Economic

Q3 funding in Europe rebounds with growth rounds leading

Europe raised €13.7bn across just over 1,300 rounds in Q3, the strongest quarter since Q2 2024. September alone brought €8.7bn. July and August reflected the familiar summer slowdown.

Growth equity provided €7bn, or 51.6% of the total, with two consecutive quarters surpassing 150 growth rounds. Data centres, AI agents, and GenAI led the activity, with more AI startups scaling with larger cheques.

Early-stage totals were the lowest in 12 months, yet they were ahead of Q3 last year. Lovable’s $200 million Series A at a $1.8 billion valuation stood out. Seven new unicorns included Nscale, Fuse Energy, Framer, IQM, Nothing, and Tide.

ASML led the quarter’s largest deal, investing €1.3bn in Mistral AI’s €1.7bn Series C. France tallied €2.7 billion, heavily concentrated in Mistral, while the UK reached €4.49 billion. Germany followed with just over €1.5bn, ahead of the Netherlands and Switzerland.

AI-native funding surpassed all verticals for the first time on record, reaching €3.9 billion, with deeptech at €2.6 billion. Agentic AI logged 129 rounds, sharply higher year-over-year, while data centres edged out agents for capital. Defence and dual-use technology attracted €2.1 billion across 44 rounds.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!

Unexpected language emerges as best for AI prompting

A new joint study by the University of Maryland and Microsoft has found that Polish is the most effective language for prompting AI, outperforming 25 others, including English, French, and Chinese.

The researchers tested leading AI models, including OpenAI, Google Gemini, Qwen, Llama, and DeepSeek, by providing identical prompts in 26 languages. Polish achieved an average accuracy of 88 percent, securing first place. English, often seen as the natural choice for AI interaction, came only sixth.

According to the study, Polish proved to be the most precise in issuing commands to AI, despite the fact that far less Polish-language data exists for model training. The Polish Patent Office noted that while humans find the language difficult, AI systems appear to handle it with remarkable ease.

Other high-performing languages included French, Italian, and Spanish, with Chinese ranking among the lowest. The finding challenges the widespread assumption that English dominates AI communication and could reshape future research on multilingual model optimisation.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Perplexity launches AI-powered patent search to make innovation intelligence accessible

The US software company, Perplexity, has unveiled Perplexity Patents, the first AI-powered patent research agent designed to democratise access to intellectual property intelligence. The new tool allows anyone to explore patents using natural language instead of complex keyword syntax.

Traditional patent research has long relied on rigid search systems that demand specialist knowledge and expensive software.

Perplexity Patents instead offers conversational interaction, enabling users to ask questions such as ‘Are there any patents on AI for language learning?’ or ‘Key quantum computing patents since 2024?’.

The system automatically identifies relevant patents, provides inline viewing, and maintains context across multiple questions.

Powered by Perplexity’s large-scale search infrastructure, the platform uses agentic reasoning to break down complex queries, perform multi-step searches, and return comprehensive results supported by extensive patent documentation.

Its semantic understanding also captures related concepts that traditional tools often miss, linking terms such as ‘fitness trackers’, ‘activity bands’, and ‘health monitoring wearables’.

Beyond patent databases, Perplexity Patents can also draw from academic papers, open-source code, and other publicly available data, revealing the entire landscape of technological innovation. The service launches today in beta, free for all users, with extra features for Pro and Max subscribers.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Nexperia chip exports may resume as China softens stance on ban

China’s Ministry of Commerce announced plans to exempt specific Nexperia orders from its export ban, aiming to stabilise the global semiconductor supply chain after the Netherlands seized control of the Chinese-owned Dutch chipmaker.

The ministry stated that exemptions would be granted when the criteria were met, encouraging affected firms to apply.

A move that follows a meeting between Chinese President Xi Jinping and US President Donald Trump in Busan, where both sides reached a framework allowing Nexperia to resume shipments under eased restrictions.

Washington reportedly agreed to pause the 50 percent subsidiary rule, which restricts exports from companies half-owned by entities on its trade blocklist. Wingtech Technology, Nexperia’s Chinese parent, has been under these restrictions since December.

Beijing’s export ban, introduced after the Dutch takeover citing national security concerns, disrupted supplies from Nexperia’s Dongguan factory, which assembles about 70 percent of its products.

China condemned the Netherlands for intervening in corporate affairs, warning that such actions deepen global supply chain instability.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

A new phase for Hyundai and NVIDIA in AI mobility and manufacturing

NVIDIA and Hyundai Motor Group will build a Blackwell-powered AI factory for autonomous vehicles, smart plants and robotics. The partners will co-develop core physical AI, shifting from tool adoption to capability building across mobility, manufacturing and on-device chips.

The programme targets integrated training, validation and deployment on 50,000 Blackwell GPUs. In parallel, both sides will back the physical AI cluster in South Korea with about $3 billion, creating an NVIDIA AI Technology Center, Hyundai’s Physical AI Application Center and regional data centres.

Hyundai will use NVIDIA DGX for model training, Omniverse and Cosmos on RTX PRO Servers for digital twins and simulation, and DRIVE AGX Thor in vehicles and robots for real-time intelligence. The stack underpins design, testing and deployment at an industrial scale.

Factory digital twins will unify data, enable virtual commissioning and improve predictive maintenance, supporting safer human-robot work. Isaac Sim will validate tasks and ergonomics before line deployment, speeding robot integration and lifting throughput, quality and uptime.

Vehicles will gain evolving features via Nemotron and NeMo, from autonomy to personalised assistants and adaptive comfort. DRIVE AGX Thor with safety-certified DriveOS will power driver assistance and next-generation safety, linking car and factory into one intelligent ecosystem.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!

Australian influencer family moves to UK over child social media ban

An Australian influencer family known as the Empire Family is relocating to the UK to avoid Australia’s new social media ban for under-16s, which begins in December. The law requires major platforms to take steps preventing underage users from creating or maintaining accounts.

The family, comprising mothers Beck and Bec Lea, their 17-year-old son Prezley and 14-year-old daughter Charlotte, said the move will allow Charlotte to continue creating online content. She has hundreds of thousands of followers across YouTube, TikTok and Instagram, with all accounts managed by her parents.

Beck said they support the government’s intent to protect young people from harm but are concerned about the uncertainty surrounding enforcement methods, such as ID checks or facial recognition. She said the family wanted stability while the system is clarified.

The Australia ban, described as a world first, will apply to Facebook, Instagram, TikTok, X and YouTube. Non-compliant firms could face fines of up to A$50 million, while observers say the rules raise new privacy and data protection concerns.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot! 

An AI factory brings Nvidia compute into Samsung’s fabs

Nvidia and Samsung outlined a semiconductor AI factory that embeds accelerated computing into production. Over 50,000 GPUs will drive digital twins, predictive maintenance, and real-time optimisation. Partners present the project as a template for autonomous fabs.

The alliance spans design and manufacturing. Samsung uses CUDA-X and EDA tools to speed simulation and verification. Integrating cuLitho into OPC reports roughly twentyfold gains in computational lithography at advanced nodes.

Factory planning and logistics run on Omniverse digital twins and RTX PRO servers. Unified analytics support anomaly detection, capacity planning, and flow balancing. Managers expect shorter ramps and smoother changeovers with higher equipment effectiveness.

Robotics and edge AI extend intelligence to the line. Isaac Sim, Cosmos models, and Jetson Thor target safe collaboration, faster task retargeting, and teleoperation. Samsung’s in-house models enable multilingual assistance and on-site decision support.

A decades-long Nvidia–Samsung relationship underpins the effort, from NV1 DRAM to HBM3E and HBM4. Work continues on memory, modules, and foundry services, plus AI-RAN research with networks in South Korea and academia linking factory intelligence with next-gen connectivity.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!

Google removes Gemma AI model following defamation claims

Google has removed its Gemma AI model from AI Studio after US Senator Marsha Blackburn accused it of producing false sexual misconduct claims about her. The senator said Gemma fabricated an incident allegedly from her 1987 campaign, citing nonexistent news links to support the claim.

Blackburn described the AI’s response as defamatory and demanded action from Google.

The controversy follows a similar case involving conservative activist Robby Starbuck, who claims Google’s AI tools made false accusations about him. Google acknowledged that AI’ hallucinations’ are a known issue but insisted it is working to mitigate such errors.

Blackburn argued these fabrications go beyond harmless mistakes and represent real defamation from a company-owned AI model.

Google stated that Gemma was never intended as a consumer-facing tool, noting that some non-developers misused it to ask factual questions. The company confirmed it would remove the model from AI Studio while keeping it accessible via API for developers.

The incident has reignited debates over AI bias and accountability. Blackburn highlighted what she sees as a consistent pattern of conservative figures being targeted by AI systems, amid wider political scrutiny over misinformation and AI regulation.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot! 

Stargate Michigan expands OpenAI’s US buildout

OpenAI will build a new campus in Saline Township, Michigan, as part of a 4.5 GW partnership with Oracle. Planned US capacity now exceeds 8 gigawatts. Investment over the next three years is expected to surpass $450 billion.

Leaders frame Stargate as a path to reindustrialise the United States while expanding access to AI benefits. Projects generate jobs during buildout and strengthen supply chains. Communities are intended to share gains.

Related Digital will develop the Michigan site, with construction expected in early 2026. More than 2,500 union construction roles are planned. A closed-loop cooling system will significantly reduce on-site water consumption.

DTE Energy will utilise existing excess transmission capacity to serve the campus. The project, not local ratepayers, will fund any required upgrades. Local energy supplies are expected to remain unaffected.

Expansion builds on previously announced sites in Texas, New Mexico, Wisconsin, and Ohio. Programmes aim to bolster modern energy and manufacturing systems. Michigan’s engineering heritage makes it a focal point for future AI infrastructure.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!

When AI LLMs ‘think’ more, groups suffer, CMU study finds

Researchers at Carnegie Mellon University report that stronger-reasoning language models (LLMs) act more selfishly in groups, reducing cooperation and nudging peers toward self-interest. Concerns grow as people ask AI for social advice.

In a Public Goods test, non-reasoning models shared 96 percent; a reasoning model shared 20 percent. Adding a few reasoning steps cut cooperation nearly in half. Reflection prompts also reduced sharing.

Mixed groups showed spillover. Reasoning agents dragged down collective performance by 81 percent, spreading self-interest. Users may over-trust ‘rational’ advice that justifies uncooperative choices at work or in class.

Comparisons spanned LLMs from OpenAI, Google, DeepSeek, and Anthropic. Findings point to the need to balance raw reasoning with social intelligence. Designers should reward cooperation, not only optimise individual gain.

The paper ‘Spontaneous Giving and Calculated Greed in Language Models’ will be presented at EMNLP 2025, with a preprint on arXiv. Authors caution that more intelligent AI is not automatically better for society.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!