1. Headline & intro
Cohere just did something many AI startups only pitch in their slide decks: it turned the hype into hard recurring revenue. With $240 million in ARR and blistering growth, the Canadian company suddenly looks less like an underdog and more like the most conventional IPO candidate among the frontier AI labs.
This piece looks at why Cohere’s numbers matter, how its enterprise‑first strategy diverges from OpenAI and Anthropic, what an IPO would signal for the broader AI bubble, and why European buyers in particular should pay attention to a Canadian player in a field dominated by US big tech.
2. The news in brief
According to TechCrunch, citing an investor memo first reported by CNBC, Cohere ended 2025 with about $240 million in annual recurring revenue, beating its own target of $200 million. The memo reportedly highlights quarter‑over‑quarter revenue growth above 50% throughout the year.
Cohere, founded in 2019 and based in Canada, develops the Command family of generative AI models aimed at business use cases. The company says these models are efficient enough to run on relatively limited GPU resources, making them attractive for enterprises that want to control infrastructure costs.
In mid‑2025, Cohere launched North, a higher‑level enterprise platform for building secure, custom AI agents and workflows on top of its models. The startup is backed by major enterprise and chip investors including Nvidia, AMD and Salesforce. CEO Aidan Gomez said in late 2025 that an IPO may come “soon”; TechCrunch notes that a 2026 listing would likely coincide with potential public debuts from OpenAI, Anthropic and SpaceX/xAI.
3. Why this matters
Cohere’s $240 million ARR is more than a vanity milestone; it’s strategic ammunition. In a market where many AI startups still monetize via credits, pilots and vague “platform fees”, ARR gives public‑market investors something familiar to underwrite. It positions Cohere closer to a high‑growth enterprise software company than a moonshot research lab.
The big winner here is any CIO who has been wary of betting their roadmap on a single hyperscaler or on consumer‑centric AI brands. Cohere offers a relatively clean story: B2B‑only, no advertising, no consumer social products, and a focus on privacy‑sensitive deployments. For heavily regulated industries, that narrative lands better than “we also run the world’s biggest chatbot.”
The losers, at least on paper, are smaller enterprise AI vendors and systems integrators whose differentiation has been “we’ll wrap OpenAI/Anthropic for you.” If Cohere goes public with strong growth and deep-pocketed backers like Nvidia and Salesforce, it can credibly push a full stack: models, platform (North) and an ecosystem of partners. That compresses the margin opportunity for middlemen.
Cohere’s emphasis on efficient models that can run on constrained GPU budgets also directly addresses the most painful bottleneck in AI today: compute. If enterprises believe they can get 80–90% of the value with far lower GPU requirements, they may choose Cohere over frontier‑of‑frontier models. That’s a subtle but important competitive reframing: from “best raw benchmark scores” to “best ROI per GPU hour.”
4. The bigger picture
Cohere’s trajectory slots into a broader pattern: the frontier model race is converging on enterprise contracts as the real battleground. OpenAI launched ChatGPT Enterprise and dedicated business APIs; Anthropic has aggressively courted corporate customers through cloud partners; Google and Microsoft now sell AI more as a feature across their productivity and cloud suites than as a standalone miracle.
The historical parallel is cloud computing itself. In the late 2000s, the headline battles were about who had the most data centres and the most exotic tech. The companies that actually won, like AWS and Azure, were the ones that translated that into standardized, boringly predictable revenue streams and developer ecosystems. Cohere is trying to replay that playbook for the model layer.
There is also a defensive element. Pure model APIs are dangerously close to a commodity, especially as open‑source models like Llama and emerging European players such as Mistral narrow the quality gap. By launching North, Cohere is climbing the stack into “workspace” and “agent” territory, where switching costs are higher and where you can justify a subscription rather than metered usage alone.
Compared to OpenAI or Anthropic, Cohere has less brand recognition but also less political baggage. It isn’t tied to a single hyperscaler the way OpenAI is to Microsoft or Anthropic is to Amazon/Google. That gives it room to position as the neutral, multi‑cloud, enterprise‑safe alternative—an angle that has historically played well in B2B infrastructure (think Snowflake vs. native cloud databases).
If a cohort of AI labs, Cohere included, actually reaches the public markets in 2026, it will be the first real stress test of whether today’s AI valuations are sustainable outside private rounds. Cohere’s solid ARR will make it a bellwether.
5. The European / regional angle
From a European standpoint, Cohere is interesting for three reasons: jurisdiction, compliance and competition.
First, as a Canadian company, Cohere sits outside the US big‑tech duopoly but within a legal environment that historically has enjoyed partial adequacy status under GDPR for commercial organisations. For European CIOs nervous about US data transfers and surveillance regimes, that makes Cohere a politically easier sell than yet another Californian hyperscaler—though any cross‑border data flow still needs careful DPA work.
Second, the upcoming EU AI Act will impose obligations on general‑purpose AI providers around documentation, risk management and transparency. Cohere’s pure enterprise focus and relative lack of consumer exposure could actually be an advantage: fewer edge‑case harms, more controlled deployments, and a customer base accustomed to audits and compliance.
Third, Europe is trying to build its own AI champions—Mistral in France, Aleph Alpha in Germany, Stability AI’s European footprint, plus countless local integrators. Cohere’s potential IPO would raise the bar for what “serious” looks like: hundreds of millions in ARR, global partnerships, and a clear path to profitability. That benchmark will shape how European policymakers talk about “strategic autonomy” in AI.
For European enterprises—and especially those in smaller markets that lack deep local AI vendors—Cohere becomes another credible non‑US‑big‑tech option alongside open‑source stacks. Expect to see it more in RFPs across banking, manufacturing and telecoms.
6. Looking ahead
If Cohere does file for an IPO in 2026, the key question will be less “how big is the market?” and more “how defensible is its slice of it?” Investors will scrutinise:
- Gross margins: Are GPU and cloud costs under control, or is revenue being bought via subsidised compute?
- Customer concentration: How much revenue depends on a handful of marquee accounts or strategic investors like Salesforce?
- Moat vs. open source: What can Cohere offer that a well‑tuned open‑source model on a European cloud cannot?
Watch for deeper partnerships with systems integrators and regional clouds in 2026; those channels will determine how present Cohere is in European boardrooms. Also expect more verticalisation of its offering—packs or agents tailored to finance, legal and industrial use cases.
The risk is twofold. On one side, AI valuations could correct before or during an IPO window, turning a trophy listing into a cautionary tale. On the other, competition from hyperscalers bundling “good enough” AI into existing contracts could cap Cohere’s pricing power. Its answer must be: better governance, better TCO, and better fit for regulated industries.
If Cohere can keep growth high while showing improving unit economics, it may become the template for what a sustainable, post‑hype AI company looks like.
7. The bottom line
Cohere’s $240 million ARR proves that you can build a large AI business without chasing consumers or memes. An IPO would not just validate its own strategy; it would also force the market to separate AI companies with repeatable, enterprise‑grade revenue from those living on narrative alone.
For enterprises in Europe and beyond, the real question is simple: do you want your core AI infrastructure controlled by the same few giants that already run your cloud, or are you willing to back a more neutral specialist like Cohere while the window is still open?
