office sits in Mountain View, California, with further sites in San Jose, Liberty Lake, Toronto and London.

How it came about

The business began in 2016, created by a team of former Google staff under Jonathan Ross, who had worked on Google's TPU accelerator, and Douglas Wightman, an ex-Google X entrepreneur who led the firm at first. Social Capital's Chamath Palihapitiya put in US$10 million of seed money during 2017. A US$300 million Series C followed in April 2021, taking the valuation past US$1 billion. Maxeler Technologies, a dataflow computing firm, was bought in March 2022. The developer platform soft-launched in February 2024, and Definitive Intelligence joined the company a fortnight later. August 2024 brought US$640 million of Series D funding. A US$1.5 billion Saudi commitment arrived in February 2025, and a further US$650 million raise was reported in May 2026.

The compute

The chip started life under the name Tensor Streaming Processor, with the internal codename Alan, before being renamed to make its purpose clearer. Its design interleaves memory with vector and matrix arithmetic blocks, arranged in a functionally sliced layout that keeps data close to where it is computed. A single core runs in a deterministic way: the compiler schedules everything explicitly, and reactive hardware such as caches, branch predictors and reorder buffers is left out. That gives predictable timing and better energy use. The first part, a 25 by 29 millimetre 14nm die clocked at 900 MHz, delivers more than one TeraOp per second per square millimetre. A second version is planned on Samsung's 4nm node.

Where the GPUs are

Around a dozen facilities were in place by 2025, spread across the United States, Canada, the Middle East and Europe. One of them is a GroqCloud data centre in Dammam, Saudi Arabia, linked to a US$1.5 billion commitment from the Kingdom announced in February 2025. Manufacturing for the next chip generation sits with Samsung Electronics at its Taylor, Texas plant, using a 4-nanometre process. That order, placed in August 2023, was the first to be booked at the new factory.

Groq
A drawing made by EveryNeocloud from what this page says. It is a diagram, not a picture of the real provider.

Customers and deals

The main route to users is GroqCloud, opened to developers in soft launch during February 2024, which offers API access and rented use of the chips. Open-source large language models run on Groq hardware and are reachable by the public through the website and a developer playground. The Kingdom of Saudi Arabia is a named backer of expansion, through a US$1.5 billion commitment tied to the Dammam site. Nvidia became a partner in December 2025 under a licensing arrangement for the inference technology.

Money and ownership

Early backing came from Social Capital, with US$10 million in 2017. A US$300 million Series C in April 2021, steered by Tiger Global Management and D1 Capital Partners, pushed valuation above US$1 billion. BlackRock Private Equity Partners led a US$640 million Series D in August 2024 at a US$2.8 billion figure. Saudi Arabia committed US$1.5 billion in 2025. Reported revenue for 2025 stands at US$500 million, while 2023 showed a net loss of US$88 million. A US$650 million round was reported in May 2026, backed by existing holders including Disruptive.

Its place in AI

The firm sits at the intersection of chip design and cloud delivery, running its own silicon rather than renting Nvidia GPUs. Its founder had a hand in Google's TPU, and the company's own part grew out of that lineage. Samsung's Taylor plant took its first ever order from Groq. In December 2025 Nvidia agreed to license the inference technology and take on several senior staff in a deal reported at about US$20 billion, described by Groq as non-exclusive, with the firm staying independent.

Questions and criticism

The Nvidia arrangement moves the founder and the president across to that company, which raises questions about how the remaining business is run. Groq has said it will carry on independently and that the licence is not exclusive. Heavy reliance on a single chip design and on outside foundry capacity is a structural risk. The 2023 accounts showed a net loss of US$88 million against 2025 revenue of US$500 million.

Questions people ask about Groq

What does Groq actually make?

It designs an AI accelerator chip called the Language Processing Unit, an application-specific integrated circuit built for inference work such as running large language models. It also supplies the surrounding hardware and software, and rents out access to the chips through its GroqCloud platform and API.

How is the Language Processing Unit different from a GPU?

It is a single-core, deterministic design. Memory sits interleaved with vector and matrix units, and the compiler schedules all execution explicitly. Caches, branch predictors and reorder buffers are absent, which gives predictable timing and helps energy efficiency on inference tasks.

Who founded the company?

Jonathan Ross, who had worked on Google's TPU, and Douglas Wightman, an entrepreneur out of Google X, started it in 2016. Wightman was the first chief executive. Ross later took the CEO role and remains in that post.

What happened with Nvidia?

In December 2025 the two firms announced an arrangement reported at roughly US$20 billion to license Groq's inference technology and move several senior executives over. Groq calls it non-exclusive and says it will keep running as an independent company.

Where does the computing sit?

About twelve data centres were operating by 2025 across the United States, Canada, the Middle East and Europe. One is a GroqCloud site in Dammam, Saudi Arabia, backed by a US$1.5 billion commitment from the Kingdom.

Good reading and sources

Where this page comes from

The facts on this page come from the Wikipedia article "Groq", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryNeocloud rewrote the text in its own words, with AI assistance, and compares every page with the article to make sure no passage is copied. It may be reused under the same CC BY-SA 4.0 licence with a link back.

It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

Photo credits on this page

The small pictures above come from Wikimedia Commons under free licences. Some are cropped to fit.

The weekly digest

Every Monday: the provider of the day, new pages on EveryNeocloud, and the conferences and anniversaries of the week ahead.