The Register

COMPUTEX 2026 It’s hard enough for startups to compete with AMD and Nvidia on chip design. The rise of rack-scale architectures has only made things harder. Companies not only have to invest in chip design but also the mechanical, thermal, and power engineering necessary to pack six dozen or more AI accelerators into a single rack that functions as one enormous GPU. At Computex last week, Delos Data, a startup funded by former Intel and Barefoot Networks execs, showed off a modular server platform aimed at giving chip startups a shortcut to rack scale. One of the challenges with the move to rack scale is actually the sheer amount of networking that needs to be enabled at the box. A typical eight GPU HGX node only needs one or two ports per GPU. By comparison, a GB300 NVL72 needs 18 400 Gbps ports per GPU. Nvidia and AMD have developed custom racks with integrated backplanes, power delivery, and cooling. Delos by comparison is keeping things relatively simple by designing a chassis that, at least from the front, looks more like a switch than a GPU server. It features 36 OSFP ports, nine for each of the four OAM sockets at the heart of the system. OAM, if you’re not familiar, is an open socket commonly used by high-performance accelerators requiring more interconnect bandwidth and power delivery than standard PCIe cards can manage. Assuming 200 Gbps SerDes, that works out to 3.6 TB/s per chip of interconnect, the same as Nvidia's new Rubin GPUs. OSFP means that customers can use standard DACs or pluggable transceivers, and switches depending on how large they want their scale-up domain to be. And while OSFP is usually associated with Ethernet, you can run just about anything you want through them, whether it be UALink, Ultra Ethernet, PCIe, or something else. From a deployment standpoint, these systems would be wired up like any other hyperscale system, just a whole lot denser. Delos isn’t the only option out there for chip startups looking for scale up reference design. AWS for example appears to be repurposing Nvidia’s MGX form factor for its Trainium 3 rack systems, while AMD’s Helios rack is now an OCP standard. Both designs would, in theory, be easier to service, but Delos argues that its modular design offers greater flexibility. “It makes it a little bit more flexible in terms of, maybe you want a scale up domain of 100 or maybe you want it a scale up domain of one,” CTO Dan Daly told El Reg. “It just depends on how many cables you want to plug in. This also allows you to go plug into different types of switches… it could be simpler switches, maybe even optical circuit switches (OCS).” Using existing packet switches from Broadcom or Marvell, such a design could support 512-1,024 accelerators in a single layer fabric depending on whether you're using 200 Gbps or 100 Gbps SerDes. Using multi-layer fabrics, OCS, and/or 2D/3D toruses, the compute domain could scale even further, all while using off-the-shelf components. While OSFP keeps things simple and easy, it also means power consumption could become problematic for larger compute domains requiring pluggable optics. In fact, this is why Nvidia has taken so long to embrace optical scale-up. Copper may not have the reach, but it uses a fraction of the power. Delos CEO Ed Doe tells us the company is already exploring versions of the system that will use near package or co-packaged optics out to MPO-style connectors rather than the OSFP. The startup isn't just doing hardware. As anyone who's done large scale networking knows, the physical and logical topologies — that is, the way devices communicate with one another on the network — can look very different depending on the workload. Delos has developed a software orchestration platform designed to facilitate the configuration and monitoring of these switched fabrics or meshes in order to enable dynamic rerouting of traffic in the event of a link failure. At Computex, this software platform, which Delos has dubbed its Nonstop AI network, was on display, allowing attendees to pull links at random and see the network react and correct itself automatically. The company's ambitions don't stop at network orchestration and systems. We're told Delos has additional products in the works, and we don't know for sure what they are, but a high radix switch design built atop merchant silicon would certainly complement its Nonstop AI systems. ®

ON CALL 你好 Nǐ hǎo, dear reader, and welcome to another installment of On Call, The Register's Friday column that shares your stories of translating technical trauma while delivering transcendent tech support. This week, meet a reader we'll Regomize as "Jackson" who told us about his time providing tech support in a university's biology department. "It was sometime in the mid-2000s and our IT group at the time consisted of myself, my boss, and a part-timer," he told On Call. "We were a very casual IT group; nothing in the way of any formal policies or standards for anything at all. If someone needed a new PC, we just ordered parts and assembled them ourselves." The department's PC fleet therefore had a diverse gene pool, with no two machines possessing the same bill of materials. "This was fine by me – I enjoyed building them and it never really caused any issues that I couldn't handle," Jackson told On Call. "Until one day we got a panicked support call from one of the secretaries who claimed that her PC just rebooted and then started talking to her." Jackson and his colleagues didn't believe a word of it until the secretary stopped talking and placed her phone next to the talking PC. "I could clearly hear a muffled voice repeating a message of some sort," Jackson told On Call. There was nothing for it but to visit the PC, which he found hung in the middle of a Power-On Self-Test, flashing an alphanumeric error code and unmistakably playing a voice through its internal speaker. In Chinese! Jackson rebooted the machine and it ended up in the same state, reciting the same message. Chinese isn't a language in which Jackson is fluent, so he had no idea what the PC was trying to tell him. "After poking around in the BIOS, I found the culprit," Jackson revealed. "This particular model of motherboard had a 'talking error BIOS' whereby certain POST codes triggered the playback of a friendly, spoken error message, with Chinese set as the default language." Jackson found the relevant BIOS settings, changed the default language to English, and the next time he rebooted the machine it helpfully let him know: "Your floppy drive may not be connected properly." In his mail to On Call, Jackson hypothesized that the PC's CMOS battery died, so the BIOS was unable to access its stored settings and reverted to factory settings that assumed the presence of a nonexistent second floppy drive. "It triggered a feature I didn't even know the motherboard had!" Jackson told On Call. Have you found yourself flummoxed by a feature you didn't know about? If so, click here to send On Call an email – we'll assume that's a feature you know well – so we can tell your story on a future Friday. ®

Enterprises that have watched Claude claw its way toward mass appeal over the past few months of capacity challenges and pricing realignment should take a closer look at Anthropic's offerings, according to International Data Corporation (IDC). The tech consultancy has been tracking Anthropic's moves over the past six months and says that the AI biz is taking credible steps toward making itself an enterprise AI provider. "Currently, no frontier model company is mature enough to be evaluated as an enterprise AI provider on its own," IDC said in a recent report. "But Anthropic is running at full speed to get there before its competitors." The report is titled "The Transformation of Anthropic (and What to Do About It)," and advises enterprises to revisit their LLM and agent evaluations with an eye toward seeing whether Anthropic might work out as a reliable technology provider. Enterprises, IDC says, remain largely unsold on Anthropic's Claude models, with only 19 percent using them extensively and 25 percent actively evaluating them. OpenAI and Google are better represented in enterprises, with about 42 percent and 38 percent of organizations using their respective products, per IDC's FERS Survey, March 2026. According to The Information, about 86 percent of Anthropic’s 2025 revenue was projected to come from enterprise sales. OpenAI, the report claims, derives just 40 percent of its revenue from business sales, though that figure ($5.2 billion) represented a higher dollar amount than Anthropic's business revenue ($3.9 billion) at the time. That was back in January, only two months after Anthropic began shifting enterprises away from seat-based pricing toward usage-based pricing. Since then, IDC says Anthropic has taken a series of steps to make itself more credible as an enterprise AI provider. "This conclusion might not be obvious: From January through May 2026, Anthropic produced well over 100 public interactions, including official announcements, release notes, blog posts, X posts, partner announcements, hiring news, policy moves, and press-covered transactions," the report says. These initiatives, such as the launch of the Claude Partner Network, have expanded distribution, bolstered brand perception, facilitated future growth, enhanced "stickiness" (aka lock-in), strengthened enterprise support, addressed the needs of specific industries, demonstrated innovation, and shored up the compute supply necessary to deliver services at scale. According to IDC, the enterprise ecosystem commonly focuses on a vendor-neutral, multi-LLM strategy. Nonetheless, the biz argues that the company has made its technology visible enough that Claude is increasingly coming up in conversations among IT decision makers. "Anthropic's transformation has just started, but the direction is clear enough for CIOs and CISOs to pay attention and reassess where Claude fits in a multi-LLM or an agentic AI Strategy," the IDC report says. ®

Palantir CEO Alex Karp doesn’t think frontier AI labs prepping for IPOs really understand what their customers need, and that ignorance is making Palantir a success. Karp had a wide-ranging, often rambling and self-interrupting sit-down (coherent compared to some of his other interviews, to be fair) with CNBC’s Sara Eisen on Wednesday in which he said that every single enterprise customer Palantir has is unhappy with frontier AI labs like Anthropic and OpenAI. Those companies, says Karp, are operating on a “hyper religion of hyper optimism” that doesn’t reflect the experiences of their customers. “They believe all problems present, past, and future, including the ones they create but don’t acknowledge, are going to be solved by them,” Karp opined. “Enterprises are fed up because they know this doesn’t actually work this way, and isn’t working.” That frustration, Karp said, is driving businesses to Palantir’s Foundry systems, which act as AI-agnostic data integration platforms for unifying disparate data sources and cognizing them with whatever LLMs a customer chooses to deploy. Pitch to prospects or not, Karp is on to something. AI projects are largely loss makers for the companies that deploy them, and have been for some time. Only 28 percent of AI use cases fully meet ROI expectations, according to a recent Gartner estimate, and most fail to ever get out of the pilot stage. Despite that, business leaders keep shoveling coal into the AI furnace to try to extract value, which, if you ask Karp, simply isn’t there unless you’re pairing those models with some decent infrastructure. Infrastructure Palantir can provide, natch. “It’s not just the man and woman on the street who are unhappy with the frontier labs,” Karp said, pointing to “every single enterprise we deal with” being frustrated with the likes of Anthropic and OpenAI’s ability to provide value for their businesses. Karp said that Palantir leadership has been debating whether they should pay potential customers to go talk to frontier labs themselves before signing a contract with his outfit. “People come out of there screaming, saying 'this could never work for me, they don’t understand the enterprise, they don’t care about my enterprise,'” he said of customers. Frontier labs, Karp opined, just want customers to "tokenmax” – that is, to view token consumption as a measure of productivity and usefulness. The charge isn’t out of left field. Google CEO Sundar Pichai even nodded to the phenomenon at I/O last month. Burning more and more tokens is getting to be expensive for companies, and OpenAI is reportedly considering reducing its per-token charge to attract more customers in its growing war with Anthropic, which Karp called the “leading frontier firm” in his interview. Karp wouldn’t give a straight answer when asked whether OpenAI, Anthropic, and other frontier labs could do what Palantir is doing, but he did imply some doubt. Sure, they have some good engineers on staff, he said, but that doesn’t matter a lick if they “don’t talk to the enterprises or understand the technical challenges” their customers are facing in deploying their models. “When you go to San Francisco and talk to them, their basic vibe is ‘we don’t have to solve your problem today because tomorrow you’re going to go away and all your problems are going to be solved,’” Karp charged. “It’s largely religious.” Karp also called out OpenAI’s recent agreement to acquire UK-based AI consulting firm Tomoro, which will form part of the newly launched OpenAI Deployment Company aimed at helping customers generate returns from their ChatGPT investments, as an attempt to replicate Palantir's success. “It’s a complete farce,” Karp said. “They don’t understand how unlikeable they are.” By that, Karp said, it’s not that AI lab leadership isn't friendly – he said he's buddies with some of them and that they’re great to chat with – but “the product doesn’t actually work and it’s very expensive.” To that end, he added, most of the things that Anthropic brags about in public, for example, are successful because they’re “running on Palantir,” Karp charged. “It is not that LLMs aren’t crucial for the world, it’s just that the implementation is where the value is, certainly in the next 7 years,” Karp explained. In essence, what the Palantir boss seems to believe is that simply tossing an LLM at business problems isn't an actual solution. What Karp had to say on CNBC was, in his usual way, boisterous, confrontational, and self-aggrandizing, but look at the rate of AI returns in the enterprise right now and you have to admit he's got at least a partial point. ®

AI may or may not be pushing lots of people out of the workforce, but Anthropic has good news as the Claude creator is creating temporary positions to promote the adoption of AI, even as CEO Dario Amodei ponders policy interventions to counter "job displacement." The AI biz has announced the launch of Claude Corps, a $150 million program that will pay 1,000 Claude Corps Fellows $85,000 (plus benefits and a token budget) for one year to help advance the missions of nonprofit organizations using generative AI. Meanwhile, the tech industry continues to take on debt to build datacenters while balancing its books by shedding employees. According to job search biz TrueUp, the tech sector this year has averaged 935 layoffs per day, up from 674 per day in 2025. Anthropic's program debuts alongside the publication of Amodei's latest musing about his optimism "that, even in a world with AIs that are better than everyone at everything, humans can live lives of deep purpose and strive to build awe-inspiring and beautiful things." Claude Corps' stated goal is to provide host organizations with valuable tools and systems and to help participating fellows "build AI skills that will serve them in their careers" – however long those careers last until AIs are better than everyone at everything. There is, of course, no guarantee that AI will surpass human cognition or folly. But Amodei likes to talk about the idling of human labor, just in case, even if that sort of chatter fuels the firebombers. Anthropic says that it is announcing Claude Corps alongside its policy framework for dealing with AI's impact on work. The framework is titled "Policy on the AI Exponential," which is the same title Amodei used for his post. The policy's call for company-endorsed regulatory intervention is predicated on the claim that "AI is advancing at exponential speed," though the document cites no evidence of exponential capability gains and offers no time frame – a necessary variable to calculate periodic gains. Judging by AI model benchmark metrics, recent AI improvement has been incremental, a rate of advancement too timid to turn heads in the attention economy. Using data from Stanford HAI's 2026 AI Index report, even impressive gains such as AI model performance on the SWE-bench Verified benchmark rising from 60 percent to nearly 100 percent of the human baseline in a single year are not, by themselves, evidence of broad "exponential" progress across AI. Alarmism aside, Claude Corps will be funded and steered by Anthropic and implemented by computer education nonprofit CodePath, which will serve as the employer of record for fellows. The 12-month-long fellowships begin with "intensive training on using Claude in non-profit settings," augmented by five hours of additional training each week. Fellows are expected to use their remaining time coaching their respective nonprofits on the ins and outs of AI workflows. The gig comes with support from a CodePath mentor and office hours from Anthropic, which may prove useful for reactivating Claude accounts that have been suspended after triggering Claude's overly sensitive safety guardrails. Some 400 nonprofits are expected to host Claude Corps Fellows over the next 12 months, including Braven (job prep for low-income students), Code the Dream (coding education), and Heartland Forward (economic growth for middle America). "If Claude Corps works, we'll have a foundation for something much larger: a model for widening AI's benefits during a period of vast economic change," Anthropic says. And if not, as New Yorker cartoonist Tom Toro put it, "Yes, the planet got destroyed. But for a beautiful moment in time we created a lot of value for shareholders." ®

Data theft and extortion group ShinyHunters claims to have exploited a critical Oracle PeopleSoft bug as a zero-day to compromise more than 100 organizations, including the University of Nottingham, across 300 vulnerable instances. A spokesperson for the cybercrime crew on Thursday told The Register that they exploited CVE-2026-35273 to break into the university’s PeopleSoft system and steal 40 GB of personal data and billing records belonging to hundreds of thousands of current and former students. ShinyHunters posted the UK university on its data leak site on Tuesday before publishing the stolen files later that same day, presumably because the school refused to pay the extortion demand. “University of Nottingham on our leak site is one of the first publicly confirmed incidents,” a ShinyHunters spokesperson told us. “We have only just started outreach to affected orgs and are actively looking to reach an agreement with affected orgs.” They didn’t say when they planned to post the other 100 or so claimed victims. PeopleSoft is a widely used enterprise software suite that large corporations and institutions use to manage their human resources, payroll and billing applications, supply chains, and student records. CVE-2026-35273 is a 9.8 CVSS-rated vulnerability that allows remote, unauthenticated attackers with network access via HTTP to compromise PeopleSoft Enterprise PeopleTools and fully take over the platform. On Wednesday, a day after ShinyHunters leaked the school’s data, the University of Nottingham confirmed the breach and Oracle issued an out-of-band security alert. It’s unclear, however, if the software provider has issued a patch to fix the security flaw. The Register reached out to Oracle, and did not receive any response to our questions. Google-owned Mandiant Chief Technology Officer Charles Carmakal, in a brief LinkedIn post on Thursday, warned that PeopleSoft was one of two zero-day vulnerabilities “actively being exploited in the wild.” “Oracle released mitigations,” Carmakal wrote. “Patches should come soon.” The other zero-day, for the record, is this Cisco Catalyst SD-WAN Manager vulnerability.®

The boffins on Google’s DeepMind team unveiled an experimental new language model this week that uses techniques originally developed for AI image generators to boost text output performance by as much as 4x when running on resource-constrained consumer hardware. It's free to download and you can run it with just 18 GB of DRAM or VRAM. The model, codenamed DiffusionGemma, is the latest addition to Google’s open weights model family. But unlike Gemma 4, which launched this spring, the 26 billion-parameter mixture of experts (MoE) model isn’t a large language model in a conventional sense. Instead, it’s actually closer to image models like Stable Diffusion or Flux. Rather than generating tokens one after another in an autoregressive fashion, DiffusionGemma generates entire paragraphs' worth of tokens at the same time. The process looks a lot like how a diffusion model turns what’s essentially static into an image through a series of denoising steps. As Google explains it, DiffusionGemma works by laying out a canvas of random tokens, and then refining them until the final output is reached. Compared to conventional LLMs, which are memory-bandwidth bound and require a lot of VRAM, diffusion models are a predominantly compute-bound workload, which is why the Chocolate Factory is positioning these models for local deployment. LLMs are autoregressive. During token generation, the model’s active parameters need to be streamed from memory for every token generated, making memory bandwidth a major bottleneck. In the cloud, inference providers balance compute and memory bandwidth by processing hundreds or thousands of requests in parallel. As you might have guessed, this isn’t something the average user running a local model on their notebook can do. However, many consumer products, like high-end graphics cards, have plenty of excess horsepower, which DiffusionGemma can take advantage of to boost output performance. Diffusion language models aren’t perfect. Google isn’t the first to explore this tech. Previous models, like DREAM or Mercury 2, demonstrated major speedups over conventional LLMs, but generally underperformed them in benchmarks for their size. DiffusionGemma doesn’t appear to be any different. According to Google, the 26 billion-parameter model falls just behind Gemma 4 12B in the GPQA-Diamond benchmark, with its main advantage being output speed, and even then it’s not as impressive as Google has made it out to be. The chart shows a roughly 2.25x speedup for DiffusionGemma over the 12B parameter LLM with speculative decode enabled. Compared to Gemma 4 26B-A4B, the speedup is nearly 4x when running a single Nvidia H100. DiffusionGemma is being released as an experimental model rather than an enterprise focused one, like we saw with Gemma 4. The model is available for download on popular model repos like Hugging Face under a highly permissive Apache 2.0 license with support already merged into popular inference engines like vLLM, MLX, and HF Transformers, with support for Llama.cpp coming soon. While local inference has largely been the domain of AI enthusiasts, companies like Google are increasingly leaning on the tech to cut cloud costs associated with their AI services. As you may recall, back in May, Google quietly began shipping a small LLM with its Chrome web browser. ®

Nightmare Eclipse, the prolific zero-day vulnerability hunter with an axe to grind against Microsoft, released yet another exploit late Wednesday that the researcher claims will spawn a command prompt that provides total access to the BitLocker volume. This bug, called GreatXML, was “an accidental discovery,” according to the researcher, who said it only took four hours to find. They claim this exploit (published on GitHub and Git-based code-hosting platforms) can bypass BitLocker on any system that has ever run a Microsoft Defender Offline scan at any point in the past. GreatXML comes just a day after Nightmare released exploit code for RoguePlanet, which allows local privilege escalation and leads to SYSTEM-level control over an affected machine. This brings the researcher’s zero-day count to eight. The earlier six - RedSun, UnDefend, BlueHammer, YellowKey, GreenPlasma, and MiniPlasma - all have patches as of this week’s Patch Tuesday event. Redmond on Wednesday told The Register that it is aware of RoguePlanet, and “actively investigating the validity and potential applicability of these claims.” The Windows giant didn’t immediately respond to our inquiries about GreatXML, including when it planned to issue a patch. Microsoft has said none of the vulnerabilities were reported via its official channels prior to being made public. The company also banned Nightmare’s earlier GitHub account, and seemingly threatened legal action before dialing back its rhetoric after steep backlash from the security community. Nightmare Eclipse, who some researchers suggest is an ex-Microsoft employee, harbors a very personal grudge against the Windows giant and its communications with bug hunters. They have promised to keep the zero-days coming, but waffle on the timing. Last month, the researcher pledged a big July 14 drop: “I will make sure your bones are shattered that day,” and then added, “nothing will be released this June (or maybe I will release smtg, depending on circumstances).” On Tuesday, they changed course. “I will be unable to mass disclose zerodays in July 14th, RoguePlanet took way more time than expected and truly drained me. I might take a break but I can't say for sure what I will be doing for next month, maybe it's nothing, maybe it's smtg.” A day later, Nightmare released the “accidental” GreatXML BitLocker bypass. According to the researcher, the BitLocker bypass first requires copying “unattend.xml” and the “Recovery” directory to the root of the recovery partition. The next step is rebooting into WinRE by Shift-clicking Restart. “If everything was done correctly, a shell with unrestricted access to the bitlocker volume will spawn,” Nightmare wrote. Also, if the scan hasn’t even been initiated on the Windows system, first you’d need to either log in and initiate it, or “figure out a way to boot into WinRE in offline scan state.” Security sleuth Will Dormann followed Nightmare’s steps to reproduce GreatXML, and said the writeup seems “flawed.” In his testing, Dormann said the command prompt appeared the next time a Defender Offline scan ran. “And in order to trigger a Microsoft Defender Offline scan, you both need to be logged in to Windows, and also have admin credentials,” he wrote on social media. “And if you've already got that level of access, you can just turn off bitlocker.” “The writeup for GreatXML suggests that the prerequisite is that Windows Defender Offline has been executed at some point in the past,” Dormann added. “And that after planting two files in WinRE, all you need to do is [Shift]-reboot into WinRE, and Windows will automatically go into Microsoft Defender Offline scan mode. But this is not the case in any of the 3 lineages of Win11 that I have handy.” ®

Datacenters got you down? Worried that even the most innocuous questions will spin up AI models running in water-guzzling, energy-sucking, planet-destroying hyperscalers? You need CrankGPT. No, we’re not talking about surrendering to AI psychosis: we’re talking about a literal hand-cranked machine loaded with a voice agent that can respond to questions and even translate speech into other languages, provided someone keeps the power flowing. There’s an onboard custom-built capacitor board to store some juice, mind you, but it only provides around 20 seconds of crank-free runtime before you’ve gotta keep crankin’ to keep it alive. That, and it takes a bit of time to get it running - according to the documentation website, it’s a 30-second process “from the moment you start cranking to the moment you’re having a conversation with CrankGPT.” According to the AI expert duo behind the device, computer scientist Katrin Tomanek and former Google Advanced Technology and Projects Group technical project lead Alex Kauffmann, CrankGPT still delivers impressive results despite the need to perform some hard physical labor for your tokens (though we’d argue some exercise for your AI might not be a bad thing). “Asking Claude to add two numbers for you is like swatting a fly with a wrecking ball,” Kauffmann told The Register in an email. This tongue-in-cheek demonstration, Kauffmann said, may be a bit of light fun, but it’s an exercise in demonstrating what his and Tomanek’s AI company, Squeez, is all about: small, private specialized AI models that, in a pinch, might not even need very much energy or a connection to the web to operate. “Squeez produces customized, efficient, and private models that can run on small, inexpensive hardware to solve specific problems,” Kauffmann explained, citing tasks like voice recognition for someone with a strong accent or speech impediment, or specially-trained, local AIs that are subject matter experts in topics like gardening or auto repair, but won’t touch subjects outside their wheelhouse. Contrary to the flashy dot-com for CrankGPT the pair have set up, Kauffmann told me, Squeez has no plans to pursue spin cycle class-powered AI stacks for dev teams, though he said if anyone wants to foot the bill, he'd be happy to give it a shot. "Off-the-shelf bike generators are shockingly expensive and they're fussy to build," Kauffmann said. Still, "a good biker can maintain a steady 120W output, so a class of twenty could power a Blackwell." Speaking of wheelhouses, what’s inside that box? If there’s a tiny computer in a 3D-printed box with a crank attached, there’s a good possibility it’s going to be a Raspberry Pi, and that’s the case here. CrankGPT’s brain is built on a stock RPi 5 with 8 GB of RAM and a cooling fan HAT, and audio input and output are handled by a dedicated I/O HAT designed for voice assistants running RPis. Power comes from the aforementioned crank, which is actually an off-the-shelf 20W switchable voltage hand crank unit built for emergency USB device charging, and is stored in the custom capacitor unit the duo built. “The neatest part of the whole thing is that you can actually feel the inference,” Kauffmann told us. “The amount of resistance the crank presents varies depending on the amount of work the board is doing, so when it's really working (generating words for instance), the crank becomes much harder to turn than when it's idling waiting for you to say something.” As for software, the device is running the most stripped-down, bare bones instance of DietPi the pair could compile, which is able to boot into a functional userspace in about three seconds. The voice agent is the truly original piece of work done for the project, as detailed in the documentation page, and was built entirely from scratch. “We wanted to understand the system end to end and have as few dependencies as possible,” the documentation page notes. It’s available on GitHub for those interested in trying it out. Speech recognition is handled by the Moonshine automatic speech recognition engine, chosen for its speed, while text-to-speech synthesis is handled by Piper, chosen again for its low-resource edge inference capabilities. As for the models running on the thinking itself, there are a few that are behind CrankGPT, with Liquid LFM2 1.2B providing a general-purpose voice agent, and Gemma 3 1B being used for translation. CrankGPT can switch between translation and various prompts (e.g., general question answering and games like two truths and a lie) via a knob on the side of the enclosure. “It’s entirely configurable,” Kauffmann told us. “We added a couple of physical inputs (the knob, a button, a switch) to make experimentation easier.” Kauffmann added that he and Tomanek were surprised by how well the translation function worked. “We did no fine tuning, it's just a two-line prompt and it works really well for high-coverage languages,” he explained. While the demonstration focuses on audio prompts and responses, Kauffmann explained that the device supports all sorts of different models, with the only real limitation being inference time and the amount of hand cranking one wants to do to get their response. “We’ve generated images (small), made poetry (bad), and written code using the same setup,” the CrankGPT makers wrote in their documentation, all with “a hand crank, a little computer, and a small stack of speech and language models running locally.” If you’re interested in building your own CrankGPT model, keep an eye on the documentation page we linked earlier in this story, as Kauffmann told us he and Tomanek are planning to release all the plans and schematics in the coming days, while the aforementioned custom voice agent is already available for tinkering. “It's a pretty straightforward setup, the only tricky part is that SBCs like the Raspberry Pi will sometimes draw enough current to trigger a little generator's overcurrent protection,” Kauffmann told us. If you have a spare $300 lying around (that’s what Kauffmann estimates the RAM pricing surge has driven the build cost up to, from the $150 he spent when building CrankGPT last year), then you, too, may soon be able to build your own completely off-grid, standalone AI box so you can keep chatting with your favorite micro LLM if and when its bigger cousins knock the grid offline. ®

Amazon, along with the rest of the industry, has gotten so used to framing everything that happens through the context of AI that it has lost the plot on their Graviton chip lineup, and along with it their own credibility. Which is a shame, because it's actually a triumph of a chip. First, the Wall Street Journal breathlessly reported that Snowflake's $6 billion AWS commitment was "for agentic computing chips." Then AWS's own press release heralded the release of their latest chips "for the Agentic AI era." In both cases, they were referring to their Graviton line. You could be forgiven for thinking this was some kind of GPU. No, that's Trainium. (Technically, Trainium isn't a GPU, nor is it a CPU, but rather a systolic array. Don't worry; most AI engineering software doesn't know what the hell that is, either.) Graviton is AWS's general purpose Arm CPU, which can be used for AI in much the same way as Excel can be used as a database. But that's far from its only, or even primary, purpose. Let's dive into what Graviton actually is. Price / Performance / Reality For the longest time, Amazon refused to issue benchmarks, competitively positioning its then-nascent Arm line against Intel. Many of us thought this meant that the results would underwhelm — so you can imagine my surprise when real-world workload tests showed 35 percent to 40 percent better performance in a wide variety of situations. It was as if Amazon had built something amazing, but was somehow embarrassed to admit it. Those days are long behind us; they trumpet in the subhead of their announcement that Graviton 5 means "apps run 35% faster, ML inference is 35% faster, and databases are 30% faster." To their credit, I was expecting those numbers to be against something ancient, but in a refreshing bout of honesty, they're comparing them to Graviton 4, itself no slouch. They are also 9 percent more expensive. Once upon a time, new generations of AWS instances were notably less expensive than their predecessors. Going from a c4.large to a c5.large meant you'd get better performance, and the instance itself was a whopping 15 percent cheaper. Upgrading was a no-brainer! That started changing, and now upgrading means the instance becomes more expensive. AWS's position is that this is an incomplete analysis, since the improved performance means you'd pay less for a given workload. In some cases, this is correct, but in others, it's akin to saying that a Ferrari offers better price performance than my Honda CR-V because I can drive it to work three times faster. Logic, as well as traffic lights, disagree. Amazon's contention is correct for customers who have large fleets of nodes that they run at high degrees of CPU utilization. Switching those fleets to the new hotness will absolutely result in a price performance improvement, provided the workload and the stars both align. However, for customers who need a fixed number of nodes (think database companies, who offer each customer of theirs a set number of replicas, or workloads of the form "each environment gets three nodes, one in each AZ"), this represents a pure 9 percent price hike going from old generations to new ones. That puts many customers in a pickle: upgrade to new instance families, or stay on the old ones and watch availability become constrained in the coming years as AWS stops racking old chips. (Hi, Amazon PR! If you're about to pop into my inbox to tell me that won't happen, I have a customer I'd love for you to have a chat with!) But this price hike isn't happening in a vacuum. It's happening against a backdrop of "an 8GB Raspberry Pi is now $175, over twice its launch price of $85." Components have become fiendishly expensive across the board as giant companies compete for capacity, and AWS has to be feeling that pressure. Two companies each asked to buy all of AWS's Graviton capacity for the year; AWS clearly has room to kick their prices into the stratosphere! Somehow, they're not only resisting the siren song of "please gouge me, business daddy," but also managing to keep availability strong for customers of all stripes; I upgraded my developer node in my tiny unremarkable AWS account yesterday, and it Just Worked. And so... Despite the nonsense marketing, I don't want to detract from just how amazing Annapurna Labs (Amazon's chip division) has been at churning out wildly performant silicon year over year. Their chips are legitimately great, and the Graviton 5 numbers are a triumph. Lost against the backdrop of "Agentic AI," the stuff underpinning all of it continues to work, improve, and largely pass by unremarked. Keep going. ®

ZTE has won three prestigious awards at Selular Award 2026, held on June 8, 2026, at Menara Peninsula Hotel, Jakarta. The awards recognize ZTE's contributions and innovations in advancing artificial intelligence (AI)-powered network technologies amid the acceleration of digital transformation and 5G development in Indonesia. ZTE's contributions to advancing AI-powered network innovation have been recognized by Selular Media Network (SMN), a leading telecommunications and technology media organization in Indonesia, through three awards at Selular Award 2026. ZTE received honors in the categories of Best AI Technology Fixed Wireless Access, Best AI Network Ecosystem, and Best Native AI Baseband. These awards reflect ZTE's capabilities across network access, ecosystem development, and core infrastructure, further strengthening its position as a technology partner supporting digital transformation and the evolution of AI-driven networks in Indonesia. The Selular Award is an annual appreciation program organized by Selular Media Network (SMN) to recognize outstanding achievements and contributions across Indonesia’s ICT and digital technology industry. As the first and most consistent telecommunications industry award since 2003, the Selular Award serves as a benchmark for excellence, honoring companies and brands that demonstrate innovation, strong performance, and meaningful contributions to Indonesia’s digital transformation. Through this award, the public and business community can identify industry leaders that continue to create value and drive progress in the digital ecosystem. This year's Selular Award carries the theme "Leading The Future: Building Exponential Value in 5G-Advanced and AI Economy", highlighting the convergence of AI and 5G-Advanced as key drivers of digital economic growth. Kevin Fang, Marketing Director of ZTE Indonesia, said: "Digital transformation today is no longer driven solely by connectivity, but also by the ability of networks to operate more intelligently, efficiently, and adaptively. Through the AI-powered innovations we have developed—from broadband access to core infrastructure—ZTE is committed to delivering network solutions that are ready to meet connectivity demands in the AI and 5G-Advanced era. These awards motivate us to continue delivering meaningful innovations that create value for the industry, our customers, businesses, and society." Indonesia's telecommunications industry is currently entering a critical phase in its digital transformation journey. According to the e-Conomy SEA 2025 report by Google, Temasek, and Bain & Company, revenue from AI-powered applications in Indonesia grew by 127% year-on-year, the highest growth rate in Southeast Asia, with 80% of users interacting with AI applications daily. This momentum reflects the growing demand for network infrastructure that is not only fast and reliable but also capable of supporting AI workloads. On the infrastructure side, GSMA Intelligence projects that 5G investment in Indonesia could contribute up to USD 41 billion to the national GDP between 2024 and 2030. This projection highlights the strategic role of 5G as a connectivity foundation that supports digital transformation and the growth of the digital economy. At the same time, the increasing adoption of AI and data-driven services is driving demand for networks that are faster, more reliable, and capable of handling greater capacity. As part of its commitment to supporting these developments, ZTE continues to deliver innovations across the entire network technology value chain, from broadband access to core infrastructure. On the access side, ZTE provides AI-powered Fixed Wireless Access (FWA) solutions designed to expand high-speed connectivity more efficiently and flexibly. The solution serves as a strategic approach to supporting broadband inclusion while addressing the growing demand for connectivity across different regions. In addition, ZTE is building an open ecosystem that integrates AI, connectivity, cloud computing, and various digital technologies within a collaborative framework involving operators and enterprises. At the core infrastructure level, ZTE embeds AI capabilities natively into the baseband, the key component responsible for network signal processing. By integrating AI directly into the baseband from the design stage, networks can analyze, optimize, and adapt operations more intelligently and in real time. This approach enables more autonomous and efficient network operations while preparing networks for the demands of the 5G-Advanced era. Moving forward, ZTE will continue to deepen collaboration with operators, enterprises, and industry partners in Indonesia while strengthening its technology portfolio, ranging from wireless access solutions and optical transport to data center infrastructure and telecommunications energy solutions. In line with Indonesia's vision of becoming one of Southeast Asia's leading digital economies, ZTE remains committed to accelerating the nation's digital transformation through AI-driven innovation, intelligent connectivity, and next-generation network technologies that benefit more industries and regions across the country. Contributed by ZTE.

It won't be making smartphones great again. The long-awaited Trump-branded smartphone has finally arrived, and it appears to be exactly what many suspected: an existing handset in gold drag. Repair biz iFixit got its hands on the Trump Mobile T1 after the device became available in May, and its teardown found the model is essentially an HTC U24 Pro with cosmetic tweaks and a Trump-friendly gold finish. It was almost exactly a year ago that the Trump Organization unveiled the Trump Mobile cellular service and heralded the coming of the T1 Phone, described as "a sleek, gold smartphone engineered for performance and proudly designed and built in the United States." Few expected the gilt gadget to live up to that promise, as there are effectively no mass-market smartphones built in the US, with the possible exception of Purism's Liberty Phone, which is priced at a challenging $1,999 for those who absolutely must have a smartphone made outside China. Despite accepting $100 deposits to pre-order the coveted handwarmer, Trump Mobile failed to deliver the device by August last year, as promised, and many started to believe it would never show up. But it arrived this May amid claims that the Trump Mobile website was leaking customer data to anyone who sent an HTTP POST request. The nerds at iFixit passed the Trump Phone through a CT scanner alongside an HTC U24 Pro to confirm that the internals of the two devices are almost an exact match. They even went so far as swapping the main board of the T1 for that of the HTC phone, and showed that it not only fits, but the phone still works. One difference iFixit noted is that the multichip package housing the 12 GB of LPDDR5 memory and 512 GB of storage is from Micron, whereas the corresponding package in HTC's phone is supplied by SK hynix. The HTC U24 Pro is a mid-range smartphone that was launched almost exactly two years ago in June 2024. It is based on the Qualcomm Snapdragon 7 Gen 3 platform, has a 6.8-inch display, and came with Android 14 at launch, whereas the Trump phone features Android 15. In other words, it's a fairly unremarkable smartphone, sprayed gold and marketed to Trump fans for a promotional price of $499. To be fair, as iFixit makes clear, this is not a bad price for a device like this, so aureate wannabes are not being overcharged here. But as iFixit also makes clear, the device may be assembled in Florida, but it was designed in China and the vast majority of its parts have been sourced from and made in China as well. ®

Online chat platform VRChat says a recent cyberattack compromised the data belonging to nearly 2.5 million users. It confirmed the “data security incident” in a report filed with Maine’s attorney general, but has not disclosed it via public channels. The company’s report confirmed that its cloud environment was accessed between May 10-12, with the unauthorized intruder making off with information concerning 2,436,782 users. This included VRChat usernames, email addresses, whether a user was a VRChat+ subscriber, login histories (including device, hardware identifiers, and IP addresses), and Steam or Meta user IDs. It does not believe passwords, credit cards or other payment information, or government IDs used for age verification were affected. “VRChat sincerely regrets that this security incident occurred,” the company stated in its disclosure. “We understand that trust between our platform and its community is earned through consistent action, and we take full responsibility for the concern this event has caused. “The security and privacy of our players' information remain our highest priority, and we are committed to doing everything within our power to protect it.” VRChat said that after it was made aware of the intrusion, it contained the threat and implemented additional security controls, as well as engaging outside security experts. And in an unusual move for US breaches, the San Francisco-based company did not offer identity theft or credit monitoring services. Offering these kinds of services is not a legal requirement, but doing so is highly common, especially regarding attacks that affect so many individuals. VRChat does not publish the total number of registered users that it has on its books, but its documentation states that “the platform has grown to millions of users,” who have collectively published tens of millions of unique pieces of content for it since its first release in 2014. The part game, part chat platform is an online, open-world chatroom where people walk around interacting with one another via their 3D avatars. It has been compared to Second Life in that users explore other users' worlds, play mini-games, and partake in casual chit-chat, with support for both virtual reality headsets and conventional PCs. You can also think of it as something similar to Meta’s vision for the metaverse, just without all the coworking and KPI meetings, and with way more users. ®

This article is aimed at bioinformatics platform leads, ML infrastructure engineers, and genomics budget owners who are now running GPU-accelerated workflows in the cloud. It's about a hidden cost problem that almost every genomics infrastructure team is paying for — and very few are actively measuring. The observations here are specific to short-read sequencing workflows, which remain the dominant data type in production genomics environments. Short-read sequencing pipelines, standard in next-generation sequencing (NGS) workflows, used to be CPU-heavy. You'd run them on a cluster, they'd grind through alignment and variant calling over hours, and the bottleneck was CPU throughput. GPU acceleration wasn't the story. That has changed. AI-driven variant calling, GPU-accelerated alignment tools like Parabricks, and deep learning models running on top of sequencing data have all moved toward the GPU, which means teams are managing serious GPU infrastructure for the first time. The cost model that comes with GPU cloud differs sharply from CPU clusters, and people are bringing CPU-era assumptions about pipeline reliability and cost accounting into a GPU environment. That mismatch is costing them. We work with a lot of these teams, and when we ask about infrastructure costs, they almost always lead with the same number: cost per sample. That's what gets reported upward, what sits in the budget. What that number hides is where things get interesting. When pipelines fail A typical short-read germline variant calling pipeline has maybe ten to 15 distinct processing steps. You start with raw FASTQ files off the sequencer, run quality control, alignment, duplicate marking, base quality score recalibration, variant calling, annotation — each step hands off to the next. These pipelines mostly run on workflow managers like Nextflow or Snakemake, which do have built-in mechanisms for resuming failed jobs. Nextflow has a flag designed to let you pick up from step eight of 11 rather than restarting from scratch. In principle, that's exactly the right solution. In practice, the problem is configuration. For that flag to work, Nextflow needs to find its cache directory — the folder that records which steps completed successfully. If the solutions architect set up the compute environment without properly configuring persistent disk space for that cache, the file isn't there when you need it, and the pipeline restarts from step one anyway. That's a setup failure rather than a tool limitation, but the result is the same: you've paid for compute you didn't get output from. When a large task fails mid-execution rather than at a clean step boundary, even proper checkpointing won't save you, because the task has to be rerun in full. A problem difficult to measure Genomics teams working with Nebius consistently report that 15 to 40 percent of their pipeline runs hit at least one failure and restart before completion. Pinning the figure down precisely is hard, and we have no definitive numbers that reflect the reality here. The range is wide because it depends heavily on how mature the infrastructure setup is. Teams with well-configured environments sit at the low end; teams newer to GPU cloud, or running on spot instances with higher interruption rates, sit at the high end. What makes this invisible is that if your metric is cost per completed sample, a failed run that eventually completes still looks like one sample at normal cost. The retry disappears from the number that gets reported. For example, a GPU-accelerated whole genome sequencing pipeline — germline variant calling — takes roughly two GPU-hours on an H200. At current on-demand rates that's about $9 of compute per sample, and that's the visible cost. Now apply a 25 percent failure rate — toward the conservative end of what teams report. For every four samples you complete, one run failed, restarted, and ran from the beginning. Your real cost per completed sample isn't $9 anymore — it's $11.25, a 25 percent hidden markup. Scale that to a team processing 2,000 samples a month: the visible compute bill says $18,000, but the real cost is $22,500. That's $4,500 a month — $54,000 a year — in compute that produced no output. For a mid-size genomics team, that's a meaningful fraction of the cloud budget, and it shows up nowhere as waste. That's before you touch storage. The hidden costs The storage picture is more nuanced than people expect. A standard whole genome generates roughly 200 gigabytes of raw FASTQ data, but that's the uncompressed figure. In practice, almost everything going into cold storage is compressed, typically down to around 30 gigabytes per sample, so the storage cost per sample is quite manageable. Where it gets complicated is retrieval. When you want to reanalyze archived samples — say, running a new cohort through an updated pipeline — you pull those compressed files back, and your infrastructure then needs to decompress them. That 30-gigabyte compressed file expands to 200 gigabytes, which means you need the disk space and memory headroom to handle the expansion. If the environment wasn't sized for it, you get failures or severe slowdowns at the decompression step, which becomes another category of hidden cost that's rarely accounted for up front. In cancer research, the numbers are much larger. Somatic mutation calling runs at 60x to 100x sequencing depth, so 600-gigabyte FASTQ files aren't unusual. Everything I've described scales accordingly. The key point: retrieval from cold storage always has a cost, regardless of where your compute lives relative to your storage. Some platforms charge for data egress between regions on top of that. Either way, the teams that haven't modeled their reanalysis frequency as a real line item are almost always surprised when they do. Tracking, tracking and tracking... Bioinformatics engineers know the failure rates, because they're the ones watching jobs fail at 2am. But by the time the numbers roll up to whoever controls the budget, it's just "cloud costs." There's no line item for "compute we paid for and got no output from." Cloud billing by service and instance type doesn't surface this. You see your GPU compute spend, your storage spend, your egress. You don't see "20% of your GPU spend this month was on runs that didn't complete." That decomposition requires deliberate instrumentation, and most teams haven't built it yet. What teams should measure instead of cost per sample Teams should measure a few things instead. First, completion rate: the percentage of pipeline runs that complete without failure or restart. That's your pipeline reliability score, directly linked to compute waste. Second, cost per attempted sample versus cost per completed sample. If those numbers are meaningfully different, you have a problem worth fixing. Third, storage retrieval frequency and the infrastructure overhead of decompression: how often you're pulling archived data back, and whether you've properly sized the disk and memory headroom for it. This is the gap between what looks cheap in the storage bill and what it costs to use the data. One thing genomics infrastructure teams should do differently starting this week Instrument your pipeline failure rate, right now, before anything else. The number itself doesn't fix anything, but it makes the problem visible. Once you can show that 15 or 25 percent of your compute spend is going toward runs that restart — with real dollar figures attached — the conversation about fixing the underlying infrastructure becomes easy to have. People move fast when they can see the waste. Everything else follows from that — better checkpointing configuration, smarter storage architecture, more stable compute — but you have to see the problem first. Discover the breakthroughs shaping the future of AI in healthcare and life sciences. Visit https://nebius.com/solutions/life-sciences-and-healthcare to learn more and register for the 2026 AI Discovery Awards ceremony: nebius.com/ai-discovery-award. Anastasia Raskolova Anastasia is a senior product manager for healthcare & kife sciences at Nebius, where she focuses on infrastructure product for drug discovery and clinical AI workflows. Before that, she spent her career building ML products across computer vision, recommendation systems, and generative AI — and stays grounded in the clinical reality through volunteering in the Emergency Department at Massachusetts General Hospital. Contributed by Nebius.

HANDS ON At WWDC this week, Apple introduced container machines, which are persistent virtual machines running Linux, bearing some resemblance to Windows Subsystem for Linux (WSL) on Microsoft's operating system. Developers using macOS, as with those on Windows, face the problem that most applications are deployed to Linux, creating a mismatch between the development machine and the deployment target. The friction is less for macOS, which, like Linux, is Unix-like, but still exists. Apple's solution builds on the Container project previewed at WWDC last year. Version 1.0 was released at this year's WWDC, complete with the new container machine feature. The project uses standard Open Container Initiative (OCI) containers, and both the containers and container machines run on lightweight virtual machines (VMs), giving strong isolation. The name "container machine" is intended to convey that the feature combines both a container and a VM. The feature uses Apple's native virtualization framework, and the command line interface integrates well with macOS. Once installed, the command container machine run will open a terminal in the default container machine. Another option is to run a command such as container machine run uname -a, which will execute in the default container machine but without leaving the macOS shell. The code is written in Swift and is open source on GitHub under the Apache 2.0 license. It uses another Swift package called containerization, which is also open source. On Windows, WSL is an important tool for developers. Could container machines have a similar impact for Mac devs? There is potential, but Apple has work to do both on features and documentation, and the project is tucked away on GitHub rather than being presented as part of macOS. We tried a brief hands-on, installing the 1.0 release from the GitHub release package on Tahoe 26.5.1. Only macOS 26 is supported. Once installed, the command container machine create is enabled, though only containers that include the /sbin/init system initialization program will work. Many container images designed for running applications, rather than being used for persistent VMs, do not include this. The solution is to build a custom container image from a Dockerfile, for which the documentation now includes examples. We used the Dockerfile supplied in a tutorial that sets up a container machine based on Ubuntu 24.04 with the Swift SDK included, followed by the steps to develop using Visual Studio Code running on macOS and connecting to the container machine via VS Code remoting. This worked and we were able to build a project on Linux and run it using VS Code and Safari on the Mac side, but debugging breakpoints were not hit. We tried again with a .NET project, for which debugging worked correctly. By default, a container machine mounts the macOS home directory with read-write permissions. This is great for accessing code or other assets from both macOS and the container machine, but not good for security. A rogue package installed on Linux, for example, could easily harvest credentials from a .ssh folder in macOS. This is configurable via the --home-mount argument. Setting access to "none" is more secure. The memory available to a container machine defaults to half the system memory. In our case that is 32 GB, but after launching the VM and starting PostgreSQL, the actual memory used, according to Activity Monitor, was only 1 GB. Additional memory is used on demand, but a limitation described in the technical overview is that memory cannot be released back to the host. In other words, memory usage will increase during use and can only be released by restarting the VM. WSL supports GUI applications via the X11 or Wayland graphic systems. An issue raised by a user about GUI applications in containers was closed on the basis that developers can install XQuartz, a project for running the X windows system on macOS, and then use container-to-host networking to connect, though we did not try this. GUI support appears not to be a goal of the project. Mac developers already have many ways to run Linux containers or VMs, including the mature ecosystem around Docker, Podman, Colima, UTM, VirtualBox, and OrbStack, to mention some contenders, as well as the option of using SSH to connect to a remote Linux VM. That means Apple has some work to do to establish its native container tools, and now container machines, as serious alternatives. On the plus side, the system is lightweight, aside from the inability to release memory, and performed well in our quick hands-on. A WWDC video has further details, alongside the documentation on GitHub. ®

NASA's sprint to save the Swift observatory has reached another milestone: Katalyst Space's LINK robotic servicing spacecraft is now installed atop its Pegasus XL launcher. The milestone came less than a year after the space agency awarded the rescue contract. The next step will be to attach the Pegasus XL to the Stargazer carrier aircraft (the last airworthy Lockheed L-1011 TriStar), which will carry it from NASA's Wallops facility to the Kwajalein Atoll in the South Pacific Ocean for launch. Launch is expected to occur later this month. The goal is to boost the Swift observatory, whose orbit is decaying faster than expected due to increased solar activity. Swift lacks thrusters to compensate for the problem, so a return to Earth in the coming months is inevitable without intervention. Engineers recently bought the vehicle a little extra time by orienting the spacecraft and reducing the science output, but there is precious little margin in the timelines. The mission is high-risk, and Swift has little to lose. However, if successful, the approach could extend the lifetimes of other craft, including the Hubble Space Telescope, which will also re-enter the atmosphere in the coming years without intervention. Although NASA rejected a proposal by its now administrator Jared Isaacman to reboost the observatory using a SpaceX Dragon spacecraft, if the mission to Swift is a success, the agency will have another, far less expensive, option to consider. Like Swift, Hubble's orbit is decaying, and there will come a point in the coming years when managers must decide whether to attempt to extend the life of the veteran observatory, devise a way of performing a controlled re-entry, or let nature take its course. Swift was one of the missions slated for the chopping block under proposed budget cuts, so a successful rescue would mark a remarkable turnaround. Extending spacecraft beyond their primary mission isn't unusual. ESA, for instance, just endorsed extensions for several veteran missions, including Mars Express, XMM-Newton, and SOHO. But a Swift-style orbital rescue is something altogether different, and one that operators of other spacecraft facing decaying orbits will be watching closely. ®

If you use Office 2019 on a Mac, your software will soon stop working properly and there's nothing you can do but buy an upgrade. From July 13, 2026, Office applications on the Apple platform could lose the ability to edit, save, or create new files. Opening and printing will still work, but otherwise it's "reduced functionality mode" time, as Microsoft puts it. The problem is due to the expiration of the certificate used to validate the user's Office license, and it will affect both Microsoft 365 subscribers on macOS, iPhone, and iPad and non-subscribers. Affected software includes Office 2021 and 2019. The fix requires an update to macOS 12 or later, or iOS 17 on an iPad or iPhone, followed by an application update, which is where the problems could start. While updates are a way of life for Microsoft 365 subscribers, they aren't for everyone. Office 2021 users can manually update – support for that product ends on October 13, 2026 – but Office 2019 users are out of luck. Support ended on October 10, 2023, and, according to Microsoft, "Because Office 2019 cannot be updated to the required version, this issue cannot be resolved by updating or reinstalling Office 2019 for Mac." The solution? Perhaps a Microsoft 365 subscription? Or switch to using Microsoft 365 on the web? The issue doesn't affect Windows or Android devices, but it is galling for Apple users who purchased Office 2019 and will soon be sent to "reduced functionality mode" with no support from Microsoft. The lack of updates is understandable, considering that support ended years ago, but turning the application into little more than a viewer due to an expired license certificate seems like poor form. Users on social media have been understandably annoyed with the situation and Microsoft's stance. One wrote, they were "completely happy with Office 2019 and saw no need to upgrade to the latest version." But now they will. Or switch to a different vendor. "This is appalling from Microsoft, will definitely not be supporting them in the future." ®

Dutch semiconductor startup Qualinx is claiming a breakthrough of sorts in European sovereign manufacturing thanks to an end-to-end semiconductor fabrication flow it is using for its new satnav chips. The firm, a spin-off from Delft University of Technology, says it has demonstrated that security-critical chips for aerospace, defense, and critical infrastructure can be designed, manufactured, and delivered entirely within Europe. Tape-out of the Qualinx QLX3xx, a family of ultra-low-power Global Navigation Satellite System (GNSS) systems-on-chip (SoCs), represents the first step on the path toward a fully automated trusted European manufacturing flow, the company claims. But Qualinx is a fabless design shop and relies on a contract manufacturer to make the chips for it. In this case, it is GlobalFoundries (GF), an international business with its headquarters in the US – so much for sovereign manufacturing. The pair say that GF's Dresden fab is establishing a European manufacturing flow with funding from the European Chips Act. This will ensure that every step of the production process occurs within the EU, so that no sensitive design data leaves the region. "This first secure product demonstrates that a fully European manufacturing path – from mask services to wafer production – is already a reality today," said Qualinx CEO Tom Trill. Qualinx is perhaps placing an emphasis on security-critical chips because there are already European semiconductor firms that design and manufacture their own products, such as STMicroelectronics. And Reg readers with long memories will recall that the UK once had its own processor company in the shape of Bristol-based Inmos, which made the Transputer, manufactured at Newport Wafer Fab (NWF) in South Wales – now sold off to US chip biz Vishay Intertechnology. The Qualinx chip will be made using GF's FDX fully depleted silicon-on-insulator manufacturing process, which we understand is a 12nm node. While advanced, this is some way behind cutting-edge processes such as Taiwanese chip giant TSMC's 2nm N2 process, now in mass production. But there has been debate about whether Europe really needs cutting-edge fabs. The European Commission's new Digital Sovereignty package proposes a Chips Act 2.0 that would fund a sovereign "AI chip factory." But as the Center for European Policy Analysis (CEPA) points out, European chip demand comes mostly from the automotive sector and industrial applications, which rely on 28/22nm technology, not cutting-edge silicon. "We are demonstrating that Europe can rely on a secure, end-to-end semiconductor manufacturing flow that meets the highest requirements of aerospace and defense," stated GF SVP and general manager Dr Manfred Horstmann. "Our partnership with Qualinx marks the first operational milestone." ®

OpenAI may be headed for Wall Street, but one analyst firm is already warning enterprise customers not to get too attached. In a note published alongside OpenAI's confidential IPO filing, Forrester urged companies to keep their AI options open, arguing that today's market leader could easily become tomorrow's cautionary tale. "Don't lock into long-term contracts; keep your architectures flexible," the firm advised. "In fact, OpenAI could become AI's BlackBerry FIFO (First In, First Out). The company that defines a category is often the one most painfully displaced by it." The caution comes as OpenAI takes its first formal step toward a public listing. Alongside its confidential SEC filing, the company published a roadmap built around three ambitions: AI systems that can accelerate research, AI that boosts economic growth, and eventually a personal AGI assistant for everyone. Forrester was more interested in a fourth question: what happens if OpenAI doesn't stay on top? The firm argues that OpenAI faces what it calls a "trifecta" of challenges: persuade consumers to use its agents instead of rivals', convince enterprises to build around its technology, and stay ahead in the race toward AGI. The enterprise battle may prove the most lucrative. "Whoever automates the dull, expensive middle of a company's operations first becomes the system of record everyone else has to rip out — and almost no one does,” Forrester said. In other words, the first company to get AI agents woven into day-to-day business processes stands a decent chance of becoming yet another piece of software that everyone complains about, but nobody can remove. However, Forrester's advice is that, rather than standardizing on a single provider, enterprises should "anchor to the capability you need — not the brand that got there first — and keep your switching costs low." The warning also comes as OpenAI reportedly weighs cutting prices to fend off growing competition from rivals, including Anthropic. If the AI market is heading for a price war, enterprises may want to think twice before chaining themselves to a single supplier. Forrester also notes that a public listing could provide customers with something they currently lack: visibility into OpenAI's finances. Once public, the company would be required to disclose far more information about the cost of training and operating its models, giving enterprise buyers a clearer picture of the economics behind the AI systems they increasingly depend on. For now, OpenAI remains the company that helped define the generative AI era. Whether it becomes the next Google, the next Microsoft, or AI's answer to BlackBerry is a question investors will soon be paying very close attention to. ®

Oracle has lifted capital spending plans above analyst estimates and expanded borrowing to chase the opportunity it says exists in building datacenters for AI workloads. Despite revenue for Q4 (ended May 31) rising 21 percent year-on-year to $19.2 billion, Oracle's share price fell as markets reacted to its increasing capex, as analysts raised concerns about how Big Red would fund the investments in datacenters. Capex for fiscal 2026 reached $55.7 billion, up from $21.2 billion a year earlier. Speaking to investors, CFO Hilary Maxson said Oracle planned to support its capital investments program by raising around $40 billion in debt and equity in fiscal 2027, including a $20 billion equity issuance already announced. "We don't anticipate raising additional debt funding in calendar year 2026," she said. Last year, Oracle raised $18 billion in debt to help fund its massive datacenter investments. Big Red's market value jumped after it declared $455 billion remaining performance obligations (RPOs) – contracted revenue not yet recognized – more than 300 percent higher than a year earlier. That figure reportedly includes $300 billion for OpenAI alone, as the LLM slinger tries to support its expansion with compute capacity. Maxson said on an earnings call this week: "In order to unlock this unique growth opportunity, we started a program of capital investments. We'll continue those investments in our fiscal year 2027, with an expected net cash outlay for capital expenditures of around $70 billion. This includes customer prepayments and timing impacts expected at around $20 billion-$25 billion, so our reported capex will be higher by this amount." CEO Clay Magouyrk said any increase in capex was not due to component prices but largely due to timing. "Part of my job is to figure out ways to actually accelerate capex. My job is to try to spend the money a little bit faster so I can get ramped revenue sometimes. Component prices in general… I think everyone knows that memory prices have definitely gone up, SSD prices, hard drive prices, etc." However, Magouyrk said Oracle had also been able to lock prices "across the spectrum, whether it be space and power costs, energy costs, people costs, component costs." Oracle added around 400 MW of capacity in Q4 – similar to the last two quarters – and expects to add nearly 1 GW of capacity in fiscal Q1 2027. One analyst told Reuters there is real demand for cloud infrastructure, but the question over how Oracle funds its datacenter expansion "is getting harder, not easier, with capex coming in well above estimates and free cash flow still negative." Oracle announced a number of new customers with its latest financial figures, including a deal for a Fusion HCM system with the US Office of Personnel Management. ®