Everything About AI
NextFin News - Cohere Inc., the Toronto-founded enterprise AI company, is in advanced talks to raise between $2 billion and $3 billion in a new funding round that would value the business at roughly $20 billion, according to people familiar with the matter. The deal, which could close as soon as next week although the timing may still slip, brings together existing backers and the Canadian government, and would crown Cohere the most valuable artificial intelligence startup headquartered outside the United States.
The headline number is only the start of the story. The real question is what a $20 billion price tag says about the AI funding market at the moment the biggest players are preparing to list.
The Deal and the Numbers Behind It
A $20 billion valuation would nearly triple the $6.8 billion mark Cohere set in August 2025, when it raised $500 million in a round led by Radical Ventures and Inovia Capital with participation from AMD Ventures, Nvidia, PSP Investments, Salesforce Ventures and the Healthcare of Ontario Pension Plan. A month later, in September 2025, the company added a further $100 million at a $7 billion valuation from the Business Development Bank of Canada and Nexxus Capital Management. Before that, in December 2024, Ottawa issued a letter of intent for up to C$240 million - roughly US$169 million - in Strategic Innovation Fund support tied to a multibillion-dollar AI data centre project, and in June 2023 the company raised $270 million at a $2.2 billion valuation.
The financial backdrop is strong enough to make the conversation plausible. Cohere told investors in a February 2026 memo that it ended 2025 with roughly $240 million in annual recurring revenue, beating its $200 million target, with quarter-over-quarter growth above 50 percent through the year and gross margins around 70 percent, up 25 basis points year over year. At $20 billion, new investors would be paying about 83 times that revenue run rate.
That multiple is the hinge of the whole story. It is high even by the standards of the 2026 AI boom, and it is being underwritten by a coalition that includes a national government - a detail that tells you as much about Ottawa's industrial policy as it does about Cohere's business. For context, the round would value Cohere at roughly the same level as Mistral AI, the French laboratory valued at about $13.7 billion as of July 2026, but on a revenue base that is smaller - implying a richer multiple than its European sovereign-AI peer.
The timing matters as much as the size. The raise comes as the AI funding market shows its first clear signs of strain. SoftBank's AI funding plans are facing scrutiny, with a $40 billion bridging loan maturing in March 2027 and lenders growing cautious about extending credit secured by private AI stakes. OpenAI's much-anticipated public listing, reported to be targeting a $1 trillion valuation, has faced delays. A Cohere deal closing now would be one of the last major private AI rounds priced before the public markets force a reckoning.
The Sovereign-AI Premium: A Structural Moat, Not a Cyclical Tailwind
The first question any buyer in this round has to answer is what it is actually paying for. Cohere's pitch is that it is not competing with OpenAI, Anthropic or Google for consumer attention. It is selling deployable, security-first models to enterprises and governments that cannot send their data to a U.S. consumer chatbot. Its North platform, launched in 2025, bundles agentic workflows with retrieval grounded in a customer's own data, and its models can run on a client's own hardware or in an isolated virtual private cloud. That architecture avoids the inference-cost burden that weighs on providers serving public users, and it is the reason gross margins sit near 70 percent.
The mechanism behind the premium is straightforward. Sovereign and regulated buyers - banks, telecoms, utilities, public agencies - face procurement and data-residency rules that effectively disqualify open consumer APIs. Cohere's cloud-agnostic, private-deployment model converts a compliance constraint into pricing power. In a June 2026 commentary on digital sovereignty, Cohere co-founder and CEO Aidan Gomez put the logic in its sharpest form:
True digital sovereignty is about choice and control. It is the ability to decide who sees your data, who modifies your systems, and who has the power to turn them off. Owning your AI means owning your future.
That is a structural moat. It does not depend on the AI funding cycle; it depends on data-residency regulation, which only tightens. The Canadian government's involvement is the tell. Ottawa has moved Cohere from a private portfolio company into a matter of national industrial policy. In May 2026 the federal government announced $66 million in support for 44 AI projects through the AI Compute Access Fund, part of a broader C$300 million sovereign compute program, and the 2025-26 budget earmarked C$925.6 million over five years for sovereign public AI infrastructure. During the May announcement, the federal AI minister said:
We are trying to build the best here in Canada.
A Canadian-owned foundation model is a strategic asset in the same category as semiconductor capacity or undersea cables - and that logic survives any single administration. The December 2024 letter of intent for up to C$240 million in Strategic Innovation Fund support, tied to a multibillion-dollar data centre project, is the financial expression of that logic: the state is not just a customer, it is a co-architect of the infrastructure layer.
The company itself has framed the capital-light model as a deliberate hedge. In its February 2026 investor memo, Cohere wrote:
By scaling compute resources proportionally to customer demand, we remain insulated from the speculative excesses surrounding the broader AI market, positioning Cohere for more sustainable growth. Our thesis is clearly resonating in the market. Our sales pipeline continues to grow as global organizations across regulated sectors choose Cohere as their trusted partner for secure AI adoption at scale.
But a moat is not a valuation. The premium is being priced as if sovereign demand were infinite and competition static. It is neither.
Cyclical Mania on Top of a Structural Floor
The clearest way to separate the two forces is to ask which part of the $20 billion would survive a funding winter. The structural floor is real. Cohere's annualized revenue grew from roughly $35 million in March 2024 to $100 million by mid-2025 and about $240 million by the end of 2025 - a compounding trajectory most software companies never see, and one where each incremental revenue dollar carries a 70 percent gross margin. That is a 7-to-1 revenue multiple in under two years, achieved while expanding margins rather than buying growth with discounted inference.
The cyclical leg is the multiple. At 83 times revenue, the round assumes the growth rate holds and that the market keeps rewarding AI scarcity. During the 2021 software boom, hypergrowth names traded at 30 to 40 times sales; the 2026 AI round has pushed that to 80-plus for a company still below $250 million in revenue. Multiples of this size are a function of capital abundance, not business quality. When capital cheapens, multiples revert first and fastest.
The historical analogue is instructive. In the late 1990s, networking and internet-infrastructure companies raised at stratospheric revenue multiples on the premise that the build-out of the internet was a once-in-a-generation structural shift. They were right about the shift and wrong about the price. The companies with real moats survived; the equity raised at the peak did not. Cohere's structural thesis may prove correct while the $20 billion entry point still loses money.
There is already evidence the cycle may be turning. Beyond SoftBank's strained financing, the broader venture market has grown selective: capital is concentrating in the one or two named leaders per category while the long tail faces down rounds. When the largest check-writers start rationing, the second and third tier feel it first - and a $20 billion valuation places Cohere squarely in the tier where every subsequent round must clear a higher bar.
The call, then, is this: the structural floor - sovereign demand, private deployment, government backing - is durable and will not mean-revert. The valuation multiple is cyclical and will. The two are being priced as one, and that is where the risk sits.
The Second-Order Effect: A Bifurcated AI Capital Market
The round's real significance is not Cohere's cap table. It is the signal it sends to the rest of the non-U.S. AI field. If a $20 billion mark clears, it becomes the reference price for peers such as France's Mistral AI, valued at roughly $13.7 billion as of July 2026 on about €300 million in annual recurring revenue, and for any sovereign-AI hopeful seeking a national champion. Capital will concentrate in the one or two companies per region that carry explicit government endorsement, because a state anchor de-risks the downside for co-investors.
The second-order consequence is a bifurcation. Labs with a government balance sheet behind them - Cohere in Canada, and similar national projects emerging in Europe and the Gulf - will find capital available even as the broader AI venture market tightens. Everyone else will face a higher cost of capital, longer diligence and down rounds. The funding winter, when it comes, will not be uniform; it will be selective, and it will favor the politically designated over the technically excellent.
This dynamic also explains why Ottawa is in the deal. The government is not merely seeking a financial return; it is buying insurance against technological dependency. That is a second-order effect most valuation debates miss: the round is as much an industrial-policy instrument as a corporate financing. For a middle power with a deep AI research base but a small domestic market, backing a national champion is the cheapest available hedge against being locked out of the foundational layer of the next computing platform.
The transmission channel runs through procurement, not just capital. Once a government has an equity stake and a strategic interest in a domestic model, its own agencies become anchor customers - a revenue stream that is sticky, long-duration and largely insulated from the private-sector spending cycle. That is the mechanism by which a state anchor supports a premium multiple: it lowers the probability of a catastrophic revenue shortfall, which in turn justifies a lower discount rate on future cash flows.
The Counter-Thesis - and the Signal That Would Break It
The strongest case against skepticism is simple: growth this fast justifies almost any multiple, because today's revenue is a poor proxy for the business this capital will build. Cohere's revenue has more than doubled year over year, margins are expanding, and its customer base spans finance, healthcare, telecoms, energy and the public sector, with partnerships including Oracle, Dell and RBC. If annual recurring revenue reaches $1 billion within three years - the pace implied by the company's own trajectory - a $20 billion entry price looks cheap, not expensive. The Canadian government's participation is not a subsidy; it is a validation of that path by the most conservative investor in the room.
This argument is coherent, but it rests on one assumption: that the growth rate holds. And growth at this scale is the hardest thing to sustain. The falsifying signal is specific. If Cohere's quarter-over-quarter revenue growth slows below roughly 20 percent for two consecutive quarters, or if the round ultimately closes below $15 billion or with less than $1.5 billion raised, the premium thesis is broken. At that point the multiple would be repricing from mania toward fundamentals, and the sovereign floor alone would not support $20 billion.
Gomez, a co-author of the 2017 "Attention Is All You Need" paper that launched the modern AI era, has consistently framed the company's mandate in measured terms. The next round will test whether that mandate can be executed at a price the public markets will accept.
What to Watch: Three Horizons, Three Scenarios
The forward look splits cleanly by horizon.
In the short term, the deal itself is the risk. Timing could slip, the size could shrink, or the terms could include protections - liquidation preferences, ratchets - that blunt the headline valuation. A $20 billion paper mark that closes at $16 billion with investor-friendly terms is a different story from the headline. Watch for the final size, the lead investor, and whether the Canadian government's contribution comes as equity, a grant, or compute credits - each carries a different signal about how much of the valuation is market-clearing versus policy-supported.
Over the medium term, the question is revenue, not valuation. Cohere needs to demonstrate that the $240 million run rate can scale toward $1 billion without margin compression. The watch items are quarterly ARR growth, net revenue retention, and whether the North platform converts pilots into multi-year contracts. If growth holds above 50 percent annually while margins stay near 70 percent, the multiple will look defensible in retrospect.
Over the long term, the structural thesis stands regardless of the entry price. Sovereign AI is a decade-long reallocation of enterprise and government spending, and Cohere's positioning - private deployment, data residency, cloud-agnostic inference - aligns with it. The company that survives the funding winter with a government anchor and a capital-efficient model will still be standing when the cycle turns.
Three scenarios frame the outcome. The base case: the round closes near the top of the range, at roughly $18 billion to $20 billion, with Canadian government participation, and Cohere uses the capital to lock in sovereign contracts ahead of a tighter 2027 funding market. The upside case: ARR compounds past $1 billion by 2028, the multiple compresses but the equity still appreciates, and an initial public offering follows at a higher absolute valuation. The downside case: the round shrinks below $1.5 billion or closes under $15 billion, growth decelerates, and Cohere joins the growing list of AI companies that raised at a peak mark and reset before listing.
The judgment condensed to one line: this is not a bet on whether AI wins. It is a bet on whether sovereign AI wins fast enough to justify a multiple that assumes it already has.
Explore more exclusive insights at nextfin.ai.
