• Video showcases China's road-mobile electromagnetic catapult launching an aircraft for the first time
• The footage comes from a now-deleted social media post from the Beijing Institute of Technology's School of Mechanical Engineering
• Leverages China's existing EMALS tech by miniaturizing what it already uses on its aircraft carrier

China seems to have solved one of the biggest challenges modern drones face: deploying without a runway anywhere on the planet, thanks to a miniaturized version of the EMALS catapults found on its newest aircraft carrier.

A short video clip that surfaced towards the tail end of last month, from a social media post by the Beijing Institute of Technology's School of Mechanical Engineering, shows three eight-wheeled trucks linked together in a Lego-like fashion to form a runway that allows a propeller-driven drone to take flight.

The footage also shows all three trucks detached, coupling with each other, and exhibiting all-wheel steering that, in principle, enables them, if given enough space, to launch drones or small aircraft in any direction.

Replacing a runway in the modern battlefield?

China's move is not one that occurred in a vacuum: the United States was the first country in the world to deploy an Electromagnetic Aircraft Launch System (EMALS), sparking an arms race when it launched its first aircraft in 2017 aboard the USS Gerald R. Ford.

China followed suit with its own aircraft carrier, the Fujian, which sports three EMALS catapults to the US's four, possibly accounting for the Chinese carrier's smaller size relative to its principal naval rival's largest.

China has, however, beaten the US to the punch when it comes to demonstrating a portable EMALS, it seems, with its three-truck solution essentially marking a first for either country.

Chinese state-owned giants account for a significant share of the 70+ organizations directly involved in an impressive achievement that sets the stage for the next generation of drone-based combat.

Despite the video making its way across social media, it was taken down by the Beijing Institute of Technology's School of Mechanical Engineering, which originally posted it. The move, however, has not stopped defense forums and Chinese analysts from offering insights into what is clearly a major technological achievement for the country.

With France and India both committed to launching EMALS (albeit on future aircraft carriers for now), only two countries currently have the tech deployed on a ship, and only one of them has a land-based, portable option in play for now: China.

• Lumo 2.0 release rebuilds Proton's privacy-first assistant with reasoning modes, image generation/recognition, cited live web search, and persistent memory
• The privacy stack mixes cryptography and policy: zero-access encryption protects stored chats and images, while inference-time protection relies on Proton's no-logs/no-training promises that have held in the past
• Proton's Lumo 2.0 Lite and Lumo 2.0 Max score 127% and 240% higher than Lumo 1.4 on the Artificial Analysis Intelligence Index, making them close in on last-generation frontier AI models

Proton has revealed Lumo 2.0, its updated AI alternative to ChatGPT, Claude, and Gemini, focusing on privacy first and foremost, a distinctly different approach from most of its competition.

The new update is not only smarter than its predecessor at what it does, but also brings a host of new capabilities: reasoning modes, image generation and recognition, live web search with citations, persistent memory, and customizable assistants.

Lumo 2.0 looks to do all of this while leveraging zero-access encryption, no-logs, no-training, a pitch that makes it appealing to privacy-focused consumers, many of whom are already customers for its VPN product lineup.

Upgrades, multiple models and faster performance

The biggest upgrade to Lumo 2.0 is that it is now multimodal, allowing it to glean information and cross-check a variety of sources without often forcing the user to defer to other AI engines for most tasks.

Proton cites a 76% faster speed for 'everyday queries' while conceding that complex tasks still take a considerable amount of time.

Users can also leverage "Custom Lumos" or purpose-built assistants that retain instructions in memory while still maintaining the encryption promise that Lumo offers, allowing users to avoid starting from scratch each time they have a query to address.

Users can use either the fast, general-purpose Lite model for everyday queries and defer to the more complex Max model for demanding work, or use Fast and Thinking modes, which offer twice the context window of its predecessor for larger workloads and greater coherence with more complex asks.

Pricing spans a free tier for what Proton calls everyday private use, a $12.99-per-month Lumo Plus plan with unlimited chats, Projects, advanced image generation and access to the most capable models, and a $14.99-per-user Lumo Professional tier for teams.

Lumo is also available to business users and offers the same upgrades discussed above, making it a significantly more powerful and smarter AI tool than it was when we last reviewed it at TechRadar.

It is important to note that while Lumo 2.0 is a huge upgrade versus its older 1.4 version, it does not come as close to frontier models as Proton might want to make it appear: its model scores a 51 on the Artificial Analysis Intelligence Index which sees current frontier models clock in as high as 59 (GPT 5.6 Sol Max) or 60 (Claude Fable 5) versus its own comparisions that show it much closer to older frontier models such as GPT 5.5 and Claude Opus 4.8.

This is not entirely surprising, given that one can find the underlying tech Lumo uses in its privacy policy. Proton states that it uses a mix of Qwen 3.5, GLM 5.2, Image-Turbo, and FireRed-Image-Edit-1.1, with GLM 5.2's scores roughly identical to the numbers it cites currently.

Despite its limitations versus newer frontier AI models, Lumo 2.0 arguably remains the most privacy-focused approach to AI available to most end users currently, and it comes considerably closer than its predecessor at what is an increasingly uphill task of late: offering a competitive privacy-centric alternative to billion-dollar proprietary AI models built by the likes of Google, Anthropic, and OpenAI.

• AMD's Versal Premium Gen 2 Memory on Package launch formally ends HBM in its adaptive SoC lineup
• The move represents a ~65% bandwidth cut from the discontinued Versal HBM's 840 GB/s, clocking in at just 288 GB/s
• AMD is painting this as a win, citing better availability, a smaller form factor, efficiency wins, and guaranteed memory supply for the next 15 years

The AI boom may have just eaten AMD's own lunch, thanks to an HBM shortage that forced it to resort to lower-bandwidth LPDDR5x for its Versal Premium Gen 2 Memory on Package offerings.

AMD recently announced the Versal Premium Gen 2 Memory-on-Package family, which leverages up to 32GB of LPDDR5x memory directly onto the chip package.

The move, which effectively cuts bandwidth by 65% compared with previous-generation Versal offerings that use HBM, is seen by many as the need of the hour as memory supplies dwindle amid overwhelming demand.

A needs-based move to lower-bandwidth memory

AMD's move is calculated, even if it comes at a significant performance cost to the chip designer. HBM memory supply is strained, and even within AMD's own lineup, its more profitable (and more demanding) Instinct datacenter GPUs have priority for HBMs' current and future iterations.

AMD is therefore both a beneficiary and a casualty of the same AI demand wave that creates opportunities on one end but constrains supply for other segments, such as its consumer-grade hardware, gaming, and SoC divisions.

AMD's Versal lineup stems from Xilinx, which it acquired in 2022. Xilinx shipped its first on-package memory FPGAs, the Virtex UltraScale+ HBM parts, in 2018 with up to 16 GB of first-generation HBM. The follow-up Versal HBM series, a variant of the Versal Premium line, supported up to 32 GB of HBM2e with 840 GB/s bandwidth.

The problem for AMD isn't just sourcing supply for its new Versal FPGAs, something it calls an Adaptive SoC (System-on-Chip), but the fact that these are extremely long-tail products. In other words, support, dedicated supply, and accessories must remain available to consumers for a long time if they are to embrace and continue working with a particular FPGA class, which complicates matters.

AMD discontinued its last-generation Versal lineup in September 2025, citing HBM2E supply constraints rather than any issues with the chips themselves, and offered no alternatives to customers, stating only that "final orders (LTB) for Adaptive SoC parts will be accepted until June 30, 2026, subject to material availability."

The new Versal lineup effectively addresses this gap, stating that it has a 15-year lifecycle and citing "memory longevity" as the reason for its pivot to LPDDR5X.

AMD's move does bring it some other advantages despite the obvious bandwidth chokepoint: LPDDR5X has better availability than HBM for the foreseeable future, and it also operates at industrial temperatures, whereas HBM tends to be stacked in ways that require advanced cooling. Not only does LPDDR5X run cooler, often passively so in most configurations, but thanks to sporting only 4 memory chips onboard, it is over 60% smaller than comparable FPGAs.

The newer FPGAs will be considerably cheaper to manufacture than their HBM alternatives in current market conditions, and with Chinese memory suppliers such as CXMT also eyeing the same market, Versal Gen 2 might be the long-stay that its predecessor was initially intended to be in a rapidly changing market.

• Xiaomi has launched its first NAS, a dual-base storage solution that comes in three different storage capacities
• The Xiaomi Smart Storage was not planned but was brought into being by customers eager to own one after a mistaken label indicated one was being designed
• The Xiaomi Smart Storage device attracted a reported 30,000 net orders within its first hour of crowdsourcing

Network-attached storage seems to be the new frontier for Chinese heavyweight Xiaomi, which has come a long way from its roots as a software developer focused on a heavily modified version of Android.

While many of Xiaomi's moves when it comes to investments and product lineups often surprise others given the breadth of its offerings, many of which are unrelated to each other, a NAS drive feels like a relatively timid product line to focus on.

The product sees Xiaomi directly compete in an industry that was previously dominated by Synology, QNAP, Ugreen, and Huawei in the region, with users essentially asking it to bring its plans to fruition.

A 'happy accident' for Xiaomi fans

The firm, whose catalog already spans a Nürburgring-bothering electric supercar, rice cookers, nose-hair trimmers, electric scooters, and the phones that started it all seems to have stumbled upon the possibility of having its own NAS, now known as 'Xiaomi Smart Storage', purely by accident.

The idea surfaced by accident in May 2025, when a schematic labeled "10G NAS" appeared in promotional imagery for the company's networking-switch range. Chinese consumers, however, reacted loudly enough that Xiaomi's ecosystem general manager, Chen Bo, publicly committed to building one, delivering a finished, crowd-funded option roughly 13 months later.

Xiaomi's crowd-funded NAS offering comes in 3 different sizes or configurations: the entry-level 4TB SKU (¥ 2299), the mid-range 8TB SKU (¥ 2899), and a top-of-the-line 16TB SKU (¥ 4699). All of these options include the dual-bay Xiaomi Smart Storage with two equal-sized hard drives.

The NAS comes fairly well-equipped, offering USB 3.0, HDMI, a 2.5-gigabit Ethernet port, and support for both 2.5-inch and 3.5-inch SATA drives, and capable hardware reportedly under the hood (a Realtek RTD1619B, a quad-core Arm Cortex-A55 clocked to 1.7GHz, with 2GB of DDR3L and 8GB of eMMC).

Xiaomi seems equally committed on the software side of the spectrum; It has also published a companion app on both Apple's App Store and Google Play, while offering support for its Mi Home ecosystem application from the get-go.

This has allowed the company to rack up an impressive 30,000 orders in the first hour the NAS drive went live on its crowdfunding site as consumers buy into Xiaomi's brand value and ecosystem promise alike, even as storage prices continue to head up thanks to AI-centric demand.

• Pentagon science advisor Joseph Jewell weighs in on the Ukraine war and integrating AI and biotech mark a paradigm shift in how the US military sees future conflicts
• Marines recently 3D-printed shaped charges from coconut husks and coffee grounds that beat conventional explosives by 25%
• Department of Defense is also trying to speed up innovation in the industry by offering as many as 500 patents free of charge to private companies

The Russia-Ukraine war has been devastating in its impact for those directly affected by the conflict, which has now entered its fifth year as both sides trade blows in what many feel is a prolonged stalemate stemming in part from a lack of manpower.

The conflict, or rather its asymmetrical nature, however, has many modern militaries keeping a sharp eye on events as they showcase what the future of combat might look like between two warring nations.

The US Assistant Secretary of War for Science and Technology, Joseph S. Jewell, recently spoke at length during the Defense One tech summit about how information from the conflict and inroads in AI and biotechnology continue to shape modern warfare as we know it.

Production inroads, AI integration and a focus on biotechnology

The summit saw the Assistant Secretary address a host of topics, including insights into how the Russia-Ukraine conflict is playing out and the lessons the US must learn from the ongoing war.

He addressed the fact Ukraine essentially willed its entire drone industry into existence because it was key to its survival while essentially keeping Russia's navy at bay for most of the conflict despite not having a similarly equipped fighting force at sea.

The line that carries from him, however, might be about Marines repurposing coffee grounds and coconut husks to make 3D-printed shape charges for the battlefield.

This underscores an important change that has already taken place on the modern battlefield, as researchers and military personnel increasingly push the boundaries to find the best way to resupply and rearm while also making advances in lethality in some cases.

Coffee grounds and coconut husks were only the tip of the iceberg; the Marines also attempted the same with plastic water bottles and even crushed volcanic rock, noting that the latter worked best.

Jewell said the field-made charge had cut the time to point-of-use by 99%, because it could be produced on the spot from materials "endemic in the Indo-Pacific," and even more interestingly, it showed "25% better focusing characteristics than conventionally manufactured high explosives."

The underlying story isn't just about how field-made charges are saving an enormous amount of time and money on future battlefields, but how military doctrine has changed to include 'patent holidays' to increase access to tech and foster innovation, for better or worse, at least on the Pentagon's end over the last few years.

While his focus was on Ukraine, similar lessons have been learned by the US in its on-again, off-again conflict with Iran, where the latter has resorted to using low-cost but high-volume weapons, including drones and missiles, to effectively force a stalemate over the Strait of Hormuz, indicating that a paradigm shift is in order where the side with the best, or most 'efficient' weapons does not necessarily win.