Cohere and Aleph Alpha Signed the Papers on a $20B Merger

On September 16, Cohere and Aleph Alpha signed a definitive business combination agreement. The merged company will operate globally as Cohere, valued at roughly $20 billion, dual-headquartered in Toronto and Berlin, with Aleph Alpha's Heidelberg office retained as a research center and combined headcount passing 1,000.

The plan was first announced in April. What changed on Tuesday is that it stopped being an intention and became a contract, subject to regulatory approval, targeted to close later in 2026.

The numbers behind the headline

Cohere was valued at $7 billion in September 2025 and reported $240 million in 2025 revenue. Aleph Alpha had raised roughly $126 million across its life. Those two figures do not add up to $20 billion on their own, which tells you the valuation is being set by what the combined entity is expected to sell, not by what either company has already built.

The capital behind that expectation is specific. Germany's Schwarz Group — the retail conglomerate that owns Lidl and Kaufland, and one of Europe's largest private companies — is committing €500 million (roughly $573 million) to Cohere's upcoming Series E. Schwarz has separately signalled a €11 billion data center program targeting up to 100,000 GPUs. Reports also point to a potential Canadian government-backed consortium investment of up to $3 billion.

Leadership is split across the two sides. Aidan Gomez remains CEO. Aleph Alpha co-CEO Ilhan Scheer takes the chief operating officer role, and co-founder Samuel Weinbach moves into chief research officer.

Why a retail conglomerate is the anchor investor

Schwarz Group is not a venture fund taking a shot at the frontier. It is a customer that decided to own part of its supplier.

That is the entire logic of the deal. Cohere never competed to make the smartest model — it competed to be the model enterprises could deploy inside their own perimeter, on their own infrastructure, without their data leaving the building. Aleph Alpha built the same thesis for German and EU public-sector buyers, who have regulatory reasons to care where inference physically runs.

European enterprises and governments are in an awkward position. The capable models are American or Chinese. The rules governing what those enterprises can do with customer data are European. For a decade, the resolution was to accept the gap and negotiate data residency addenda. The Cohere-Aleph Alpha merger is a bet that a meaningful number of buyers would rather pay for a lab whose corporate domicile and compute footprint they can point to on a map.

"Transatlantic sovereign AI" is the phrase both companies are using, and it is a strange one — sovereignty that spans two continents is a contradiction if you take the word literally. What it means in practice is a supplier that is neither in Silicon Valley nor in Hangzhou, run out of jurisdictions where Canadian and EU procurement rules apply.

The gap this doesn't close

Nothing in this agreement changes the capability picture at the top of the market.

Mozilla's State of Open Source AI report, published the same week, put the open-to-closed capability gap at about 4.4 months on METR task-horizon data. That is the distance between the frontier and freely downloadable weights. A merged Cohere is not competing at that frontier — it is competing at a layer below it, where models are good enough and the differentiator is deployment, compliance, and support.

That layer is real and it is large. It is also the layer most exposed to open weights. If the strongest open model sits a few points behind the closed leader at a fraction of the price, and eight of the top ten models by tokens on OpenRouter in August were open-weight — seven of them Chinese — then the question a European bank asks is not "American or European lab?" It is "why am I paying a lab at all when I can run GLM or DeepSeek on my own hardware and hire an integrator?"

The merged company's answer has to be that it sells something an integrator can't: accountability, a roadmap, indemnity, and a named counterparty that a regulator will accept. That is a services business wearing a frontier lab's valuation.

What the deal actually buys

Three things worth naming.

Scale on the sales side. Over 1,000 employees across North America and Europe, with existing enterprise relationships on both continents, is a materially larger go-to-market surface than either company had alone. In a market where the technical differences between adequate models are shrinking, distribution is the asset.

A domestic compute story. Schwarz's data center program gives Cohere something most challenger labs lack — a path to serving European customers on European infrastructure without renting it from a US hyperscaler. Whether 100,000 GPUs materializes on the stated timeline is a separate question, but the commitment exists and it is denominated in euros.

Regulatory positioning ahead of a tightening cycle. Ursula von der Leyen spent this week previewing a Digital Fairness Act and warning that frontier models could enable attacks "we never thought possible." A European-headquartered lab with a Canadian arm is well placed if EU procurement starts to weight domicile.

What to watch

Whether the Series E closes at the stated size. Schwarz's €500 million is committed. The Canadian consortium figure is reported, not signed. The gap between the two determines whether the $20 billion mark has capital behind it or is a paper number attached to a merger ratio.

Regulatory clearance in Berlin and Ottawa. A transaction that moves German research assets into a Canadian-domiciled parent will draw scrutiny under exactly the sovereignty logic the deal is selling.

The first large public-sector win. Sovereign AI as a category has produced more announcements than contracts. A named EU government agency running production workloads on Cohere infrastructure would convert the thesis into revenue. Until then, this is a $20 billion bet that where a lab is incorporated will matter more than how good its model is.