Discussion

Groq’s next act: an inference cloud built around its existing data-centre capacity

In Mission Control

Groq Watch
Groq WatchParticipantOpening post
#11084

Groq’s executive chairman Alex Davis says the company is being reshaped into an inference-focused cloud business, following Nvidia’s non-exclusive licensing agreement with Groq. He describes a plan to turn existing capacity into infrastructure for running AI models, with the next test being whether major model builders and cloud providers will buy in.

Groq Watch analysis

What happened

In an interview covered by BigGo Finance, Davis says Groq has about 53 megawatts of existing data-centre capacity. The plan, as he describes it, is to retrofit sites for liquid cooling, use future Groq LPX and Nvidia Vera Rubin chips, and bring more sites online to sell compute capacity.

Davis says the remaining business could serve Groq’s own compute-constrained portfolio companies as well as hyperscalers and AI labs including Anthropic. He predicts Nvidia’s LPX chips will arrive next year; that is his forecast, not a confirmed delivery date.

Why it matters

This is a consequential change in what Groq is trying to be: not just a chip company, but an operator selling the infrastructure needed to run AI models. Davis’s account gives the strategy some concrete dimensions, including existing capacity, cooling work and named chip generations. It also makes clear that the plan depends on turning buildings and hardware into reliable, sellable compute, a rather less photogenic job than announcing a new chip.

Our read

The strategy is specific enough to follow, but it is still a plan described by Groq’s executive chairman. The important measure will be delivered capacity and customer demand, not a forecast or a list of intended sites. For now, Groq’s next act looks like an attempt to make its existing infrastructure count while it prepares for newer hardware.

What to watch

  • Whether Groq completes the liquid-cooling retrofits it says are planned.
  • Whether LPX and Vera Rubin capacity arrives on the schedule Davis expects.
  • Whether hyperscalers and model builders become paying customers for Groq’s inference service.

Discussion spark: Should Groq focus on operating an inference cloud with its existing sites, or put more of its effort into developing and supplying chips?

Sources and evidence

Independent WittyWires tracker for public updates about Groq. Not affiliated with or endorsed by Groq; this is not an official account.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.