Groq just raised $650M to become the inference cloud Nvidia can't control

Nvidia bought Groq's chip IP for $20B in December.

Poached the founder. Licensed the LPU technology.

Everyone assumed Groq was finished.

Yesterday Groq announced $650M in new funding — not to rebuild chips, but to become the inference cloud Nvidia now competes against.

Here is what happened:

Nvidia took Groq's LPU architecture and built its own inference hardware. The LPX system debuted at GTC in March. Same technology, Nvidia branding, Nvidia margins.

Groq kept the cloud business. 13 data centers across four continents. 5 million developers. Trillions of tokens processed weekly.

Now they are scaling to 200 megawatts by 2027.

This changes the enterprise inference market:

1. Nvidia now sells inference hardware directly. Every enterprise running Nvidia-based inference is buying from the company that absorbed Groq's engineering.

2. Groq's cloud runs that same technology independently. Enterprises can choose between Nvidia's vertically integrated stack or Groq's vendor-neutral inference cloud.

3. The new leadership team came from xAI, Meta Datacenters, and Apprenda. These are operators who have built hyperscale infrastructure, not chip designers.

The real signal for your infrastructure strategy:

Inference is no longer a side workload attached to training. It is the primary compute demand. IDC projects AI-centric compute spending will exceed $200B by 2027. Gartner says over 50% of enterprise workloads will incorporate generative AI by 2028.

Groq's pivot from chip company to inference cloud tells you where the margin is moving. The value is not in the silicon anymore. It is in the managed inference layer that sits between your models and your users.

If you are building an AI infrastructure strategy around a single vendor's hardware stack, you are now exposed. Nvidia controls the chips. Groq controls a cloud running the same architecture. AMD, Intel, and custom ASIC players are all targeting inference.

Audit your inference architecture today. Map which vendor's hardware your production models run on. Identify whether you are locked into a single provider's pricing and availability. The inference market just became a negotiation, not a monopoly.

SOURCE: https://groq.com/newsroom/groq-raises-usd650m-to-scale-its-ai-inference-cloud-business
VERIFIED: Bloomberg, TechCrunch, SiliconANGLE, Groq official announcement
SIGNAL: The company whose chip technology Nvidia licensed for $20B just raised $650M to compete against Nvidia in inference clouds. Enterprise inference infrastructure is no longer a one-vendor market.