🤖🦾 What’s the Real Opportunity for Computing in Europe?

This week, discussions around DeepSeek’s low or not-so-low training costs continued, even as new models make headlines and the field of reasoning LLMs expands. At the same time, the debate over scaling AI infrastructure is heating up, with questions about the feasibility of building ever-larger data centers and how much big tech can realistically invest in these projects.

Also, we hosted our first Future of Quantum Computing meetup in Berlin, and a huge thank you goes out to everyone who attended, as well as Kipu Quantum and Quantistry for their support. Tomorrow, we’ll announce another major event—stay tuned and follow us on LinkedIn to be the first to hear about it!

Spotlights

🦾 Sustainable SOT-MRAM memory technology could replace cache memory in computer architecture in the future (TechXplore)

Researchers at Johannes Gutenberg University Mainz have developed an advanced Spin-Orbit Torque Magnetic Random-Access Memory (SOT-MRAM) technology that achieves over a 50% reduction in energy consumption, a 30% increase in efficiency, and a 20% decrease in the input current required for magnetic switching while ensuring data retention for more than a decade.

⚛️⚡️ Quandela’s Hybrid Photonic Approach Reduces Resource Requirements for Fault-Tolerant Quantum Computing (Quantum Insider)

By efficiently utilizing semiconductor quantum emitters to generate photonic qubits, this approach decreases the number of components needed for a logical qubit by a factor of 100,000 compared to traditional photonic methods. Quandela's quantum computers are projected to operate below 1 MW, making them more energy-efficient than conventional high-performance computing centers.

🦾 Can We Build a Five Gigawatt Data Center? (Lynette Bye, Asterisk Magazine)

Building a five-gigawatt data center to fuel AI’s insatiable hunger for compute is an engineering and political nightmare. The biggest roadblocks? Power—securing a stable supply without waiting decades for nuclear permits—and sheer infrastructure scale, from cooling to chip availability. Without radical permitting reforms and energy breakthroughs, the dream of a five-gigawatt AI beast by 2030 might stay just that—a dream.

🤖 European AI group set to develop new open LLMs (Techerati)

A consortium of leading European AI organizations, businesses, and universities is collaborating to develop open-source large language models (LLMs) tailored to European standards. This initiative aims to democratize access to high-quality AI technologies, enabling European enterprises to compete globally and public organizations to offer impactful services.

Headlines

🤖 How to Scale Your Model: A Systems View of LLMs on TPUs (GitHub Book by Google DeepMind researchers)

🤖 OpenAI hits back at DeepSeek with o3-mini reasoning model (Ars Technica)

🤖 S1: The $6 R1 Competitor? “The model is just below state of the art, but it can run on my laptop.” (Tim Kellogg’s Blog)

🤖 Understanding Reasoning LLMs: Methods and Strategies for Building and Refining Reasoning Models (Ahead of AI)

🤖 New report by leading semiconductor analyst Dylan Patel shows that DeepSeek spent over $1 billion on its compute cluster (David Sacks on X)

🧠 Neuromorphic weighted sum with magnetic skyrmions: Integrating magnetic skyrmion properties into neuromorphic computing promises advancements in hardware efficiency and computational power (arXiv)

🧠 Novel processor uses magnons to crack complex problems: a smart "universal" device that uses spin waves ("magnons") to perform multiple data processing tasks with exceptional energy efficiency (Phys Org)

⚡️ Accelerating high-performance AI workloads with photonic chips: An overview of how engineers achieve efficiency and scalability with photonic chips (EEWorld Online)

🦾 Fujifilm invests €25 million in semiconductor production in Belgium (The Brussels Times)

🦾 Chips War 2025: US and European industries are complementary — and need to join forces to beat China (CEPA)

🦾 Could Finland become Europe’s semiconductor powerhouse? (Innovation News Network)

🦾 International Electron Device Manufacturing Conference 2024 Round-Up: TSMC N2 + Next-Gen SoIC, Intel EMIB-T, Meta 3D Stacked Memory, CFET, 2D Materials, and More (Semianalysis)

🦾 CHERI Protects Memory at the Hardware Level: CHERI focuses on memory protection by ensuring that data cannot be accessed outside the bounds of what is permitted (EETimes Europe)

⚛️ Distributed quantum computing across an optical network link (Nature)

⚛️ Topological Quantum Computing: Rethinking the fundamental physics used to create a qubit (Nokia Bell Labs)

Funding News

⚡️ €1M Grant – Pixel Photonics: multi-mode single photon detection (Tech EU)

⚡️ $30M Series A – Salience Labs: photonic switches to connect AI clusters that is compatible with existing infrastructure while delivering high bandwidth, low latency, and significant power savings (BusinessWire)

🦾 US$700 million Round – Wuxi SJ Semiconductor: China fosters another IC unicorn as BYD fund moves in

Deep Dive: The Real Opportunity for Computing in Europe

Last week, I had the chance to speak with GLOBSEC, a global think tank, about the latest U.S. restrictions on AI chips and their impact on Europe. After interviewing 140+ founders and industry experts on the Future of Computing, I think the real opportunity lies in next-generation computing:

💡 Love these insights? Forward this newsletter to a friend or two who can subscribe here