Alibaba Cloud Puts $3M Behind Omarchy as AI Startup Funding Keeps Flowing

The deal is $1 million a year for three years, according to the Omarchy project's own announcement on 28 September. That brings total backing for the foundation to roughly $21.7 million. Alibaba Cloud joins DigitalOcean and Meta Superintelligence Labs as founding corporate patrons.

The money is only part of it. Alibaba Cloud will help build Omarchy China, covering a local CDN, hosting, a website, meetups and community support. The two sides will also tune Omarchy to work with Alibaba's Qwen models out of the box. The announcement ties the collaboration to Qwen Book, which Alibaba Cloud describes as aiming to create a native AI agent computer, with an "OS as Harness" philosophy that integrates models, operating systems, applications, cloud services and hardware. The first Qwen Book release is based on ARM, with x86 compatibility also promised.

Corporate patrons writing cheques to open-source Linux distributions is not the same business as venture funding. But the same week produced a steady trickle of early-stage AI rounds in Europe, and the patterns are worth reading together.

Small cheques, specific targets

On 24 September, Tech.eu reported that Reply Next, a Spanish startup founded by telecommunications engineer Miguel Peñate, raised €400,000 in pre-seed funding. The round was led by 4Founders Capital, with business angels including Juanjo Mostazo and David Alcobero taking part. Reply Next builds an AI platform for brands with hundreds or thousands of physical locations. It pulls reviews, ratings, messages, searches and recommendations from Google Maps, Booking, the App Store, Google Play and AI-powered search engines into one place. The company claims nine out of ten companies do not know how much revenue their stores generate through Google Maps, 70% of brands do not respond to users on public channels and fewer than 10% of businesses appear in AI-powered search results. The funding goes to technology and sales hires, product development and new AI agents.

Also on 24 September, Tech.eu reported that Eindhoven-based OWOW Venture Studio secured €2.65 million from eight family offices and angel investors. OWOW backs early-stage B2B startups in manufacturing, OEMs and construction, with investments of €100,000 to €250,000 per startup and plans to add at least three ventures a year for three years. It initially targeted €2 million before raising the figure on investor interest. Its first two investments are Synthgen, which generates synthetic data for training AI models used in robots and machines, and Caddy Clubhouse, a British platform connecting golfers with caddies. Robin Dohmen leads the studio.

Those are modest sums next to the headline numbers coming out of the United States. They are also much more concrete about what the money buys.

Agents that write code, and agents that check it

The more interesting cluster this week sits one layer up: tooling for teams already running AI coding agents. Canary, a Y Combinator-backed startup, launched on 24 September with a pitch aimed squarely at that problem. Its site says Canary is the independent tester for code your agents write: it runs the app, tries to break every change and reports what would have reached production. The company says its tool works with Claude Code, Cursor, Codex and other agents, installs via an npm CLI, and files failures back into GitHub, Linear, Sentry, Datadog, Notion and Slack. Its marketing claims bug reports on new code go down, though the site does not publish named customers or audited figures.

A quieter GitHub project takes a different angle on the same coordination problem. Foremerge, posted on Hacker News on 21 September, describes itself as an open-source coordination protocol for coding agents built above Git. Agents announce what they are about to touch, and a small SQLite database inside the project's .git folder lets every agent on the machine see the same picture. If two plans collide, Foremerge names both agents and explains the clash before either writes code. Its README is explicit about limits: version 0.5.0 is a pre-1.0, local-first MVP, schemas may change, published benchmark results do not yet exist, and coordination between machines is out of scope.

There is a sceptical reading of all this. The most detailed AI funding data in the dossier comes from two Tech.eu posts on 24 September, and neither round tops €3 million. The loudest claims about bug reduction come from vendor marketing pages. The most technically specific project, Foremerge, admits it has no benchmarks. The Omarchy announcement, meanwhile, is a foundation blog post, not a filing, and the $21.7 million total is the project's own figure.

None of that makes the money fake. It does mean the useful signal this week is not the size of any single round but where the capital is pointing: localisation for China, digital presence for physical stores, synthetic data for robots, and an entire cottage industry of tools that assume your code is now written by agents you do not fully supervise.