Actualités

The metals industry has experienced transformational moments before: the Bessemer process, flash smelting, electric arc furnaces, and modern continuous casting all changed what was possible. Yet, for more than a century, most extraction processes rely on techniques and pathways that would be familiar to metallurgists from generations ago. Processes have become larger and control systems smarter, but the fundamental science behind separating metals from ores has changed less than many realize. Today’s materials challenges are different from those of the 20th century, but why are critical metals still produced using largely the same scientific foundations? EvoMaterials, a technology platform developed by Outokumpu—the global leader in sustainable stainless steel and the producer of ferrochrome with 67% lower carbon footprint than the industry average—is exploring that question by rethinking how metals can be extracted: “We decided to go beyond the current constraints and look fundamentally at what’s the best way to do this with the benefit of the scientific engineering knowledge and hindsight that we have today,” says Caspar Stinn, technical director for extractive metallurgy at Outokumpu. “We’re looking at how we can reimagine metals production.” A supply chain problem disguised as a materials problem As economies expand and the world becomes increasingly electrified, data and scientific models point to rising demand for many minerals, driven by technologies ranging from data centers to aerospace, water treatment, energy systems, and defence. According to Antoine Allanore, professor of metallurgy at the Massachusetts Institute of Technology (MIT) this coincides with many Western economies having reduced their domestic processing capacity for the very metals those industries depend on. Metallurgy is now strategically important, shaped not only by cost and availability but also by carbon footprint, geographic origin and supply-chain resilience. That reality is particularly relevant for chromium and other specialty metals used in high-temperature and extreme environments. Chromium-based materials play a critical role in aerospace systems, power generation, corrosion-resistant infrastructure and defense technologies, where materials must perform under demanding conditions. Beyond optimization For decades, metallurgical innovation largely focused on improving yields, energy efficiency and process control. EvoMaterials is based on the idea that some metallurgical processes may benefit from entirely different chemical pathways. The platform originates from research into sulfidation-based processing technologies, initially pursued to reduce carbon emissions associated with ferrochrome production. Laboratory work revealed broader potential. Lab-scale trials demonstrated that these alternative process chemistries could potentially do more than lower emissions. They could recover valuable metals from lower grade resources and industrial side streams while fundamentally altering how separation processes are performed. The opportunity is to rethink how mineral resources are used and how much value can be extracted from them, bringing in greater resource efficiency and stronger industrial competitiveness. Understanding traditional smelting vs. new proprietary technology Source: Outokumpu Development of the technology began in the United States in 2021 for chromite from Outokumpu’s Kemi mine in Finland, the only chrome mine in the European Union and one of the few such operations across the Western world. Since then, Outokumpu has evolved the technology through extensive research and development, combining new process chemistries with proprietary metallurgical expertise, supported in part by a non-exclusive patent license agreement from MIT. “The technology is a new application for separating metallic elements from materials,” explains Allanore. Historically, that separation has been achieved using carbon- and energy-intensive approaches. The new method creates a different chemical environment that enables more efficient extraction with fewer losses and fewer unwanted by-products, recovering critical materials from resources considered waste. Innovation happens at the boundaries EvoMaterials sits at the intersection of several scientific disciplines, a combination that strongly resonates with Stinn, who completed his PhD in Materials Science and Engineering at MIT, where his research focused on low-carbon metals processing and recycling technologies. Modern materials innovation draws on a growing mix of chemistry, materials science, physics, computational modeling, and engineering. With advanced simulation tools and AI helping researchers identify promising directions, scientific discovery still depends on experimentation and testing in the lab—on getting your hands dirty and doing the hard work. “Science today is much more interdisciplinary than it was, and the collaboration across disciplines reveals new scientific opportunities,” says Stinn. Engineering can translate those opportunities into more efficient and cost-competitive metals production. Building a technology platform A technically significant aspect of EvoMaterials is that it’s not a single process, but a technology platform. Allanore compares EvoMaterials to platform technologies in other industries, where a single scientific breakthrough can enable multiple downstream applications. Once a new extraction pathway has been mastered for one material system, the same scientific principles may be adapted to entirely different metals and mineral resources. The initial application becomes only the starting point. “The future and the potential impact might be much broader than the original and current developments that are happening,” Allanore explains. That possibility is increasingly important as industries seek secure supplies of nickel, chromium, molybdenum, and other strategic materials needed for advanced manufacturing. The future of materials companies The broader implication extends beyond metallurgy. Companies that develop proprietary process technologies may create value through production but also through intellectual property and scientific expertise. Based on Outokumpu’s technology leadership, that transition could create entirely new categories of value, enabling the production of high-value critical materials from ores and resource streams that previously appeared uneconomical or inaccessible. The result would be a fundamental shift in how metals are manufactured, where sustainability, resource efficiency, supply security, and economic value become outcomes of the same technological architecture rather than separate objectives. The next step for EvoMaterials is pilot-scale validation, with a pilot plant in New Hampshire to begin operations in the first half of 2027. The facility will scale the proprietary technology from laboratory to one-ton production, providing a step toward industrial-scale deployment targeted for 2030. Beyond producing low-CO2, chromium-enriched ferrochrome and chromium metal, the platform could be adapted to other strategic materials, including nickel and molybdenum. The work raises a broader question—how many of today’s metallurgical processes would look the same if they were invented from scratch? Or could those

Then you train the neural network to reverse this process, commonly using a mean squared error objective.

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. Nancy Pelosi’s replacement is ready to regulate AI —Grace Huckins As a California state senator, Scott Wiener authored legislation aimed at curbing the risks of advanced AI. Now, after winning a tight congressional race for Nancy Pelosi’s open San Francisco seat, he could play a key role in setting federal AI policy. I spoke with Wiener only a few hours after he had been declared the victor. But he was already thinking about AI risk: during our 20-minute conversation, he used the phrase “recursive self-improvement” half a dozen times.  “We need to slow this down,” he told me. “Because it’s moving very quickly, and we could sleepwalk into a disaster.” Still, he describes himself as only “cautiously optimistic” about Congress’s ability to rein in AI. Here’s what he plans to do about it. The backlash about data centers might really just be about electricity  —James O’Donnell As voters went to the polls across the US last week, I was in Ashburn, Virginia, part of what’s known as Data Center Alley—the world’s largest data center hub. It’s also where the backlash sweeping the country arguably first began to take shape. Big Tech has framed the data center uproar as a culture war over progress. But a few days in Ashburn convinced me the real fault line may be electricity: drawing it from the grid or generating it onsite provokes different forms of blowback. The industry can improve its image, but it can’t escape the consequences of its enormous appetite for electricity. Find out why.  This story is from The Algorithm, our weekly AI newsletter. Sign up to receive it in your inbox every Monday. The must-reads I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology. 1 The FBI is investigating Nvidia CEO Jensen Huang’s affairBut Trump says he doesn’t think Huang needs to worry. (AP News)+ Huang hired a PR firm known for handling sex scandals last year. (Wired $)+ The allegations will likely spur new chip export restrictions. (Reuters $)+ They also raise questions about Huang’s lobbying on China. (Verge)+ And threaten the fate of the economy. (WP $) 2 France has accused Iran of hacking its supermarket fridgesProsecutors are investigating the leaders of an Iranian cyber unit. (NPR)+ Hackers are shifting to manipulating physical equipment. (Axios)+ Is your smart kitchen next? (Wired $) 3 The Iran war has triggered a drop in global emissionsThe main driver is energy supply constraints. (New Scientist $)+ But the decline will likely prove temporary. (Gizmodo)+ Our climate models have a big blind spot. (MIT Technology Review) 4 The FDA has ordered GLP-1s to carry warnings about blindnessIt says the drugs may increase the risk of vision loss. (NBC News)+ GLP-1s could also exacerbate eating disorders. (MIT Technology Review) 5 Australia is mulling criminal penalties for rogue AIThe proposals follow an AI agent hacking government systems. (BBC) 6 The Pentagon is planning to patrol the MoonThe project’s first five satellites could launch in 2029. (Ars Technica) 7 Two US cities have voted to ditch Flock surveillance camerasYork, Maine, and League City, Texas, have voted to ban them. (Vox)+ What Flock’s defenders are missing. (MIT Technology Review) 8 Constant surveillance is changing how children grow upTracking apps can fuel anxiety and undermine resilience. (Atlantic $) 9 A tiny RNA molecule could explain how life beganSome researchers believe life emerged from self-copying molecules. (Quanta) 10 A sci-fi novel has won the Booker Prize for the first timeThe End of Everything was awarded the prestigious honor. (Gizmodo) Quote of the day “Nvidia is not an ordinary company and Jensen is not an ordinary ⁠CEO; he is Nvidia, clearly sets the long-term strategy and direction, and pulls most of the company’s strings himself.” —Stacy Rasgon, an analyst at ​Bernstein, tells Reuters why it will be so difficult to replace Jensen Huang as Nvidia CEO. One More Thing China wants to restore the sea with high-tech marine ranches Just a short ferry ride from the Chinese port city of Yantai sits Genghai No. 1, a strange offshore complex that combines a tourist attraction with an industrial fish farm. It even has a seven-story visitor center shaped like a cartoon starfish. But its main purpose is to breed fish and release them into the ocean.  Genghai is part of China’s ambitious marine ranching program. Since 2015, the country has built 169 demonstration ranches and released at least 167 billion juvenile fish and shellfish. But the approach has also proved controversial. Dive into China’s divisive plans to restore the seas with marine ranches. —Matthew Ponsford We can still have nice things A place for comfort, fun, and distraction to brighten up your day. (Got any ideas? Drop me a line.) + “Earth artist” David Popa has created a new land-art portrait in the mountains of Norway.+ The winners of Wildlife Photographer of the Year capture both the power and fragility of nature.+ A Minnesota man who stopped a runaway school bus got surprised with a $1,000 cheque from Superman.+ After more than two decades without regular star shows, Frank Lloyd Wright’s only planetarium has been restored.

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. Opinion: The climate case for AI has not been made  —By Mary Nichols, distinguished counsel at the UCLA School of Law and the former chair of the California Air Resources Board, and Jeremy Tamanini, the founder of Dual Citizen, a sustainability research and advisory firm This week, as countries meet in Antalya, Turkey for the United Nations climate conference, one topic is likely to dominate the conversation: artificial intelligence. The host has already called on countries to commit to ensuring that AI is designed, powered, and deployed in ways that support climate action. But so far, industries around the world aren’t doing nearly enough to harness the technology to actually drive down the greenhouse-gas emissions heating up the planet. AI companies like to make very optimistic claims about AI’s potential to help address climate change. The reality is far less promising. Read the full story. Meet the ecologist revealing the invisible networks that revive rainforests   Tropical forests are our planet’s biggest terrestrial carbon sink, accounting for around 40% of carbon stored on land. Forest ecologist Daisy Dent, a senior scientist at ETH Zurich, is part of an international network of researchers working to decipher the networks of animal and plant interactions that enable some forests to recover after destruction.  The goal is a low-cost, high-impact alternative to typical carbon-­crediting schemes, which plant huge areas with a single species of tree.  For over a decade, her research has helped show that these large “monocrop” programs could be expensive mistakes—and that more resilient forests can rebound naturally in just a few decades, re-creating ecosystems that suck up vast quantities of carbon dioxide.  Learn more about Dent, her work, and her impact on how we understand the role rainforests play in regulating our climate.  —Matthew Ponsford The must-reads I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology. 1 Nvidia is investigating its own bossJensen Huang has admitted to intimacy with a suspected Chinese spy. (NYT $)+ Shares in Nvidia were down 5% in pre-market trading. (CNBC)+ Huang’s children are also employed by Nvidia. (WSJ $) 2 Democrats are planning a blitz of investigations into Big TechReportedly starting with Apple, Meta, and Musk’s companies. (WP $)+ They’re hoping to harness the data center backlash to rein in the AI sector. (Vox $)+ Here’s why the midterms sparked an AI stock selloff. (BBC) 3 The UN annual climate meeting starts todayIt’s the first to be held after confirmation the planet will exceed 1.5 °C warming. (Nature)+ How to study a swiftly melting climate disaster. (MIT Technology Review) 4 Connecting fridges to the internet looks like a bad move nowWe still don’t know who is behind recent hacks, or even if they’re connected. (The Atlantic $)+ Experts say they should be a “wake up call” for the industry. (NPR)+ Are we in the foothills of a hacking epidemic? (The Guardian)+ AI is already making online crimes easier. It could get much worse. (MIT Technology Review) 5 Tesla is axing the Cybertruck RIP, we hardly knew ye. (Wired $)+ The Cybertruck was on our “worst tech flops of 2025” list. (MIT Technology Review) 6 Iran might attack subsea cables if the US escalates againThat’s the message it’s sending via several loitering tankers. (FT $) 7 Chinese AI labs are struggling to poach Silicon Valley workersStartups like MiniMax and Moonshot have an uphill battle as they try to fill new US offices. (Rest of World)+ China’s AI models have Trump’s AI world at war with itself. (MIT Technology Review) 8 Online piracy is back But this time, archivist zeal is at least as big a draw as accessing content for free. (The Verge) + Inside the race to save our online lives from a digital dark age. (MIT Technology Review) 9 How do you land on a moon with almost no gravity?It’s a huge challenge to land on Martian moon Phobos—here’s how Japan’s going to attempt it. (IEEE Spectrum) 10 Meet Philadelphia’s delivery robot saboteurCarolyn Marvin is going to court for blocking them, and it’s turning her into a local hero. (The Guardian) Quote of the day “Leading NVIDIA is the privilege of my life. My responsibility is to serve you and be worthy of your trust. I have fallen short of that responsibility.”  —Jensen Huang apologizes to Nvidia employees for his affair with a suspected Chinese spy in a mass email sent yesterday, the Washington Post reports. One more thing STUART BRADFORD AI is coming for music, too AI models are marching into a creative field that is uniquely vulnerable to disruption: music.   AI-generated creative works—from orchestra performances to heavy metal—are poised to suffuse our lives more thoroughly than any other product of AI has done yet.  The songs are likely to blend into our streaming platforms, party and wedding playlists, soundtracks, and more, whether or not we notice who (or what) made them.  Here’s why AI is set to forever change how we make music.  —James O’Donnell We can still have nice things A place for comfort, fun, and distraction to brighten up your day. (Got any ideas? Drop me a line.) + French bardcore band Corseval has given ABBA a medieval makeover.+ A stunning collection of prehistoric engravings has been found in a Spanish cave.+ This browser game flips a retro classic on its head: you control the asteroids, not the spaceship.+ From a caiman in a wig to a leopard doing a handstand, the Comedy Wildlife Awards shortlist is a delight.

Overview This post is divided into eight parts; they are: • The Checkpoint Fixes the Attention Architecture • A Closer Look at Attention Memory • Multi-Head Attention • Multi-Query and Grouped-Query Attention • Sliding-Window and Mixed Attention • Multi-Head Latent Attention • What the Inference Runtime Can Change • From Cache Size to Serving Capacity The Checkpoint Fixes the Attention Architecture The checkpoint fixes the model configuration; the inference engine does not change it.

Overview This post is divided into five parts; they are: • How Floating-Point Numbers Work • Common Precision Formats • Loading a Model at a Different Precision • Mixed Precision and Sensitive Operations • Measuring the Result How Floating-Point Numbers Work A floating-point number contains a sign, an exponent, and a significand (often called the mantissa).