Editor's note
Today's thread is concentration — of capital and of compute. Mistral raised €3B in Europe's largest-ever tech round, while China laid out a national plan to grow AI compute roughly sixfold by 2030. Google, meanwhile, admits its EU search revamp lowers quality even as its generative UI pulls single-purpose utility traffic inside the results page. With a proof-of-concept AI-built WeChat worm rounding out the day, the ground is shifting under capital, search and security alike.
AITechCrunch
Mistral Raises €3B Series D — Samsung-Led, Valuation Above €21B
French AI firm Mistral said on September 8 it raised €3B in a Series D at a post-money valuation above €21B, the largest equity round ever by a European tech company. Samsung Electronics led, with EQT-managed Scaleup Europe Fund and PSG Equity as co-leads and BlackRock-managed funds among new investors.
Context
Its prior round was a €1.7B Series C at €11.7B in September 2025, led by ASML — so the valuation roughly doubled in a year. The proceeds go to frontier research, scaling compute for training, and commercial expansion. It underscores both Europe's push for sovereign AI and the concentration of capital in the sector.
HaLVision's take
With US firms dominating, a well-funded independent European player widens vendor choice — an indirect plus for SMBs. For firms that weigh data sovereignty and GDPR, a Europe-built model becomes a real option to evaluate. Still, a valuation this steep reflects frothy AI investment; judge it on shipping products, pricing and availability, not headlines.
Read source ↗AISouth China Morning Post
China Targets 9,800 EFLOPS of AI Compute by 2030 in New Five-Year Plan
China's MIIT published its 15th Five-Year Plan for the information and communications industry, targeting 9,800 EFLOPS of intelligent computing by 2030 — up from 1,590 EFLOPS in 2025. It anticipates cumulative information-infrastructure investment of about 3.8 trillion yuan.
Context
An EFLOPS is a quintillion floating-point operations per second. The plan calls for GPU clusters in the 10,000-to-100,000-card range and inference facilities tuned to specific uses. It formalizes a national strategy to keep building domestic compute despite US export controls.
HaLVision's take
Compute demand ripples through global GPU and memory supply, affecting cloud pricing and lead times everywhere. As national AI-infrastructure gaps widen, SMBs should avoid over-relying on one region — spreading across multiple clouds and mixing in lighter models hedges both cost and supply risk.
Read source ↗WebSearch Engine Land
Google Overhauls EU Search — Calls It the Biggest Quality Drop in 29 Years
On September 8 Google reorganized its European search results to comply with the Digital Markets Act, trimming real-time prices from hotel, airline and restaurant carousels and demoting specialized results. Google itself described the change as the largest reduction in quality of service in Search's 29-year history.
Context
The move follows a €460M self-preferencing fine in July. Google claims earlier DMA remedies already cut European businesses' free direct-booking traffic by about 30%. It is a concrete case of the trade-off between regulation and search experience surfacing as an actual UI change.
HaLVision's take
Japan is out of scope, but a search surface that narrows the path to direct booking and comparison is instructive for any business that relies on search for reservations or foot traffic. It is another reminder to reduce platform dependence and invest in owned channels — your own site, Google Business Profile and email.
Read source ↗WebSearch Engine Roundtable
Google Expands Interactive Tools in AI Overviews — Counters and Calculators on the Fly
Google is bringing the generative UI it built for AI Mode into AI Overviews, assembling interactive tools — such as calculators and a character counter — directly inside results. The shift absorbs demand for simple single-purpose utility sites into the search page itself.
Context
Generative UI dynamically builds layouts and small tools within an answer; examples shown include a character counter, a pH scale and a mortgage calculator. Traffic that single-function sites once captured may increasingly stay inside the search surface.
HaLVision's take
Sites that depend on the search-then-visit-a-utility path could see clicks erode as AI Overviews resolve the task in place. Tool pages need to differentiate on things the results page cannot easily replicate — proprietary data, expert context, saving and integrations. SEO now has to be rebuilt around the assumption that the answer ends on the results page.
Read source ↗AICalif
Researchers Demo an AI-Built Zero-Click WeChat Worm That Spreads via Calls
Security firm Calif used AI to find a serious WeChat vulnerability and build WeWorm, a proof-of-concept worm that infects with no user interaction, spreads over calls across iOS and Android, and propagates through saved contacts. Tencent patched it after responsible disclosure and says it has no evidence users were affected.
Context
The AI found the bug in July and a working cross-platform demo was finished in mid-August, with disclosure and coverage on September 8. The key point is that AI accelerated the path from discovery to weaponization, with the tool built in a little over a week — a concrete sign that attacker costs are falling.
HaLVision's take
This is a demonstration, not a live breach, but it sharpens how AI can speed up attacks. Basics — prompt patching, multi-factor authentication and least privilege — become even more cost-effective defenses for SMBs. Treat a vendor app's vulnerability as something that can reach you, and revisit your patch cadence now.
Read source ↗