For informational purposes only. Not financial advice, not a recommendation to buy or sell any security. Standing Order is a directory and research project, not an advisory service.
Thesis
Groq is the most prominent dedicated AI-inference silicon company, built around the LPU (Language Processing Unit) — a deterministic, single-core-cluster architecture optimized for low-latency token generation rather than NVIDIA-style training-and-inference generality. Founded by Jonathan Ross (lead architect of Google's first TPU), the company has raised over $1B from BlackRock, Cisco, Samsung Catalyst, and a Saudi Aramco-led 2024 round at meaningful valuation marks.
GroqCloud has emerged as a credible inference destination for open-weight models with throughput characteristics that NVIDIA-on-Hopper struggles to match at price. The investment case is that inference becomes a separate, much larger silicon TAM than training, and Groq's architecture is positioned to take share. Risks: Cerebras, SambaNova, Tenstorrent, and NVIDIA's own Blackwell-inference SKUs all compete for the same workloads.
Catalysts
- GroqCloud token-throughput and customer mix
- Datacenter capacity build-out pace
- Enterprise on-premise design wins
- Funding round valuations and path to public listing