The problem
A bank, hospital network, or government ministry wants generative AI over its own documents — but compliance forbids sending customer records to a foreign consumer chatbot, and frontier models demand GPU budgets only hyperscalers can afford. Generic AI either fails the security review or fails the finance review.
How it works
Cohere trains the Command family of LLMs — deliberately efficiency-tuned so they run on limited GPUs — plus an Embed/Rerank stack for retrieval-augmented generation over enterprise data. Customers call the models via API or run them privately inside their own VPC, on-prem data centre, or sovereign cloud. North, launched in summer 2025, sits on top: a workspace where business teams build custom AI agents and document workflows without data ever leaving the enterprise boundary. CEO Aidan Gomez co-authored Google’s “Attention Is All You Need” transformer paper before founding the company in 2019 with Ivan Zhang and Nick Frosst.
Pain points
Data-residency laws and procurement rules that disqualify public AI clouds; GPU scarcity and inference bills; hallucinating chatbots that cannot cite company sources.
Business model
Enterprise platform subscriptions plus usage-based API pricing, weighted toward multi-year private-deployment contracts — the segment that drove the jump from $35M annualized revenue in March 2024 to $240M ARR in 2025.
Challenges
OpenAI, Anthropic, and Google are all chasing the same enterprise budgets with far larger balance sheets; the $5.5B valuation needs continued 50%-plus quarterly growth to look cheap; and every private deployment is operationally heavier than a single public cloud.
Funding
- Raised: $500M Series D (Jul 2024) led by PSP Investments with Cisco, Fujitsu, AMD, and Export Development Canada (Nvidia and Salesforce also back the company).
- Valuation: $5.5B at the July 2024 round.
Latest — February 2026
Cohere told investors it beat its $200M ARR target for 2025, closing the year at $240M with quarter-over-quarter growth above 50% all year, and CEO Aidan Gomez has said an IPO could come “soon” — potentially pitting a Toronto listing against the big US AI debuts expected in 2026.