Newsletter
Microsoft and Cloudflare both fell because of ordinary internal changes. Radu Georgescu on why the volume of change entering production is the real exposure.
November 16, 2026
·
5
min read

The European private markets briefing you can't afford to miss. Deals, signals, and moves, decoded every Monday by SeedBlink.

On Thursday, I could not send a work email from the United States. The problem was not the distance.
Outlook, Teams and Copilot were down for more than a day, with reports from more than 70 countries. By Microsoft’s account it began with a change to its own network, after which servers started failing their automated health checks and dropping out of service. Not an attack, and the company was explicit that AI had nothing to do with it. The full report has yet to appear.
Last November it was Cloudflare, for just under six hours, because a change to database permissions left one configuration file twice the size the software reading it could handle. Their worst outage since 2019.Neither was exotic. Both were ordinary changes to systems too complex for any one team to fully model. And there was no European version that stayed up, and no American one either. We were all out together, which is worth remembering before anyone draws conclusions about whose cloud it ought to be.
My own conclusion is about volume. The amount of change going into production is rising. Ten years ago, the thing we all worried about was a person making a mistake. We now ship code faster than any team can read it, and we are starting to let software act on its own inside live systems. The surface where things can go wrong grows with the speed.
Where I would be looking right now is the least glamorous layer there is: mission-critical infrastructure, and security for agentic systems. The tools that let an organisation run something it cannot fully inspect, in businesses where somebody will eventually be asked who approved what, what the system changed, and how it was stopped. It is probably the most interesting place to be in technology at the moment.
I believe I would rather own the brakes than the engine.


Outages pass. The exposure they reveal does not, and it is measurable.
This autumn, UK financial firms reported 169 cloud and compute relationships to the Bank of England and the FCA, in a survey of 196 institutions. Of those relationships, 37% were described as both highly critical and hard to substitute. Only 35% of that group had an exit plan that had been tested.
The knowledge gap sits beside it. Where a high-materiality system comes from a third party, 81% of firms rely on the vendor’s assurances and the contract, while 24% say they have deep knowledge of the system themselves.
The direction of travel is the part that matters for anything being built now. Third-party systems account for 39% of the AI systems these firms reported, but 53% of those they classify as high-materiality. The systems classified as most material are increasingly the ones firms rely on but understand least.
Supervisors are raising related concerns from their side of the table. Pedro Machado of the ECB told a conference in Frankfurt in February that governance is “uneven across institutions and use cases”, with fragmented ownership, unclear accountability and insufficient challenge from risk, compliance and internal audit. AI, he said, “does not dilute responsibility. If anything, it raises the bar.”
Put the two together and the shape of a market appears. Institutions have adopted faster than they have built the ability to answer for what they adopted, and the regulated ones now have supervisors asking. As scrutiny increases, the opportunity may lie less in adding new capabilities and more in providing evidence of control: inventories, materiality classifications, ongoing monitoring, and exit plans that have been tested rather than simply documented.
For an investor, that turns a vague concern into a concrete question: does a company know which of its suppliers are critical, and when did it last test its ability to operate without one? For many organisations, that test remains unperformed, and running it costs money.

Radu Georgescu joins the board of FlowX.AI
Radu Georgescu, SeedBlink’s Chairman of the Board, has taken on an additional role as a board member at FlowX.AI, a SeedBlink alumnus.
Founded in 2020 and now a team of more than 150, FlowX.AI has raised over $43 million for a multi-agent platform used by banks, insurers, logistics operators and governments, with a catalogue of more than 220 prebuilt enterprise-grade AI agents.
Radu has long brought his entrepreneurial perspective to SeedBlink, working alongside founders and investors to support the growth of Europe’s tech ecosystem. He will now bring that experience to FlowX.AI as the company continues its international expansion.
“FlowX.AI has the vision, product, the team and the customers to become a leader in enterprise AI, and I’m glad to help,” Radu said of the appointment.

Ukraine 🇺🇦
UFORCE raises $700M Series A at $6B valuation
Ukraine-founded and London-headquartered UFORCE raised $700 million at a $6 billion valuation to scale production of uncrewed sea, land and air platforms and the software that coordinates them, in a Series A co-led by Valor Equity Partners and the UK’s National Wealth Fund, with the US-Ukraine Reconstruction Investment Fund also participating.
France 🇫🇷
Orakl Oncology raises $34M Series A
Paris-based Orakl Oncology, spun out of the Gustave Roussy cancer centre in 2023, raised $34 million co-led by Norrsken VC and Singular to grow lab copies of patients’ tumours, test drugs on them, and train models that predict how those drugs will perform in trials. In July it won the rights to test four of Sanofi’s cancer drugs.
Italy 🇮🇹
Nebuly raises $21M Series A
Turin-based Nebuly raised $21 million led by Planven, with Exor Ventures and CDP Venture Capital among the backers, to measure what a company’s AI agents are actually worth, tracking time saved and revenue influenced rather than prompts and users. Customers include Oura, Iveco Group and Zeiss.

What Jensen Huang’s resignation means for startups
Nvidia’s chief executive resigned last week and the shares have since lost more than $855 billion in market value. Tunguz sets out why the shock travels further than one company: Nvidia has $61 billion of equity at cost across 229 private startups and 13 public companies, carries $164.5 billion in lease guarantees and cloud capacity backstops, and last quarter put 112% of its operating cash flow into equity stakes and acquisitions.
Anthropic puts Claude into meetings, whiteboards and spreadsheets
Claude now joins video calls as a named participant that drafts the plan under discussion, alongside new shared whiteboards and spreadsheets. Worth reading less for the features than for the direction of travel: the AI labs are now building office software and competing for it with Microsoft and Google.
Serial unicorn founders from Europe
Dealroom has pulled together 27 founders of European birth or nationality who started two or more companies that reached a billion dollars, from Nord Security and Oxylabs in Lithuania to Ganymed and BioNTech in Germany. Nineteen built both companies in Europe, five built both abroad. Dealroom calls it a selection rather than a ranking, and counts any company recorded at a billion at any point, including a few that fell back afterwards.
Written by

Denisa Lacatus
Communication and Content Specialist
TABLE OF CONTENT
Share this article