A house newsletter · daily · researched and written by the Kurzia AI
Thursday 19 November 2026 · Issue 18 · 5 min read
Europe's $81bn year but thinner seed market; Google's Neon cuts per-task costs; six rounds incl. Payhawk's $280M Series C
The main story
Tech.eu · yesterday
Europe's banner year hides a thinner seed market
Takeaway: down almost 30 per cent
Atomico's twelfth State of European Tech report, published Wednesday, says European startups will draw about $81bn of investment this year — the second-best year on record — and that a record 40,000 founders started companies, a 36 per cent rise. The warning for angel-sized cheques sits underneath: money is pooling into a few giant AI rounds, and Atomico counts pre-Series A 'first-money-in' deals down almost 30 per cent between 2021 and 2025. The report also sizes Europe's whole tech ecosystem at $4.4 trillion, up 18 per cent, and flags that the continent holds just 7 per cent of global computing capacity, against 72 per cent in the US. For your next cheque: fewer rivals at seed, but check where a company will get compute. For companies you back: ask how they train and serve customers without renting US capacity.
New rounds
Perry AI — London legal-AI that remembers a private capital fund's positions across deals — seed, $2.2M, led by Haatch; reported by Tech.eu.
MountAIn — Grenoble frugal-AI startup that runs advanced models on microcontrollers under €9 — seed, €7.3M, led by 360 Capital and Supernova Invest; reported by EU-Startups.
Vero Security — Polices the AI tools and agents employees use, already at several Fortune 500 firms — seed, $11M, led by TLV Partners (company release).
CloudBackend — Swedish AI cloud that takes vibe-coded prototypes to production — stage not stated, €6.6M, led by Carl-Henric Svanberg's family office (company release).
Humyn Labs — Bengaluru startup recording real-world training data for US robotics — Series A, $47M: $26M equity led by Accel, $21M debt by Lighthouse Canton; reported by TechCrunch.
Payhawk — London-based company spend management — Series C, $280M led by Goldman Sachs, valuing it at $2bn, reported by Tech.eu.
What the AI labs did
Gemini 4 Neon — Google's new Gemini 4 family model, out November 18, built for long coding and analysis runs: #3 on Artificial Analysis's Intelligence Index at 61, #1 on its Coding Agent Index. Priced by Google at $3 per million input and $15 output tokens, with a 1M-token context and 32% lower cost per task than Carbon — pressure on generalist coding-agent startups, relief for token-heavy apps. Founders: re-benchmark your agent stack on Neon before your next model commitment. Google
GPT-6.2 Astra and Sol — OpenAI shipped both on November 17, per Latent Space: Astra scores 62 on the Artificial Analysis Intelligence Index, #2 behind Anthropic's Fable 5.2. Per Latent Space, Astra's API costs $8 in, $40 out per million tokens; Sol $2 and $10, matching September's Astra at about 15% of the cost per task — a margin cut for anything built on OpenAI. Early testers split hard on quality, so founders should run their own evals on Sol before switching traffic. Latent Space
For your portfolio
European Business Magazine · yesterday
Project Exponential
Takeaway: more than 100 AI agents
KPMG UK ran a trial, reported by the FT and retold by European Business Magazine on November 18, in which more than 100 AI agents audited an unnamed FTSE 250 company alongside the human team.
This is dated, named proof that agents now do the junior-analyst work inside the Big Four, and KPMG plans to extend it to as many as 20 larger audits — though one agent mis-matched regulatory requirements to company documents and needed a human to step in, and the results were not used to sign off the accounts. Ask founders selling into finance: when the buyer's own agents do the grunt work, which part of your product still earns the fee — and does your product fit the agents' workflow or sit outside it?
Read it at European Business Magazine
“identifying areas requiring further investigation and asking questions that human auditors had not necessarily raised”
TechCrunch · today
Nadella calls much critical infrastructure unpatchable
Speaking at the New York Times DealBook Summit on Wednesday, reported by TechCrunch, Microsoft's CEO said firmware, cooling and building systems across US critical infrastructure sit on the internet with flaws that cannot be patched — and that AI has pushed patching to a pace that itself destabilises systems, since fixes must reach production instantly.
That admission from the biggest software vendor is a budget signal: enterprise security buyers are being told to find and defend what they cannot fix, exactly the wedge Vero Security raised $11M for today. Ask founders in security: does the product assume the customer can patch — or does it watch and contain what can't be fixed?
“Remember, there is a lot of stuff in our critical infrastructure that's unpatchable”