Save this picture!VARID A VR-AR Toolkit for Inclusive Design.. Image © Foster + PartnersOver the last decade, architectural design has relied on 2D methods of representation, such as elevations, sections, and floor plans, paired with digital renderings of 3D models. While these tools are essential to convey geometry and intent, they remain limited by their two-dimensional format. Even the most realistic renderings, created through programs like SketchUp, Revit, or AutoCAD, still flatten space and distance the viewer from the lived experience of a project. Recently, architects have begun to explore immersive technologies as a way to bridge this gap between drawing and experience, offering new ways to inhabit and assess spatial proposals.What are AR, VR, and MR?Extended Reality (XR) can be classified into three main types: Augmented Reality (AR), Virtual Reality (VR), and Mixed Reality (MR), each offering varying levels of immersion in digital environments. At one end of the spectrum, AR enhances the real world with digital content, while at the other, VR fully immerses the user in a completely virtual environment, blocking out the physical world. MR lies between these extremes but is essentially a more detailed classification of AR based on the type of display used. Their research proposes the following classification: Class 1 display refers to monitor-based systems, where users view the real world through a screen equipped with a camera that captures the environment and overlays digital information, such as in the Apple Vision Pro, which uses passthrough cameras. In contrast, Class 2 and 3 systems use head-mounted displays (HMDs) with see-through lenses that superimpose 3D models onto the user's view, like the Microsoft HoloLens. In 2020, Trimble combined the HoloLens with a hard hat, creating the Trimble XR10, which makes this technology usable in the construction site. For clarity, this text will refer to Class 1 systems as AR and Class 2 and 3 systems as MR moving forward. Related Article Using Augmented Reality In Bamboo Architecture Save this picture!How do Users Perceive Space?Architectural design is not only about defining space, but also about anticipating how people will perceive and move through it. The way users interpret a space depends not just on geometry, but also on intuition, their individual knowledge, and experiences. Kevin Lynch described this as a space's "legibility," or how easily it can be understood and organized mentally, while Ittelson (1978) emphasizes how users explore, categorize, and systematize spatial elements into a coherent whole. The user first explores an area to orient themselves and move around, then they will develop a taxonomy of the space elements to mentally organize it, and finally, they put everything together into a system that tells the brain why things are happening and how they relate to each other. Research suggests that immersive environments such as mixed reality can simulate this faithfully, allowing architects and clients alike to engage with a design not as an abstract plan, but as a place to walk through, observe, and interpret.Save this picture!Which One Improves Design Understanding: 2D Drawings or MR?Based on the above, a study made by the National Taiwan University in 2021 explored this topic by conducting an experiment where participants were brought to a room and were divided into two groups. The first would analyze an interior design proposal of the space using printed architectural drawings and colored renderings. The second group was asked to do the same but only used the explorable MR 3D model seen through an MR headset, in this case: The HoloLens. After the exploration was done, users would sit down, and researchers would ask questions about the space. For example, the general understanding of the elements in the architectural program, how well people perceive length and sizes of objects, perception and understanding of textures and materials, and knowledge of demolition or renovation of specific elements. A total of 42 people participated in the research, with an average age of 26 years, various ranges of architectural drawing understanding, and from diverse cultural backgrounds, including Africa, the Middle East, Asia, the Americas, and Europe. The results shed light on several topics for architects looking into implementing this technology in their work.Save this picture!First, the study suggests that MR technology allowed users to understand around 85% of the overall design proposal compared to 2D methods (which allowed participants to obtain only around 75% of the information). At the same time, they also concluded that MR does not fully replace 2D; in fact, it's about balance. Both MR and 2D are suitable for identifying spaces and general layout, identifying where activities can be performed, and identifying heights. However, 2D plans are especially good for specific measurements of the space (Length and width), understanding the demolition plan, and identifying countable elements in the design, like the number of lamps, switches, or sockets. On the other hand, MR was better for understanding how elements in the space interact with each other (Like if the columns were wrapped by a specific material). MR  was especially useful for quickly identifying the specific materials and textures of the design and visually understanding size in terms of width, and mentally perceiving certain properties of materials like roughness, smoothness, warmth, or coldness.Save this picture!How Can We Integrate MR into our Current Design Review Workflows?MR has the potential to facilitate inclusive and interdisciplinary collaboration by bridging the gap between technical and non-technical stakeholders. Clients or end users with limited experience in reading architectural drawings often struggle to visualize how a space will look or function. AR, especially through Mixed Reality headsets, can mitigate this by allowing them to engage with the space intuitively. Given the transparent property of the MR lenses, non-architect users can experience the spatial and material qualities of a design proposal directly on-site, making it easier to identify potential issues such as circulation conflicts, scale misinterpretations, or material inconsistencies. This allows them to give feedback that is grounded in their own perceptual experience rather than abstract interpretations. This can help to democratize the design review process and can lead to more informed, client-centered decisions. For architectural teams, combining MR with traditional tools might mean that their detailed technical evaluations (e.g., clearances, counts, and demolition plans) are complemented by a richer experiential understanding from the client, which can lead to more holistic and user-validated design outcomes.Save this picture!This article is part of the ArchDaily Topics: What Is Future Intelligence?, proudly presented by Gendo, an AI co-pilot for Architects. Our mission at Gendo is to help architects produce concept images 100X faster by focusing on the core of the design process. We have built a cutting edge AI tool in collaboration with architects from some of the most renowned firms such as Zaha Hadid, KPF and David Chipperfield.Every month we explore a topic in-depth through articles, interviews, news, and architecture projects. We invite you to learn more about our ArchDaily Topics. And, as always, at ArchDaily we welcome the contributions of our readers; if you want to submit an article or project, contact us.

Klarna said it used an AI doppelganger of CEO Sebastian Siemiatkowski to report its updated quarterly earnings. Klarna 2025-05-22T02:31:59Z Save Saved Read in app This story is available exclusively to Business Insider subscribers. Become an Insider and start reading now. Have an account? Klarna said it used an AI avatar of its CEO to report quarterly earnings in a YouTube video. Klarna brands itself as an AI company and has "streamlined" its workforce by 40% since 2022. Klarna's Q1 results show revenue growth but see a spike in net and credit losses. Buy now, pay later services company Klarna said it used an AI doppelganger of CEO Sebastian Siemiatkowski to report its quarterly earnings on Monday.The AI avatar appeared in a video on Klarna's official YouTube channel to deliver earnings highlights. It wore a brown jacket reminiscent of one in Siemiatkowski's corporate headshots, and aside from a lack of blinking and suspect"Our AI-first strategy is driving exceptional returns, we're outpacing competitors, our merchant network is scaling rapidly, and our next-gen products are reshaping money management for millions," a presumably human Siemiatkowski said in a press release.The move comes as Klarna, which last month put its IPO on ice due to economic uncertainty, tries to brand itself as an AI company. In the earnings press release, Klarna said it has "streamlined" its workforce by around 40% since 2022.In 2022, 800 employees were fired, while some were quietly offered an exit package last year after being placed into a "talent pool." In February 2024, Klarna announced that itsKlarna has recently ramped up partnerships with platforms like Walmart, eBay, and DoorDash, but consumer watchdogs have long been concerned about the potential for overspending under BNPL services. Under the Biden administration, theBNPL providers as credit card lenders, which required stricter protections around disclosures and disputes.The Federal Reserve found in 2024 that users of BNPL services are more likely to rely on high-interest financing tools and are more financially fragile.LendingTree, an online lending marketplace, also found in an April survey that 41% of BNPL users in the US paid late over the last 12 months, up from 34% a year ago.Klarna's latest Q1 results also show that an increasing number of people may not have been paying their loans. While revenue grew 13% year over year and it reached 100 million active users, Klarna also doubled its net losses from $47 million in Q1 2024 to $99 million in Q1 2025 — a 110% increase.During the May 19 earnings call, Klarna attributed the spike in losses to several one-off costs related to depreciation, share-based payments, and restructuring.Klarna's consumer credit losses have also jumped, which its Q1 financial report said is "driven by the accelerated expansion of Pay Later and Fair Financing products." Klarna's first quarter saw a 17% year-on-year increase in credit losses from $117 million to $136 million.Klarna did not immediately respond to requests for comment.

But the added value... Google has set its new Ultra plan at $50 more than any other top-tier AI subscription. While it boasts exclusive tools and integrations, much of the offering repackages features that are already available in other Google services. If competitors follow this lead, AI subscriptions could become significantly more expensive – without offering significantly more value. Google introduced its AI Ultra subscription at I/O 2025, calling it the most advanced option for users who want "the absolute best" of Google AI. Priced at $249.99 per month, it's $50 or more higher than top-tier plans from OpenAI ($200/mo) and Anthropic ($100/mo). New subscribers get a 50-percent discount for the first three months, making it temporarily cheaper than OpenAI. Google frames Ultra as a premium-tier product for creative professionals, developers, and academics who need extended access to its largest models and experimental tools, including the latest Gemini 2.5 Pro model. However, the pricing reflects more than just higher usage limits – it sets a new baseline for what AI companies may soon charge. For $250, Ultra includes access to Google's latest versions of Gemini, the new Deep Think mode in Gemini 2.5 Pro, and early access to the Veo 3 video model, which is quite impressive. It also unlocks high-resolution video generation in Flow, a cinematic video creation tool built around Veo, Imagen, and Gemini. The package offers top-tier usage limits across these tools and others like Whisk Animate and NotebookLM, with more enhancements promised later this year. Many of these features aren't exclusive to Ultra. Google is rolling out Flow to the $20-per-month AI Pro subscription with fewer capabilities. Some Gemini integrations, such as those in Gmail, Docs, and Chrome, are already free or included in lower tiers. So, Ultra subscribers are essentially paying to beta test features that will soon become standard. Google Ultra includes 30 terabytes of Google One cloud storage, a YouTube Premium subscription, and Project Mariner, a task manager handling up to 10 concurrent threads. While these add some value, they aren't AI-specific. YouTube Premium and extra storage are available separately, making this bundle feel more like padding than a core upgrade. For users primarily seeking AI capabilities, paying extra for external subscriptions they may not want or use offers limited additional value. // Related Stories Meanwhile, tools like Whisk and NotebookLM, framed as Ultra upgrades, are also available in basic forms elsewhere. Even Veo and Flow, which sound exclusive, are already seeing partial rollout to AI Pro users. The most meaningful perk may be higher usage caps, but the practical value of those limits remains unclear – a frustratingly common trait across the AI industry. If Google's Ultra tier catches on, it could nudge competitors like OpenAI to introduce similarly priced or bundled offerings – even if the functionality remains essentially identical. That would shift the market from a race to deliver the best AI to a race to monetize "premium" features, regardless of whether they provide real differentiation. Google may be positioning AI Ultra as an elite creative toolkit, but it may also be redefining "premium" as just more expensive.

The Linux Foundation and Meta are putting some numbers behind how open-source AI (OSAI) is driving innovation and adoption.The adoption of AI tools is pretty much everywhere now, with 94% of organisations surveyed already using them. And get this: within that crowd, 89% are tapping into open-source AI for some part of their tech backbone.A paper released this week by Meta and the Linux Foundation stitches together academic brainpower, industry frontline stories, and global survey data to showcase an ecosystem that’s buzzing thanks to being open and affordable.If there’s one thing that jumps off the page, it’s the money talk. Cost savings, folks, are a huge deal here. Unsurprisingly, two-thirds of businesses are saying that open source AI is just plain cheaper to get up and running compared to proprietary. So, it’s no shocker that almost half of them point to these savings as a big reason for going the open-source route.We’re not talking about trimming a few coins here and there. Researchers reckon companies would be shelling out 3.5 times more cash if open-source software simply vanished. As AI digs its heels deeper into everything we do, the financial muscle of open-source is only going to get stronger, potentially even overshadowing traditional open-source software’s impact.But this isn’t just about pinching pennies; it’s about unleashing brains. The report points out that AI can slash business unit costs by over 50%, which, as you can imagine, opens the door for revenue boosts. When open AI models are out there for cheap, or even free, it levels the playing field. Suddenly, developers and businesses of all sizes can jump in, play around, and rethink how they do things.Often it’s the smaller players, the agile startups and medium-sized businesses, that are diving headfirst into open-source AI more so than the big corporate giants. And since these are often the places where groundbreaking ideas and new products are born, it really hammers home how vital OSAI is for keeping the innovation engine chugging and helping those plucky, cutting-edge firms compete.And if you want a textbook example of how going open can turbocharge things, look no further than PyTorch. The report digs into how Meta’s decision to shift its heavyweight deep learning framework to an open governance model, under a non-profit, turned out to be a masterstroke.The report leans on a close look by Yue and Nagle (2024), who tracked what happened next. Once PyTorch flew the Meta nest, contributions from Meta itself “significantly decreased.” Sounds a bit off, right? But actually, it signalled a healthy move away from one company calling the shots.What really ramped up was input from “external companies, especially from the developers of complementary technology, such as chip manufacturers.” Meanwhile, the actual users, the developers building stuff with PyTorch, kept their engagement steady – “no change.”It’s a clear win. As the researchers put it, this kind of shift for major OSAI software “promotes broader participation and increased contributions and decreases the dominance of any single company.” It’s a powerful testament to what report authors Anna Hermansen and Cailean Osborne found: “engagement in open, collaborative activities is a better indicator of innovation than patents.”This isn’t just theory; it’s making waves in massive sectors. Take manufacturing. Open-source AI is set to be a game-changer there, mostly because its open code means you can bend it and shape it to fit. This flexibility allows AI to slot neatly into factory workflows, automating tasks and smoothing out order management. A 2023 McKinsey report, flagged in the study, even predicts AI could pump up to $290 billion extra into advanced manufacturing.Then there’s healthcare. In places like hospitals and local clinics, where every penny and every minute counts, free and flexible tools like open-source AI can literally be lifesavers. Imagine AI helping with diagnoses or flagging diseases early.McKinsey thinks the global healthcare sector could see up to a $260 billion boost in value once AI is really rolled out. A 2024 analysis even showed that open models in healthcare can go toe-to-toe with the proprietary ones—meaning hospitals can get tailored, privacy-friendly OSAI without skimping on performance.And it’s not just about the tech; it’s about the people. The report mentions that AI-related skills could see wages jump by up to 20%. That’s a big deal and really underlines why we need to be thinking about training and development for this new AI era.Hilary Carter, SVP of Research at The Linux Foundation, said: “The findings in this report make it clear: open-source AI is a catalyst for economic growth and opportunity. As adoption scales across sectors, we’re seeing measurable cost savings, increased productivity and rising demand for AI-related skills that can boost wages and career prospects.“Open-source AI is not only transforming how businesses operate—it’s reshaping how people work.”So, the takeaway? Open AI models are fast becoming the standard, the very foundation of future breakthroughs. They’re pushing growth and healthy competition by making powerful AI tools available without an eye-watering price tag.The Linux Foundation’s report isn’t just cheerleading; it’s laying out the hard numbers to show why open-source AI is absolutely crucial for a robust, stable, and forward-looking economy.Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is co-located with other leading events including Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.Explore other upcoming enterprise technology events and webinars powered by TechForge here.

May 20, 2025The Hacker NewsPenetration Testing / Risk Management In the newly released 2025 State of Pentesting Report, Pentera surveyed 500 CISOs from global enterprises (200 from within the USA) to understand the strategies, tactics, and tools they use to cope with the thousands of security alerts, the persisting breaches and the growing cyber risks they have to handle. The findings reveal a complex picture of progress, challenges, and a shifting mindset about how enterprises approach security testing. More Tools, More Data, More Protection… No Guarantees Over the past year, 45% of enterprises expanded their security technology stacks, with organizations now managing an average of 75 different security solutions​. Yet despite these layers of security tools, 67% of U.S. enterprises experienced a breach in the past 24 months​. The growing number of deployed tools has a few effects on the daily operation and the overall cyber posture of the organization. Although it seems obvious, the findings tell a clear story - more security tools do mean better security posture. However, there is no silver bullet. Among organizations with fewer than 50 security tools, 93% reported a breach. That percentage steadily declines as stack size increases, dropping to 61% among those using more than 100 tools. Alert Fatigue Is Real The flip side of larger security stacks is that CISOs and their teams must contend with a much larger influx of information. Enterprises managing over 75 security solutions now face an average of 2,000 alerts per week — double the volume compared to organizations with smaller stacks, and those with over 100 tools receive over 3000 (3x the alerts). This in turn, puts much more emphasis on effective prioritization, otherwise, critical threats may get buried in a sea of alerts. In this environment, where alert volumes are high and time to triage is short, organizations benefit most when they can frequently test for exploitable gaps, so they know which issues truly matter before threat actors find them first. Software-Based Pentesting Gains Ground Trust in software-based security testing is growing rapidly. Only 5-10 years ago, many enterprises would never have permitted automated tools to run pentests in their environments for fear of causing outages, but sentiment is changing. As CISOs continue to recognize the advantages of software in scaling adversarial testing and keeping pace with constantly changing IT environments, software-based pentesting is becoming the standard. Over half of enterprises now use these tools to support in-house testing, driven by trust in their reliability and the need for scalable, continuous validation strategies. Today, 50% of CISOs cite software-based pentesting solutions as their primary method for uncovering exploitable gaps​. Insurance Providers Become Unexpected Influencers Beyond internal management and Boards of Directors, a surprising new force is shaping security strategy: Cyber insurance providers. 59% of CISOs admitted that they have implemented at least one cybersecurity solution that they were not previously considering as a result of their cyber insurers. It's a clear sign that insurers aren't just pricing risk, they're actively prescribing how to reduce it, and reshaping enterprise security priorities in the process.​. Low Confidence in Government Support While governmental agencies like CISA (in the US) and ENISA (in the EU) play an important role in threat visibility and coordination, confidence in government cybersecurity support is surprisingly low. Only 14% of CISOs believe the government is adequately supporting the private sector's cyber challenges​, while 64% feel that government efforts, though acknowledged, are insufficient​. 22% believe that they cannot rely on the government at all for cybersecurity help. To benchmark your organization's pentesting practices, budgets, and priorities against other global enterprises, register for the webinar on May 27, 2025 where senior security analysts will discuss the key findings. Alternatively, get the full 2025 State of Pentesting Report and see all the insights for yourself!Note: This article was written and contributed by Jay Mar Tang, Field CISO at Pentera. Found this article interesting? This article is a contributed piece from one of our valued partners. Follow us on Twitter  and LinkedIn to read more exclusive content we post. SHARE    

When we set out to write a story on the best available estimates for AI’s energy and emissions burden, we knew there would be caveats and uncertainties to these numbers. But, we quickly discovered, the caveats are the story too.  This story is a part of MIT Technology Review’s series “Power Hungry: AI and our energy future,” on the energy demands and carbon costs of the artificial-intelligence revolution. Measuring the energy used by an AI model is not like evaluating a car’s fuel economy or an appliance’s energy rating. There’s no agreed-upon method or public database of values. There are no regulators who enforce standards, and consumers don’t get the chance to evaluate one model against another.  Despite the fact that billions of dollars are being poured into reshaping energy infrastructure around the needs of AI, no one has settled on a way to quantify AI’s energy usage. Worse, companies are generally unwilling to disclose their own piece of the puzzle. There are also limitations to estimating the emissions associated with that energy demand, because the grid hosts a complicated, ever-changing mix of energy sources.  It’s a big mess, basically. So, that said, here are the many variables, assumptions, and caveats that we used to calculate the consequences of an AI query. (You can see the full results of our investigation here.) Measuring the energy a model uses Companies like OpenAI, dealing in “closed-source” models, generally offer access to their  systems through an interface where you input a question and receive an answer. What happens in between—which data center in the world processes your request, the energy it takes to do so, and the carbon intensity of the energy sources used—remains a secret, knowable only to the companies. There are few incentives for them to release this information, and so far, most have not. That’s why, for our analysis, we looked at open-source models. They serve as a very imperfect proxy but the best one we have. (OpenAI, Microsoft, and Google declined to share specifics on how much energy their closed-source models use.)  The best resources for measuring the energy consumption of open-source AI models are AI Energy Score, ML.Energy, and MLPerf Power. The team behind ML.Energy assisted us with our text and image model calculations, and the team behind AI Energy Score helped with our video model calculations. Text models AI models use up energy in two phases: when they initially learn from vast amounts of data, called training, and when they respond to queries, called inference. When ChatGPT was launched a few years ago, training was the focus, as tech companies raced to keep up and build ever-bigger models. But now, inference is where the most energy is used. The most accurate way to understand how much energy an AI model uses in the inference stage is to directly measure the amount of electricity used by the server handling the request. Servers contain all sorts of components—powerful chips called GPUs that do the bulk of the computing, other chips called CPUs, fans to keep everything cool, and more. Researchers typically measure the amount of power the GPU draws and estimate the rest (more on this shortly).  To do this, we turned to PhD candidate Jae-Won Chung and associate professor Mosharaf Chowdhury at the University of Michigan, who lead the ML.Energy project. Once we collected figures for different models’ GPU energy use from their team, we had to estimate how much energy is used for other processes, like cooling. We examined research literature, including a 2024 paper from Microsoft, to understand how much of a server’s total energy demand GPUs are responsible for. It turns out to be about half. So we took the team’s GPU energy estimate and doubled it to get a sense of total energy demands.  The ML.Energy team uses a batch of 500 prompts from a larger dataset to test models. The hardware is kept the same throughout; the GPU is a popular Nvidia chip called the H100. We decided to focus on models of three sizes from the Meta Llama family: small (8 billion parameters), medium (70 billion), and large (405 billion). We also identified a selection of prompts to test. We compared these with the averages for the entire batch of 500 prompts.  Image models Stable Diffusion 3 from Stability AI is one of the most commonly used open-source image-generating models, so we made it our focus. Though we tested multiple sizes of the text-based Meta Llama model, we focused on one of the most popular sizes of Stable Diffusion 3, with 2 billion parameters.  The team uses a dataset of example prompts to test a model’s energy requirements. Though the energy used by large language models is determined partially by the prompt, this isn’t true for diffusion models. Diffusion models can be programmed to go through a prescribed number of “denoising steps” when they generate an image or video, with each step being an iteration of the algorithm that adds more detail to the image. For a given step count and model, all images generated have the same energy footprint. The more steps, the higher quality the end result—but the more energy used. Numbers of steps vary by model and application, but 25 is pretty common, and that’s what we used for our standard quality. For higher quality, we used 50 steps.  We mentioned that GPUs are usually responsible for about half of the energy demands of large language model requests. There is not sufficient research to know how this changes for diffusion models that generate images and videos. In the absence of a better estimate, and after consulting with researchers, we opted to stick with this 50% rule of thumb for images and videos too. Video models Chung and Chowdhury do test video models, but only ones that generate short, low-quality GIFs. We don’t think the videos these models produce mirror the fidelity of the AI-generated video that many people are used to seeing.  Instead, we turned to Sasha Luccioni, the AI and climate lead at Hugging Face, who directs the AI Energy Score project. She measures the energy used by the GPU during AI requests. We chose two versions of the CogVideoX model to test: an older, lower-quality version and a newer, higher-quality one.  We asked Luccioni to use her tool, called Code Carbon, to test both and measure the results of a batch of video prompts we selected, using the same hardware as our text and image tests to keep as many variables as possible the same. She reported the GPU energy demands, which we again doubled to estimate total energy demands.  Tracing where that energy comes from After we understand how much energy it takes to respond to a query, we can translate that into the total emissions impact. Doing so requires looking at the power grid from which data centers draw their electricity.  Nailing down the climate impact of the grid can be complicated, because it’s both interconnected and incredibly local. Imagine the grid as a system of connected canals and pools of water. Power plants add water to the canals, and electricity users, or loads, siphon it out. In the US, grid interconnections stretch all the way across the country. So, in a way, we’re all connected, but we can also break the grid up into its component pieces to get a sense for how energy sources vary across the country.  Understanding carbon intensity The key metric to understand here is called carbon intensity, which is basically a measure of how many grams of carbon dioxide pollution are released for every kilowatt-hour of electricity that’s produced.  To get carbon intensity figures, we reached out to Electricity Maps, a Danish startup company that gathers data on grids around the world. The team collects information from sources including governments and utilities and uses them to publish historical and real-time estimates of the carbon intensity of the grid. You can find more about their methodology here.  The company shared with us historical data from 2024, both for the entire US and for a few key balancing authorities (more on this in a moment). After discussions with Electricity Maps founder Olivier Corradi and other experts, we made a few decisions about which figures we would use in our calculations.  One way to measure carbon intensity is to simply look at all the power plants that are operating on the grid, add up the pollution they’re producing at the moment, and divide that total by the electricity they’re producing. But that doesn’t account for the emissions that are associated with building and tearing down power plants, which can be significant. So we chose to use carbon intensity figures that account for the whole life cycle of a power plant.  We also chose to use the consumption-based carbon intensity of energy rather than production-based. This figure accounts for imports and exports moving between different parts of the grid and best represents the electricity that’s being used, in real time, within a given region.  For most of the calculations you see in the story, we used the average carbon intensity for the US for 2024, according to Electricity Maps, which is 402.49 grams of carbon dioxide equivalent per kilowatt-hour.  Understanding balancing authorities While understanding the picture across the entire US can be helpful, the grid can look incredibly different in different locations.  One way we can break things up is by looking at balancing authorities. These are independent bodies responsible for grid balancing in a specific region. They operate mostly independently, though there’s a constant movement of electricity between them as well. There are 66 balancing authorities in the US, and we can calculate a carbon intensity for the part of the grid encompassed by a specific balancing authority. Electricity Maps provided carbon intensity figures for a few key balancing authorities, and we focused on several that play the largest roles in data center operations. ERCOT (which covers most of Texas) and PJM (a cluster of states on the East Coast, including Virginia, Pennsylvania, and New Jersey) are two of the regions with the largest burden of data centers, according to research from the Harvard School of Public Health.  We added CAISO (in California) because it covers the most populated state in the US. CAISO also manages a grid with a significant number of renewable energy sources, making it a good example of how carbon intensity can change drastically depending on the time of day. (In the middle of the day, solar tends to dominate, while natural gas plays a larger role overnight, for example.) One key caveat here is that we’re not entirely sure where companies tend to send individual AI inference requests. There are clusters of data centers in the regions we chose as examples, but when you use a tech giant’s AI model, your request could be handled by any number of data centers owned or contracted by the company. One reasonable approximation is location: It’s likely that the data center servicing a request is close to where it’s being made, so a request on the West Coast might be most likely to be routed to a data center on that side of the country.  Explaining what we found To better contextualize our calculations, we introduced a few comparisons people might be more familiar with than kilowatt-hours and grams of carbon dioxide. In a few places, we took the amount of electricity estimated to be used by a model and calculated how long that electricity would be able to power a standard microwave, as well as how far it might take someone on an e-bike.  In the case of the e-bike, we assumed an efficiency of 25 watt-hours per mile, which falls in the range of frequently cited efficiencies for a pedal-assisted bike. For the microwave, we assumed an 800-watt model, which falls within the average range in the US.  We also introduced a comparison to contextualize greenhouse gas emissions: miles driven in a gas-powered car. For this, we used data from the US Environmental Protection Agency, which puts the weighted average fuel economy of vehicles in the US in 2022 at 393 grams of carbon dioxide equivalent per mile.  Predicting how much energy AI will use in the future After measuring the energy demand of an individual query and the emissions it generated, it was time to estimate how all of this added up to national demand.  There are two ways to do this. In a bottom-up analysis, you estimate how many individual queries there are, calculate the energy demands of each, and add them up to determine the total. For a top-down look, you estimate how much energy all data centers are using by looking at larger trends.  Bottom-up is particularly difficult, because, once again, closed-source companies do not share such information and declined to talk specifics with us. While we can make some educated guesses to give us a picture of what might be happening right now, looking into the future is perhaps better served by taking a top-down approach. This data is scarce as well. The most important report was published in December by the Lawrence Berkeley National Laboratory, which is funded by the Department of Energy, and the report authors noted that it’s only the third such report released in the last 20 years. Academic climate and energy researchers we spoke with said it’s a major problem that AI is not considered its own economic sector for emissions measurements, and there aren’t rigorous reporting requirements. As a result, it’s difficult to track AI’s climate toll.  Still, we examined the report’s results, compared them with other findings and estimates, and consulted independent experts about the data. While much of the report was about data centers more broadly, we drew out data points that were specific to the future of AI.  Company goals We wanted to contrast these figures with the amounts of energy that AI companies themselves say they need. To do so, we collected reports by leading tech and AI companies about their plans for energy and data center expansions, as well as the dollar amounts they promised to invest. Where possible, we fact-checked the promises made in these claims. (Meta and Microsoft’s pledges to use more nuclear power, for example, would indeed reduce the carbon emissions of the companies, but it will take years, if not decades, for these additional nuclear plants to come online.)  Requests to companies We submitted requests to Microsoft, Google, and OpenAI to have data-driven conversations about their models’ energy demands for AI inference. None of the companies made executives or leadership available for on-the-record interviews about their energy usage. This story was supported by a grant from the Tarbell Center for AI Journalism.

Empowering engineering teams with more tools for building AI factories, NVIDIA today announced a significant expansion of the NVIDIA Omniverse Blueprint for AI factory digital twins, now available as a preview. The blueprint features new integrations across the AI factory power, cooling and networking ecosystems with industry leaders Delta Electronics, Jacobs and Siemens, joining existing partners Cadence, Schneider Electric with ETAP and Vertiv. This growing ecosystem unifies the design and simulation of billions of components required to build digital twins of AI factories. The expanded blueprint will equip engineering teams to design, simulate and optimize entire AI factories in physically accurate virtual environments, enabling early issue detection and the development of smarter, more reliable facilities. Built on reference architectures for NVIDIA GB200 NVL72-powered AI factories, the blueprint taps into Universal Scene Description (OpenUSD) asset libraries. This allows developers to aggregate detailed 3D and simulation data representing all aspects of the data center into a single, unified model, enabling them to design and simulate advanced AI infrastructure optimized for efficiency, throughput and resiliency. The Omniverse Blueprint for AI factory digital twins unifies AI factory power, cooling and networking components in one simulation. AI Factory Ecosystem Teams With NVIDIA The Omniverse Blueprint for AI factory digital twins brings together diverse partners and tools to optimize the design, simulation, deployment and operations of AI factories. Today, NVIDIA announced that new partners are contributing to the framework. Siemens is building 3D models according to the blueprint and engaging with the simulation-ready, or SimReady, standardization effort, while Delta Electronics is adding models of its equipment. Because these are built with OpenUSD, users get accurate simulations of their facility equipment. Jacobs is helping test and optimize the end-to-end blueprint workflow. They join leaders in data center power and cooling solutions like Schneider Electric with ETAP and Vertiv, which contribute SimReady assets to populate the digital twin of the AI factory with 3D models of power, cooling and mechanical systems. “As AI factories continue to scale at an unprecedented pace, the energy demands they generate are reshaping the entire digital infrastructure landscape,” says Tanuj Khandelwal, CEO of ETAP. “Using the Omniverse Blueprint and SimReady assets, customers can test and optimize energy efficiency for the complexity and intensity of their AI workloads before even breaking ground.” Connections to the Cadence Reality Digital Twin Platform and ETAP provide thermal and power simulation, enabling engineering teams to test and optimize power, cooling and networking long before construction begins. These contributions help NVIDIA and its partners reshape how AI infrastructure is built to achieve smarter designs, avoid downtime and get the most out of AI factories. “Digital twins are fundamental to meet the escalating global demand for AI factories,” said Ben Gu, corporate vice president of R&D for multiphysics system analysis at Cadence. “The integration of the Cadence Reality Digital Twin Platform with the NVIDIA Omniverse Blueprint transforms the entire engineering process to design AI factories more efficiently and operate them more effectively than ever before. We are excited to continue our full-stack collaboration with NVIDIA.” Building SimReady Assets for AI Factories The OpenUSD-based models within the blueprint are inherently SimReady, designed from the ground up to be physics-based. This is especially valuable for developing and testing physical AI and agentic AI within these AI factories, enabling rapid and large-scale industrial AI simulations of power and cooling systems, building automation and overall IT operations. SimReady standardization workflow enables developers to view standardized simulations of thermal airflow within a digital twin environment. A key enhancement to this blueprint is the SimReady standardization workflow. Originally developed as a SimReady standardization proposal to streamline NVIDIA’s internal creation of OpenUSD assets, this now publicly available, industry-agnostic resource offers standardized requirements and processes for developing SimReady capabilities. It empowers data center developers and owners to efficiently establish, optimize and rigorously test their own digital twins of critical infrastructure, particularly for electrical and thermal management within AI factories. A Smarter Road to AI Infrastructure The expansion of the NVIDIA Omniverse Blueprint for AI factory digital twins marks a significant leap forward in how engineers design, simulate and build the sophisticated infrastructure required for industrial AI. By providing a unified and physically accurate digital twin, built on the robust foundation of OpenUSD and guided by SimReady standardization, this blueprint enables the industry to de-risk development, optimize performance and accelerate the deployment of next-generation AI factories. Learn more about NVIDIA Omniverse and preview the Omniverse Blueprint for AI factory digital twins. Watch the COMPUTEX keynote from NVIDIA founder and CEO Jensen Huang, as well as NVIDIA GTC Taipei 2025 sessions. See notice regarding software product information. Featured image courtesy of Cadence, ETAP, Schneider Electric, and Vertiv.

Why it matters: As the keyboard space continues to evolve, Cherry's latest innovations show how far switch technology has come. By combining advances in sensing, efficiency, and feel, Cherry is refining its legacy and reshaping what users can expect from future keyboards – whether that means quieter typing, longer battery life, or a more satisfying press. German keyboard maker Cherry has unveiled a "bold new chapter" in its history at Computex 2025, headlined by the debut of its first inductive switch and three new additions to its MX mechanical lineup. The announcements mark a clear push to modernize its catalog with improved durability, efficiency, and typing feel across multiple use cases. Leading the charge is the Cherry IK, a next-generation analog switch built on patented inductive sensing. Instead of relying on physical contact for actuation, the switch detects electromagnetic changes, enabling contactless operation. It supports RGB lighting and customizable actuation points, delivering a more versatile typing experience. Compared to traditional magnetic switches, the IK reportedly uses 50 percent less power and just 5 percent of what Hall effect switches consume, representing a substantial leap forward for wireless keyboard battery life. Cherry plans to launch the IK switch this fall. The company also introduced three MX mechanical switches targeting different user needs. First up is the MX Honey, a silent tactile switch tailored for office environments. It's part of the MX2A family and features a reworked spring design with factory-applied lubrication for smoother keystrokes. For fans of ultra-light linear switches, the MX Blossom is now Cherry's lightest mechanical offering, requiring just 35 centinewtons of actuation force. Like the MX Honey, it shares the MX2A's upgraded internals. Finally, the MX Falcon caters to heavy typists and keyboard enthusiasts who want a strong tactile experience. Drawing inspiration from classic keyboards and the distinct click of typewriters, this switch features Cherry's first long-pole stem design, perfect for those who crave more pronounced typing feedback. // Related Stories With these launches, Cherry is diversifying its switch lineup – moving from mainstream mechanical designs toward specialized, high-performance innovations aimed at professionals and hobbyists. All three MX switches debut in prebuilt keyboards this June, with standalone kits expected to follow.

Industrial AI is transforming how factories operate, innovate and scale. The convergence of AI, simulation and digital twins is poised to unlock new levels of productivity, flexibility and insight for manufacturers worldwide — and NVIDIA’s collaboration with Siemens is bringing these technologies directly to factory and shop floors, making advanced automation more accessible. Matthias Loskyll, head of virtual control and industrial AI at Siemens Factory Automation, joined the NVIDIA AI Podcast to discuss how Siemens’ work with NVIDIA is reshaping manufacturing, as the industry hits a turning point. Manufacturing companies are facing a shortage of skilled labor, widening skills gaps as experts retire and increased demand for resilient, efficient production. At the same time, AI advancements offer ways to automate tasks previously deemed too complex or variable for traditional programming — and digital twins open a path to designing and optimizing safe, efficient interactions between AI-powered robots and smart spaces. Siemens’ Inspekto, an AI-driven visual quality inspection system, enables even small manufacturers to automate defect detection in their production lines. Inspekto can be trained in under an hour using as few as 20 product samples, making it ideal for fields like electronics and metal forming. Meanwhile, automaker Audi is using industrial AI in its car body shops, where 5 million welds are made daily. Training AI models to automate weld-spot inspection and integrating them with Siemens’ Industrial AI Suite helped Audi achieve up to 25x faster inference directly on the shop floor, where the defects can be addressed. Siemens is creating an AI-driven vision software enabling robots to handle arbitrary, previously unseen objects. The company is also developing Industrial Copilots with NVIDIA NIM microservices to bring generative AI-powered assistance directly to shopfloor operators and service technicians. Loskyll noted that the Industrial Copilots will run on premises to keep sensitive production data secure while enabling rapid troubleshooting and process optimization. To learn more about the latest in industrial AI, watch the COMPUTEX keynote by NVIDIA founder and CEO Jensen Huang. Hear more from Siemens at NVIDIA GTC Paris, running June 10-12. Time Stamps 1:00 – Overview of NVIDIA’s collaboration with Siemens. 5:00 – Challenges faced by manufacturing companies. 15:00 – How Inspekto makes automated visual quality inspection more accessible. 24:00 – How Audi achieved up to 25x faster inference with Siemens’ Industrial AI Suite. 37:00 – Future directions with industrial copilots and AI-enhanced robotics. You Might Also Like…  Yum! Brands, the World’s Largest Restaurant Company, Advances AI Adoption Yum! Brands, the parent company of KFC, Taco Bell, Pizza Hut and Habit Burger & Grill, is partnering with NVIDIA to streamline order taking, optimize operations and enhance service across its restaurants. Joe Park, chief digital and technology officer at Yum! Brands, Inc. and president of Byte by Yum!, shares how the company is further accelerating AI deployment. Roboflow Helps Unlock Computer Vision for Every Kind of AI Builder Roboflow’s mission is to make the world programmable through computer vision. By simplifying computer vision development, the company helps bridge the gap between AI and people looking to harness it. Cofounder and CEO Joseph Nelson discusses how Roboflow empowers users in manufacturing, healthcare and automotive to solve complex problems with visual AI. NVIDIA’s Jacob Liberman on Bringing Agentic AI to Enterprises Agentic AI enables developers to create intelligent multi-agent systems that reason, act and execute complex tasks with a degree of autonomy. Jacob Liberman, director of product management at NVIDIA, explains how agentic AI bridges the gap between powerful AI models and practical enterprise applications.