Groq
AI inference chip and cloud company that licensed its core tech to Nvidia.
On the record
Checkpoint: Nvidia's Groq 3 LPX rack commercial shipment on the Vera Rubin platform
The call
Groq turned its chip design into a Nvidia payout and a smaller neocloud business, competing on price against the company that now owns its own technology.
Groq makes specialized computer chips for running artificial intelligence software quickly, but in late 2025 it sold the rights to its main invention to Nvidia for about 20 billion dollars, and its top leaders left to join Nvidia. Now Groq runs a data center cloud service where it rents computing power to customers, but it has no special advantage anymore, Nvidia dominates this market with its own chips, big companies like Google and Amazon build their own AI chips, and a competitor called Cerebras just went public and is also trying to sell AI computing power. The real test will come in the second half of 2026 when Nvidia starts selling computers that use Groq's own licensed technology, which could either prove Groq's original idea was valuable or show that Nvidia can improve it and take all the customers.
Why
01
AI pressure: Cerebras Systems. Cerebras IPO and OpenAI supply deal pull sovereign and frontier lab inference contracts Groq was chasing. Could AI do the core job? Yes, likely.
02
Bigger players: Nvidia. Nvidia licensed Groq's own LPU design, hired its founder and president, and ships a rival Groq 3 LPX rack in H2 2026. Could a giant copy it? Yes, easily.
03
What protects it. Cash from the Nvidia deal and a 5 million plus developer base buy time, not any remaining chip exclusivity.
What would change this call
Commercial shipment volume and customer uptake of Nvidia's Groq 3 LPX rack in H2 2026. Shows whether Nvidia's own product built on licensed Groq IP displaces Groq's neocloud offering.
Groq's neocloud revenue and gross margin figures reported through 2027. Shows whether Groq's pivot to commodity inference hosting can sustain above the ~$500M revenue line.
Announcements of new sovereign or frontier-lab compute contracts naming Cerebras versus Groq. Shows whether Cerebras is pulling large customers away from Groq's pipeline.
Groq's neocloud revenue and gross margin figures reported through 2027. Shows whether Groq's pivot to commodity inference hosting can sustain above the ~$500M revenue line.
Announcements of new sovereign or frontier-lab compute contracts naming Cerebras versus Groq. Shows whether Cerebras is pulling large customers away from Groq's pipeline.
Key numbers
- $20B Nvidia license, ~$17B cash through 2026
- 5M+ developers claimed
- Headcount down 43% YoY to 315
Named risks
- Core LPU IP now non-exclusively licensed to Nvidia, leaving Groq with no proprietary tech moat.
- Headcount fell 43% year over year to 315 by June 2026 after executive exodus to Nvidia.
- Revenue figures wildly inconsistent across trackers ($172.5M ARR vs ~$500M vs earlier >$2B projections).
- Neocloud pivot is a commodity token-serving business facing thin margins against well-funded rivals.
- Cerebras IPO and reported OpenAI supply deal could pull sovereign and frontier-lab contracts from Groq's pipeline.
Rivals, as read in July 2026
| Rival | Direction | Evidence |
|---|---|---|
| Nvidia | losing to | Licensed Groq's own LPU IP, ships competing Groq 3 LPX rack H2 2026, took CEO/president Dec 2025 |
| Cerebras | losing to | Cerebras $5.55B Nasdaq IPO 2026-05-14 at $56-66B valuation, reported OpenAI supply deal |
| SambaNova | contesting | Rival cut 15% staff April 2025 but raised $350M Series E at $2.2B Feb 2026 |
| Hyperscaler custom silicon (Google TPU, Amazon Trainium, Meta MTIA) | losing to | In-house chips reduce reliance on third-party inference providers like Groq |
Questions people ask
- Is Groq profitable in 2026?
- Profitability is not disclosed. Revenue figures conflict sharply: Sacra cites a $500M 2025 projection versus ~$90M in 2024, while Latka's tracker cites $172.5M ARR as of mid-2025 (2026-07-03), suggesting Groq is still scaling, not confirmed profitable.
- Did Nvidia buy Groq?
- No. On 2025-12-24 Groq signed a non-exclusive licensing deal, not an acquisition, giving Nvidia rights to Groq's LPU chip tech for roughly $20 billion (~$17B cash through end-2026). Founder/CEO Jonathan Ross and president Sunny Madra moved to Nvidia as part of the deal.
- Who is Groq's biggest competitor now?
- Nvidia itself, which licensed Groq's own LPU technology and is shipping a competing 'Groq 3 LPX' rack in H2 2026 per its 2026-04-02 GTC announcement. Cerebras is also a major rival, having completed a $5.55B Nasdaq IPO on 2026-05-14 at $56-66B valuation.
- Could Nvidia replace Groq's core business?
- It largely already has for the chip business: Nvidia now owns a license to Groq's LPU IP and unveiled the 'Groq 3 LPX' rack for its Vera Rubin platform on 2026-04-02, shipping H2 2026, directly competing with what was Groq's original hardware product.
- How much did Groq raise in its 2026 funding round?
- Groq closed a $650M growth round on 2026-06-22, led by Disruptive and Infinitum, explicitly to fund its pivot into an AI inference cloud (neocloud) business, still valued around $6.9B per its last priced round.
- What happened to Groq's founder Jonathan Ross?
- Jonathan Ross, Groq's founder and CEO, moved to Nvidia as part of the December 24, 2025 licensing deal in which Nvidia acquired rights to Groq's LPU inference technology for roughly $20 billion, alongside president Sunny Madra.
Timeline
- 2026-06-22Groq closes $650M round to fund neocloud pivot; claims 5M+ developers
- 2026-05-14Cerebras completes $5.55B Nasdaq IPO at $56-66B valuation
- 2026-04-02Nvidia unveils Groq 3 LPX rack for Vera Rubin platform, shipping H2 2026
- 2026-02-01Groq distributes ~$7.6B to shareholders, ~64/share, first Nvidia-deal payout
- 2025-12-24Groq signs ~$20B Nvidia LPU licensing deal; CEO Ross, president Madra leave to Nvidia
- 2025-09-17Groq closes $750M Series E at $6.9B post-money valuation
- 2025-07-29Groq cuts 2025 revenue projection from >$2B to ~$500M per The Information
- 2025-04-29Groq becomes official Meta Llama API inference partner
- 2025-02-01Groq secures $1.5B Saudi HUMAIN commitment for Dammam data center
Revision history
20 July 2026
First assessed. Protection under test. The text above is fixed as of this date and is not edited afterwards.
In the same band
Sources
- Groq revenue, valuation & funding (2026-02-09)
- Groq Raises $650M to Scale Its AI Inference Cloud Business (2026-06-22)
- Inside NVIDIA Groq 3 LPX (2026-04-02)
- Groq slashes 2025 revenue projections to $500 million (2025-07-29)
- $172.5M ARR, $6.9B Valuation (2026-07-03)
- Groq - Headcount, Funding & Open Jobs (2026-06-20)
- Groq Raises $750 Million as Inference Demand Surges (2025-09-17)
- Meta and Groq Collaborate to Deliver Fast Inference for the Official Llama API (2025-04-29)
- AI chip company Groq raises $750m at $6.9bn valuation (2025-09-18)
- Independent AI Chip Companies Challenging NVIDIA in 2026 (2026-05-15)
- LLM Inference Hardware: An Enterprise Guide to Key Players (2026-02-28)
- Llama 4 Live Day-Zero on Groq at Lowest Cost (2025-04-05)