Stack L2

Groq

Inference-only LPU maker at $6.9B post-money, scaling sovereign clouds via Bell, HUMAIN, Equinix

HQ US

Fit Score

avg of 4 segments

76/100

Growth

Comp

—

Learning

Stability

Culture

Recent Business Signals

No business signals recorded in the last 90 days.

Strategic Position

What they do best

Deterministic-latency, high-throughput open-model inference — evidenced by day-one OpenAI gpt-oss serving at ~500 tok/s (signal 7d8465ab) and Hugging Face provider status (signal 9edb3fe6).

Their bet

Becoming the default sovereign/non-NVIDIA inference utility — backed by Bell's 500 MW target (33060c3f), HUMAIN Dammam (bc6a227e) and the Helsinki/Equinix footprint (2fa4a708).

Top risk

Sovereign deals are MW commitments, not delivered revenue; if the next 6 months show no capacity milestone beyond Bell's first ~7 MW Kamloops site or the HUMAIN Dammam build, the announced pipeline is overstated.

Compared to peers

Direct competitor

Cerebras Systems

Cerebras also chases record-fast cloud inference but via wafer-scale single-chip; Groq's bet is many small deterministic LPUs plus a broader sovereign-deployment footprint.

Substitute

AMD

Buyers can default to AMD MI300/MI350 merchant GPUs for inference with mature ecosystem pull; Groq must win on latency/throughput-per-dollar, not optionality.

Why someone would join

1.Closed $750M on 2025-09-17 at $6.9B post-money with Samsung and Cisco on the cap table, and is hiring a Principal Hardware Architect (2026-02-26) plus Senior Silicon Design Engineer (2026-04-22) — funded next-gen LPU silicon work.
2.Commercial pull is concrete: day-one OpenAI gpt-oss partner (2025-08-05, ~500 tok/s), Hugging Face provider reaching 5M+ developers (2025-06-16), and Bell Canada's 500 MW program — capacity demand, not a research project.

Recent Hiring (60 days)

RTL design
1
ASIC design
1
deterministic architecture
1
compiler development
1
ML compilers
1

Reverse-Hype Watch

!Bell Canada 'Bell AI Fabric' states 'build-out target of up to 500 MW' but only '~7 MW Kamloops, BC' is operational (2025-05-28) — 500 MW target ~70x demonstrated live capacity, largely aspirational vs current customer-backed deployment.
!Capacity claim partly unbacked: Bell 'Bell AI Fabric' states 'build-out target of up to 500 MW' but only 'first ~7 MW supercomputing centre in Kamloops, BC' is live; the 500 MW figure is a target, not deployed customer-backed capacity.
!Capacity target outruns delivery: Bell AI Fabric 'stated build-out target of up to 500 MW' vs only 'first ~7 MW supercomputing centre in Kamloops' operational (2025-05-28) — ~70x gap between target and live site.
!Bell 'Bell AI Fabric' states 'build-out target of up to 500 MW' but only '~7 MW Kamloops live' — stated target capacity ~70x current live deployment, unbacked by proportional contracted volume.

→ Get this data as JSONLast updated: May 16, 2026

Frequently asked about Groq →