There's a sentence in Anthropic's IPO filing, published on Monday, that every European founder building on a model should read twice. The company says it builds "the foundation upon which other companies and software developers can create products and services", and then that, "as we continue to develop our own first-party applications, we may from time to time compete with enterprise software and application companies in select categories."
Europe spent the week arguing about exactly that.
On 15 October Mistral's Arthur Mensch published a LinkedIn essay titled "Build or surrender", calling the idea of leaving frontier models to the US and China "a fundamental mistake", TNW reports. Whoever owns a closed model, he argued, sets the price, the uses allowed and which versions stay available, and Europe already pays more than €260 billion a year to American cloud and software providers. On 20 October he added that apps built on foreign models would leave Europe with "leftovers".
The other camp has a CEO with a number. Legora's Max Junestrand told Politico that efforts to build a "competitive frontier lab right now feel like very wishful thinking". On 21 October his company raised $550 million at an $8.8 billion post-money valuation, led by Atreides. Its ARR is past $200 million and the US is now its largest market by revenue. Its CFO, David Eckstein, says durable revenue takes "healthy gross margins", which is the number the next paragraph is about.
Here's the mechanism as I see it. An application company's gross margin is what's left after it pays for the intelligence underneath, so it lives with two things it doesn't control, the supplier's price list and its roadmap. The price list has been moving the app layer's way: Anthropic's filing says "lower prices and improved capability that unlocked new use cases" helped its growth speed up early this year, and first-half revenue reached $16.3 billion, up from $1.2 billion a year earlier. The roadmap is the sentence above. Call it the margin test: for any European AI company, ask who prices its main input, and whether that supplier has said it may compete.
Europe's own money hasn't picked a side. The EU-backed Scaleup Europe Fund is in Mistral, tech.eu says, and led Pigment's $250 million round last week. And if Anthropic's listing goes ahead, its retail tranche in "certain member states of the European Economic Area" would let some European savers own a piece of the supplier too.
So, for anyone building on someone else's model: which of your categories counts as "select"?
LSEG agreed to buy Octus for $3.535 billion. London's exchange group is paying cash for a New York credit-data company that turns covenants and deal terms into structured, searchable data, which customers reach "via Octus’ platform, an MCP connector or via AI workflows". LSEG says Octus is expected to generate over $300 million of revenue this year at an adjusted EBITDA margin of around 40%. Here the data owner is the supplier, and the AI workflow is just how its data gets delivered.
Daytona became a $1 billion company. Founded by three Croatians and headquartered in New York, it runs its 20-person technical team mainly from Split and Zagreb, and The Recursive counts it as Croatia's third unicorn. It rents isolated computers where AI agents run code and use software. FirstMark led the $60 million Series B. tech.eu lists Anthropic, OpenAI, Nvidia and Cursor among the companies it supplies, so the model layer in this week's lead is also a Croatian startup's customer.
Enact raised a $30 million pre-seed. The Zurich company, founded in March by Manuel Yves Galliker, formerly of 1X and Physical Intelligence, builds robots that sense force and touch through their own actuators instead of relying on cameras. Redalpine, Notion Capital and firstminute capital invested, with Arthur Mensch, Yann LeCun and Hugging Face's Thomas Wolf as angels, tech.eu reports. Redalpine's Harald Nieder talks of building "a data moat". Mistral's founder, in other words, is betting on data a camera can't collect.
Bending Spoons closed its purchase of Miro. The Milan acquirer completed the all-cash deal on 26 October, and at closing some former Miro shareholders put about $295 million of their proceeds into roughly 6.2 million new Bending Spoons shares. Per EconomyUp, the enterprise value is $1.355 billion and Miro brings about $600 million of ARR, which by my arithmetic is about 2.3 times ARR. Bending Spoons' acquisition screen, EconomyUp notes, weighs the risk of disruption by AI explicitly. The low multiple is probably at least partly what that risk costs a seller.
Also this week: in London, Nous raised $17 million for a household-admin agent; in Leeds, adsilico raised £10 million to test heart valves and stents on virtual patients; in Eindhoven, Photon Bridge raised €10 million for AI-cluster lasers; in Madrid, Seaya held a €600 million first close for an EIF-backed growth fund; and in Milan, Rulen raised €1.8 million to automate tax work and the Z_One Fund held a first close above €35 million, against a €60 million target.
Britain's £600 million-plus AI package comes with up to £200 million of lab credits. The Chancellor, John Healey, set it out on 24 October: £165 million for a National AI Biosecurity Initiative, including a £100 million AI Biosecurity Lab and £60 million more for the Sovereign AI Fund. The lab also gets "up to £200 million in AI model credits from OpenAI and Anthropic". AI minister Kanishka Narayan says "We cannot afford to rely entirely on technology built overseas." Run the margin test: the lab's model credits, worth up to twice its £100 million of government money, come from two American labs.
Publishers asked Brussels and London to stop Meta's new terms. The European Publishers Council wrote to Teresa Ribera and Henna Virkkunen on 19 October, Politico reports, asking them to check the terms, which publishers say would let Meta copy their websites to build AI, against EU copyright and competition law. A Commission spokesperson cited EU law, "including competition rules and the Digital Markets Act", and the CMA's Sarah Cardell named Meta a possible candidate for Britain's new tech rules. On 23 October Meta pushed the effective date from 30 October to 20 November, Digiday reports. Writing to the competition boss marks this as a market-power fight, fought in Meta's terms of business.
Parliament's centrist coalition split over AI liability. At the plenary on 22 October, Socialist and Green MEPs called for a new AI Liability Act, after the Commission withdrew its liability proposal last year. The Greens' Kim van Sparrentak: "If a phone charger explodes in your house, we never call it innovative." The center-right's Arba Kokalari, who led the work on the AI omnibus, asked instead for "less bureaucracy, more private investments, better access to data and computing infrastructure". With the center-right holding the simplification line, this is the opening position in the next fight, and there is no bill yet.
Washington tied steel tariffs to the DMA again. The US has renewed a warning that it won't lower tariffs on EU steel unless Brussels softens its enforcement of the Digital Markets Act, two people told Politico. Brussels has fined Apple, Meta and Google a combined €1.59 billion under the DMA since April 2025. The DMA is also the rulebook the Commission's spokesperson pointed to over Meta's terms, so any AI case Brussels brings under it would land on the same table as steel.
UK Budget, 28 October. The Chancellor gave the date on 21 October, and the AI package above is meant to be part of it. The line to read is the Sovereign AI Fund's: whether £60 million more is the whole of it, or the start of money that backs British AI companies in cash rather than credits.
Capgemini's third-quarter revenues, 30 October. Capgemini's numbers come a week after the FT reported that Anthropic and Hg plan an AI consultancy for European companies. Listen for whether AI work grows faster than the rest, and who Capgemini says it competes with.
AI Act enforcement rules, feedback closes 9 November. The Commission's draft implementing regulation sets how it runs proceedings under Articles 75c and 75d of the AI Act: access to the file, procedural safeguards and limitation periods for penalties. It had 9 responses on Monday. These are the defense rights a company gets when the Commission itself runs an AI Act case against it.
Siemens' fourth quarter, 12 November. Siemens reports its fiscal 2026 results in Munich, after Berlin signed a declaration of intent with Nvidia on physical AI on 19 October. The same day, per Nvidia, Siemens said a fully virtual production ramp-up at its Erlangen factory cuts engineering effort by up to 85%. Its industrial numbers are where physical AI either shows up as revenue or stays a keynote.
If you’re building something, reply and tell me what it is, I read everything. If you’re in Turin, Italy, let’s have an espresso!
I write two other newsletters, and they come from the same place this one does: twenty-odd years at agencies and scaleups as a creative and later a CMO, mostly where marketing and new technology collide.
AI-Ready CMO is the daily one for marketing leaders. What changed in AI today, what it means for your team, what to do about it.
AI-Ready Ads is daily too, and short: one example a day of AI used well in advertising.
Both are free. If this one is useful to you, those probably are too.