Every foundation model headline in 2026 is about OpenAI or Anthropic. Almost nobody is talking about the third company that actually might IPO first — a Toronto-born AI lab that decided, from day one, to never build a consumer product at all.
Cohere doesn't have a chatbot with hundreds of millions of monthly users. It doesn't have a splashy consumer app fighting for App Store rankings. What it has is a ~$7B valuation, $150M+ in ARR, and — per Crunchbase's 2026 IPO candidate list — a "probable" path to public markets this year. That combination makes Cohere one of the more interesting, least-discussed stories in the entire AI IPO conversation.
The Numbers: $7B Valuation, $150M+ ARR
Cohere's last private round priced the company at roughly $7 billion — a number that looks almost modest next to OpenAI's hundreds of billions or Anthropic's ~$180B mark, but is still a top-five independent foundation model valuation globally. Against $150M+ in ARR, that implies a revenue multiple in the mid-40s, which is expensive by traditional SaaS standards but cheap relative to how the market has priced frontier AI labs burning far more cash for far less revenue transparency.
~$7B
latest private valuation
2025 round
$150M+
annual recurring revenue
enterprise contracts
2019
founding year, Toronto + SF
Gomez, Zhang, Frosst
What matters more than the valuation itself is the composition of Cohere's revenue. Enterprise contracts tend to be sticky, multi-year, and priced on deployment scale rather than free-tier-to-paid conversion — the exact opposite of the consumer subscription churn dynamics that make OpenAI's and Anthropic's consumer revenue lines harder to underwrite for a public market.
Who Built Cohere
Cohere was founded in 2019 by Aidan Gomez, Ivan Zhang, and Nick Frosst. The pedigree here is not incidental — Gomez co-authored "Attention Is All You Need," the 2017 Google Brain paper that introduced the transformer architecture. Every model people currently argue about — GPT, Claude, Gemini, Llama — descends architecturally from that paper. Frosst spent years at Google Brain working alongside Geoffrey Hinton. This isn't a team that stumbled into AI during the ChatGPT hype cycle; it's a team that helped invent the underlying math a decade earlier.
The Toronto/San Francisco dual base also matters. Cohere is the flagship company in Canada's AI research cluster — anchored by the University of Toronto, the Vector Institute, and Hinton's legacy — and it has become something of a national champion, with PSP Investments (a major Canadian pension fund) as a backer alongside more typical Silicon Valley money.
The Product Lineup: Built for Business, Not Chat
Cohere ships three core products, and none of them are a chatbot:
Command
The generative model family — used for internal drafting, summarization, and workflow automation inside enterprise software, not as a standalone consumer app
Embed
Embedding models for semantic search and retrieval — the backbone of enterprise RAG (retrieval-augmented generation) systems that let a company search its own documents intelligently
Rerank
A model purpose-built to reorder search results by relevance — a narrow but extremely high-value piece of enterprise search infrastructure that most labs don't bother building
That's a deliberately narrow lineup. Cohere isn't trying to be everything to everyone — it's trying to be the default answer to three specific enterprise problems: generate text internally, search internal data intelligently, and rank that search output well. It's unglamorous. It's also exactly the kind of infrastructure large regulated companies actually buy.
The Real Differentiator: Deployment, Not Just Model Quality
Model benchmarks get all the attention, but Cohere's actual competitive edge is deployment flexibility. Most large enterprises — banks, insurers, government agencies, healthcare systems — are legally or contractually barred from sending sensitive data to a third-party API sitting outside their infrastructure. Cohere built its entire go-to-market around solving that constraint:
On-premises deployment
Models run entirely inside the customer's own data center or VPC — no data ever leaves the enterprise's security perimeter, which is often a hard requirement for banks and government contracts.
Multi-cloud support
Cohere runs on AWS, Azure, Oracle Cloud, and Google Cloud rather than locking customers into a single hyperscaler — a meaningful advantage for enterprises with existing multi-cloud commitments.
Data privacy guarantees
Customer data is contractually excluded from model training, addressing the single biggest legal objection enterprise counsel raises about adopting generative AI.
Nvidia partnership
A close relationship with Nvidia — both as an investor and infrastructure partner — gives Cohere preferential access to compute at a moment when GPU supply is still the binding constraint for every AI lab.
Who Backs Cohere
The investor list reads like a checklist of exactly who you'd want backing an enterprise infrastructure company: Nvidia (compute access and strategic alignment), Salesforce Ventures (direct line into enterprise CRM and business software distribution), PSP Investments (a Canadian pension fund giving the company patient, long-duration capital), and Inovia Capital (the Canadian VC firm that's been closest to Cohere since its earliest rounds).
That combination — strategic compute partner, strategic distribution partner, and patient institutional capital — is not the typical late-stage AI cap table dominated by momentum-chasing growth funds. It suggests investors underwriting Cohere on enterprise fundamentals rather than consumer hype multiples, which is exactly the profile that tends to travel better through an actual IPO roadshow.
Competing Against Giants With a Fraction of the Capital
Cohere competes against OpenAI, Anthropic, Google, and Meta's open-source Llama models — every one of which has dramatically more capital, more compute, and more researchers. That's the uncomfortable part of this story. Cohere isn't winning benchmark leaderboards, and it doesn't have the resources to out-scale Google or Meta on raw model capability. Its bet is narrower and, I think, smarter: don't compete on being the best model in the world, compete on being the model an enterprise can actually deploy, govern, and defend to a compliance team.
Meta's open-source Llama models are arguably the bigger long-term threat here, not OpenAI. A regulated enterprise that wants on-premises deployment and doesn't need frontier-level capability can increasingly just run Llama for free and fine-tune it internally. Cohere's durable moat has to be the surrounding enterprise tooling, support, and trust relationships — not the raw model weights, which are a commodity that gets cheaper every quarter.
My Take
I like Cohere's positioning more than its headline valuation multiple. The company made a real strategic choice in 2019 — before ChatGPT existed, before "foundation model" was a category anyone talked about — to build for enterprises instead of consumers. That's a bet that looked conservative for years and now looks prescient, because the consumer AI race has become a capital-destruction contest that only two or three players can survive, while the enterprise AI market is still being won contract by contract, RFP by RFP, in a way that rewards exactly the deployment flexibility Cohere has spent six years building.
The IPO case is genuinely reasonable, but it's not a slam dunk. $150M+ ARR at a $7B valuation is a real business with a real revenue multiple — not a story stock priced purely on model benchmarks and hype. That's a much easier story to sell to public market investors than OpenAI's or Anthropic's burn profile, and it's the reason I'd put decent odds on Cohere actually filing before either of the two much larger, much more famous labs get anywhere near a listing. The risk is that Cohere never gets to be the story — it stays the "also mentioned" company in every AI IPO roundup, permanently overshadowed by the two names everyone already knows.
If I were underwriting this, the question I'd want answered isn't "is the model good enough" — it's whether Cohere can keep growing ARR at a rate that justifies the multiple once Llama-based open-source deployments get good enough that enterprises stop paying for a proprietary alternative. That's the real competitive clock, and it's ticking faster than most people watching the OpenAI-versus-Anthropic headline fight seem to realize.
Cohere is the enterprise AI bet nobody is watching closely enough.
A $7B valuation, $150M+ ARR, and a "probable" 2026 IPO tag from Crunchbase — built by the team that literally invented the transformer.
Whether it actually files this year or slips to 2027, Cohere is proof that the AI IPO story isn't just a two-horse race between OpenAI and Anthropic.
Track upcoming tech listings on the IPO Tracker, monitor private valuations on Unicorns, and benchmark fund performance on the VC Performance dashboard at Value Add VC. Originally published in the Trace Cohen newsletter.
Get VC data most people never see — free.
Weekly benchmarks, valuations, and fund data. No spam, unsubscribe anytime.