Groq Reportedly Raises 650 Million Dollars for an Inference Cloud

What this is about

According to a TechCrunch report from 29 May 2026, the AI company Groq is raising up to 650 million US dollars from existing investors. The background is notable: in December 2025, Groq licensed its chip technology to Nvidia in an arrangement valued at around 20 billion dollars, with part of its leadership moving to Nvidia. The remaining team is now building a company with no chip business of its own that offers AI inference as a cloud service.

What the financing actually means

Inference is the running of an already-trained AI model, meaning the actual answering of requests. Groq plans to sell this inference as a cloud service, a category the industry often calls a "neocloud." Unlike before, the company no longer sells its own hardware but provides compute as a service. According to TechCrunch, the round is effectively backstopped: existing investors Disruptive and Infinitum have reportedly agreed to cover the full 650 million dollars should other investors decline their pro-rata shares. It is important to note that this is a report about an ongoing financing, not an officially closed deal.

Why it matters

The case shows a split between two bets: building AI chips and running AI inference as a service. The chip technology has been sold; the inference business continues as a standalone company. For the market, this signals that inference-cloud services are becoming their own highly valued category alongside the pure training of models. For companies running AI applications, more competition among inference providers could, over time, affect prices and latency.

In plain language

Imagine a bakery that spent years building and selling its own ovens. Now it sells the oven blueprints to a large manufacturer and instead opens a bakery that only bakes and sells bread. Baking bread is the inference here: the daily delivery of results, not the building of the machine.

A practical example

A software vendor in Munich runs an AI chatbot for 50,000 end customers and pays monthly for inference at a cloud provider. If an additional specialized inference provider with fast response times entered the market, the company could compare offers, redirect part of the load as a test, and weigh response times against costs. Before switching, a load test, a comparison of prices per million tokens, and a review of the data-protection and location model would be sensible.

Scope and limits

First, the financing is at this point a report and not an officially confirmed deal; the amount and terms may change. Second, building an inference-cloud business without its own chips means the company depends on purchased or licensed hardware, which can affect margins. Third, named personnel details such as a new management team are taken from reporting and should be checked directly against primary sources before any business decisions.

SEO & GEO keywords

Groq, AI inference, neocloud, inference cloud, Nvidia, 650 million dollar funding, LPU, AI infrastructure, Disruptive, Infinitum, AI compute, 2026