Longtime Slashdot reader zuki writes: According to a recent report over at Tom's Hardware, a number of those among early buyers who have been able to put the highly-coveted $4,000.00 DGX Spark mini-AI workstation through its paces are reporting throttling at 100W (rather than the advertised 240W capacity), spontaneous reboots, and thermal issues under sustained load. The workstation came under fire after John Carmack, the former CTO of Oculus VR, began raising questions about its real-world performance and power draw. "His comments were enough to draw tech support from Framework and even AMD, with the offer of an AMD-driven Strix Halo-powered alternative," reports Tom's Hardware.

"What's causing this suboptimal performance, such as a firmware-level cap or thermal throttling, is not clear," the report adds. "Nvidia hasn't commented publicly on Carmack's post or user-reported instability. Meanwhile, several threads on Nvidia's developer forums now include reports of GPU crashes and unexpected shutdowns under sustained load."

An anonymous reader quotes a report from Ars Technica: On Tuesday, Westinghouse announced that it had reached an agreement with the Trump administration that would purportedly see $80 billion of new nuclear reactors built in the US. And the government indicated that it had finalized plans for a collaboration of GE Vernova and Hitachi to build additional reactors. Unfortunately, there are roughly zero details about the deal at the moment. The agreements were apparently negotiated during President Trump's trip to Japan. An announcement of those agreements indicates that "Japan and various Japanese companies" would invest "up to" $332 billion for energy infrastructure. This specifically mentioned Westinghouse, GE Vernova, and Hitachi. This promises the construction of both large AP1000 reactors and small modular nuclear reactors. The announcement then goes on to indicate that many other companies would also get a slice of that "up to $332 billion," many for basic grid infrastructure. The report notes that no reactors are currently under construction and Westinghouse's last two projects ended in bankruptcy. According to the Financial Times, the government may share in profits and ownership if the deal proceeds.

A global shortage of jet engines is threatening the rapid expansion of AI data centers, as hyperscalers like OpenAI and Amazon scramble to secure aeroderivative turbines to power their energy-hungry AI clusters. With wait times stretching into the 2030s and emissions rising, the AI boom is literally running on jet fuel. Tom's Hardware reports: Interviews and market research indicate that manufacturers are quoting years-long lead times for turbine orders. Many of those placed today are being slotted for 2028-30, and customers are increasingly entering reservation agreements or putting down substantial deposits to hold future manufacturing capacity. "I would expect by the end of the summer, we will be largely sold out through the end of '28 with this equipment," said Scott Strazik, CEO of turbine maker GE Vernova, in an interview with Bloomberg back in March.

General Electric's LM6000 and LM2500 series -- both derived from the CF6 jet engine family -- have quickly become the default choice for AI developers looking to spin up serious power in a hurry. OpenAI's infrastructure partner, Crusoe Energy, recently ordered 29 LM2500XPRESS units to supply roughly one gigawatt of temporary generation for Stargate, effectively creating a mobile jet-fueled grid inside a West Texas field. Meanwhile, ProEnergy, which retrofits used CF6-80C2 engines into trailer-mounted 48-megawatt units, confirmed that it has delivered more than 1 gigawatt of its PE6000 systems to just two data center clients. These engines, which were once strapped to Boeing 767s, now spend their lives keeping inference moving.

Siemens Energy said this year that more than 60% of its US gas turbine orders are now linked to AI data centers. In some states, like Ohio and Georgia, regulators are approving multi-gigawatt gas buildouts tied directly to hyperscale footprints. That includes full pipeline builds and multi-phase interconnects designed around private-generation campuses. But the surge in orders has collided with the cold reality of turbine manufacturing timelines. GE Vernova is currently quoting 2028 or later for new industrial units, while Mitsubishi warns new turbine blocks ordered now may not ship until the 2030s. One developer reportedly paid $25 million just to reserve a future delivery slot.

NextEra Energy and Google have partnered to restart Iowa's long-shuttered Duane Arnold nuclear plant, marking the first major U.S. attempt to revive a decommissioned reactor. "We expect Duane Arnold to be back online in early 2029, and the plant will provide more than 600 MW of clean, safe, 'always-on' nuclear energy to the regional grid," said Google in a blog post. Reuters reports: Under the 25-year agreement, the tech giant will purchase power from the 615-MW plant for its growing cloud and AI infrastructure in the state, while also driving significant economic investment to the Midwest region. One of the plant's minority owners, Central Iowa Power Cooperative (CIPCO), will purchase the remaining portion of the plant's output on the same terms as Google, NextEra said. The utility added that it had also signed agreements to acquire CIPCO and Corn Belt Power Cooperative's combined 30% interest in the Duane Arnold plant, bringing NextEra's ownership to 100%.

Qualcomm has entered the AI data center chip race with its new AI200 and AI250 accelerators, directly challenging Nvidia and AMD's dominance by promising lower power costs and high memory capacity. CNBC reports: The AI chips are a shift from Qualcomm, which has thus far focused on semiconductors for wireless connectivity and mobile devices, not massive data centers. Qualcomm said that both the AI200, which will go on sale in 2026, and the AI250, planned for 2027, can come in a system that fills up a full, liquid-cooled server rack. Qualcomm is matching Nvidia and AMD, which offer their graphics processing units, or GPUs, in full-rack systems that allow as many as 72 chips to act as one computer. AI labs need that computing power to run the most advanced models.

Qualcomm's data center chips are based on the AI parts in Qualcomm's smartphone chips called Hexagon neural processing units, or NPUs. "We first wanted to prove ourselves in other domains, and once we built our strength over there, it was pretty easy for us to go up a notch into the data center level," Durga Malladi, Qualcomm's general manager for data center and edge, said on a call with reporters last week.

North Dakota experienced an almost 40% increase in electricity demand "thanks in part to an explosion of data centers," reports the Washington Post. Yet the state saw a 1% drop in its per kilowatt-hour rates.

"A new study from researchers at Lawrence Berkeley National Laboratory and the consulting group Brattle suggests that, counterintuitively, more electricity demand can actually lower prices..." Between 2019 and 2024, the researchers calculated, states with spikes in electricity demand saw lower prices overall. Instead, they found that the biggest factors behind rising rates were the cost of poles, wires and other electrical equipment — as well as the cost of safeguarding that infrastructure against future disasters... [T]he largest costs are fixed costs — that is, maintaining the massive system of poles and wires that keeps electricity flowing. That system is getting old and is under increasing pressures from wildfires, hurricanes and other extreme weather. More power customers, therefore, means more ways to divvy up those fixed costs. "What that means is you can then take some of those fixed infrastructure costs and end up spreading them around more megawatt-hours that are being sold — and that can actually reduce rates for everyone," said Ryan Hledik [principal at Brattle and a member of the research team]...

[T]he new study shows that the costs of operating and installing wind, natural gas, coal and solar have been falling over the past 20 years. Since 2005, generation costs have fallen by 35 percent, from $234 billion to $153 billion. But the costs of the huge wires that transmit that power across the grid, and the poles and wires that deliver that electricity to customers, are skyrocketing. In the past two decades, transmission costs nearly tripled; distribution costs more than doubled. Part of that trend is from the rising costs of parts: The price of transformers and wires, for example, has far outpaced inflation over the past five years. At the same time, U.S. utilities haven't been on top of replacing power poles and lines in the past, and are now trying to catch up. According to another report from Brattle, utilities are already spending more than $10 billion a year replacing aging transmission lines.

And finally, escalating extreme-weather events are knocking out local lines, forcing utilities to spend big to make fixes. Last year, Hurricane Beryl decimated Houston's power grid, forcing months of costly repairs. The threat of wildfires in the West, meanwhile, is making utilities spend billions on burying power lines. According to the Lawrence Berkeley study, about 40 percent of California's electricity price increase over the last five years was due to wildfire-related costs.
Yet the researchers tell the Washington Post that prices could still increase if utilities have to quickly build more infrastructure just to handle data center. But their point is "This is a much more nuanced issue than just, 'We have a new data center, so rates will go up.'"

As the article points out, "Generous subsidies for rooftop solar also increased rates in certain states, mostly in places such as California and Maine... If customers install rooftop solar panels, demand for electricity shrinks, spreading those fixed costs over a smaller set of consumers.

Long-time Slashdot reader schwit1 shares this article from Interesting Engineering: Bill Gates-backed TerraPower's innovative Natrium reactor project in Wyoming has cleared a critical federal regulatory hurdle. The US Nuclear Regulatory Commission (NRC) has successfully completed its final Environmental Impact Statement (EIS) for the project, known as Kemmerer Unit 1, and found no adverse impacts that would block its construction.

The commission officially recommended that a construction permit be issued to TerraPower subsidiary USO for the facility in Lincoln County.

This announcement marks a significant milestone, making the Natrium project the first-ever advanced commercial nuclear power plant in the country to successfully complete this rigorous environmental review process... The first-of-a-kind design utilizes an 840 MW (thermal) pool-type reactor connected to a molten salt-based energy storage system. This storage technology is the plant's most unique feature. It is designed to keep the base output steady, ensuring constant reliability, but it also allows the plant to function like a massive battery. The system can store heat and boost the plant's output to 500 MWe when demand peaks, allowing it to ramp up power quickly to support the grid. TerraPower says it is the only advanced reactor design with this unique capability. The Natrium plant is strategically designed to replace electricity generation capacity following the planned retirement of existing coal-fired facilities in the region.

While the regulatory process for the nuclear components continues, construction on the non-nuclear portions of the site already began in June 2024. When completed, the Natrium plant is poised to be the first utility-scale advanced nuclear power plant in the United States.
The next step for the construction permit application is a final safety evaluation, which is anticipated by December 31, 2025, according to announcement from TerraPower, which notes that the project is being developed through a public-private partnership with the U.S. Energy Department.

"When completed, the Natrium plant will be the first utility-scale advanced nuclear power plant in the United States."

Texas's electricity demand has surged to record highs in 2025 but renewable energy is meeting the challenge. According to new data from the Energy Information Administration, solar output has quadrupled since 2021, wind continues steady growth, and battery storage is increasingly stabilizing the grid during evening peaks. Electrek reports: ERCOT, which supplies power to about 90% of the state, saw demand jump 5% year-over-year to 372 terawatt hours (TWh) -- a 23% increase since 2021. No other major US grid has grown faster over the past year. [...] The biggest growth story in Texas power generation is solar. Utility-scale solar plants produced 45 TWh from January through September, up 50% from 2024 and nearly four times what they generated in 2021 (11 TWh). Wind power also continued to climb, producing 87 TWh through September -- a 4% increase from last year and 36% more than in 2021.

Together, wind and solar supplied 36% of ERCOT's total electricity over those nine months. Solar, in particular, has transformed Texas's daytime energy mix. From June to September, ERCOT solar farms generated an average of 24 gigawatts (GW) between noon and 1 pm -- double the midday output from 2023. That growth has pushed down natural gas use at midday from 50% of the mix in 2023 to 37% this year. The report notes that while natural gas is still Texas's dominant power source, it isn't growing like it used to. "Gas comprised 43% of ERCOT's generation mix during the first nine months of 2025, down from 47% in the first nine months of 2023 and 2024," reports Electrek.

Apple has begun shipping U.S.-made AI servers from a new factory in Houston, Texas -- part of its $600 billion investment in American manufacturing and supply chains. CNBC reports: Apple Chief Operating Officer Sabih Khan said on Thursday that the servers will power the company's Apple Intelligence and Private Cloud Compute services. Apple is using its own silicon in its Apple Intelligence servers. "Our teams have done an incredible job accelerating work to get the new Houston factory up and running ahead of schedule and we plan to continue expanding the facility to increase production next year," Khan said in a statement. The Houston factory is on track to create thousands of jobs, Apple said. The Apple servers were previously manufactured overseas.

IBM announced that its quantum error-correction algorithm can now run in real time on standard AMD field-programmable gate array (FPGA) chips -- a major step toward making quantum computing more practical and affordable. Reuters reports: In June, IBM said it had developed an algorithm to run alongside quantum chips that can address such errors. In a research paper seen by Reuters to be published on Monday, IBM will show it can run those algorithms in real time on a type of chip called a field programmable gate array manufactured by AMD.

Jay Gambetta, director of IBM research, said the work showed that IBM's algorithm not only works in the real world, but can operate on a readily available AMD chip that is not "ridiculously expensive." "Implementing it, and showing that the implementation is actually 10 times faster than what is needed, is a big deal," Gambetta said in an interview. IBM has a multi-year plan to build a quantum computer called Starling by 2029. Gambetta said the algorithm work disclosed Friday was completed a year ahead of schedule.

The amount of energy required to supply the data centers powering AI is so vast that meeting that need may be more than a lifetime away, according to a senior executive at Apollo Global. From a report: "The gap between what AI is demanding and what we have everywhere in the world on the grid in terms of generation and transmission is huge and will not be closed in our lifetime," Dave Stangis, who has led and developed Apollo's sustainability strategy over the past four years, said in an interview.

That means sustainable energy investors need to accept that renewables alone aren't enough to power the AI age, he said. The comments encapsulate a new approach across the finance industry, where the economics of the energy transition -- a concept intended to represent the shift to a low-carbon future -- are becoming merged with the economics of an unprecedented boost in supply. "So what is happening around the world, there's no doubt about it, is what you might call energy addition," Stangis said. "The world is scrambling to add every source of power."

Google is migrating all its internal workloads to run on both x86 and its custom Axion Arm chips, with major services like YouTube, Gmail, and BigQuery already running on both architectures. The Register reports: The search and ads giant documented its move in a preprint paper published last week, titled "Instruction Set Migration at Warehouse Scale," and in a Wednesday post that reveals YouTube, Gmail, and BigQuery already run on both x86 and its Axion Arm CPUs -- as do around 30,000 more applications. Both documents explain Google's migration process, which engineering fellow Parthasarathy Ranganathan and developer relations engineer Wolff Dobson said started with an assumption "that we would be spending time on architectural differences such as floating point drift, concurrency, intrinsics such as platform-specific operators, and performance." [...]

The post and paper detail work on 30,000 applications, a collection of code sufficiently large that Google pressed its existing automation tools into service -- and then built a new AI tool called "CogniPort" to do things its other tools could not. [...] Google found the agent succeeded about 30 percent of the time under certain conditions, and did best on test fixes, platform-specific conditionals, and data representation fixes. That's not an enormous success rate, but Google has at least another 70,000 packages to port.

The company's aim is to finish the job so its famed Borg cluster manager -- the basis of Kubernetes -- can allocate internal workloads in ways that efficiently utilize Arm servers. Doing so will likely save money, because Google claims its Axion-powered machines deliver up to 65 percent better price-performance than x86 instances, and can be 60 percent more energy-efficient. Those numbers, and the scale of Google's code migration project, suggest the web giant will need fewer x86 processors in years to come.

Samsung has officially launched the Galaxy XR, the first Android headset powered by Google's new Android XR platform. Priced at $1,800 without controllers, the device features dual 4.3K Micro-OLED displays, a Snapdragon XR2+ Gen 2 chip, extensive camera tracking, and deep Gemini AI integration. Ars Technica reports: Galaxy XR is a fully enclosed headset with passthrough video. It looks similar to the Apple Vision Pro, right down to the battery pack at the end of a cable. It packs solid hardware, including 16GB of RAM, 256GB of storage, and a Snapdragon XR2+ Gen 2 processor. That's a slightly newer version of the chip powering Meta's Quest 3 headset, featuring six CPU cores and an Adreno GPU that supports up to dual 4.3K displays. The new headset has a pair of 3,552 x 3,840 Micro-OLED displays with a 109-degree field of view. That's marginally more pixels than the Vision Pro and almost three times as many as the Quest 3. The displays can refresh at up to 90Hz, but the default is 72Hz to save power.

Like other XR (extended reality) devices, the Galaxy XR is covered with cameras. There are two 6.5 MP stereoscopic cameras that stream your surroundings to the high-quality screens, allowing the software to add virtual elements on top. There are six more outward-facing cameras for headset positioning and hand tracking. Four more cameras are on the inside for eye-tracking, and they can scan your iris for secure unlocking and password fill (in select apps). Samsung says the Galaxy XR has enough juice for two hours of general use or two and a half hours of video. That's not terribly long, but you may not want to wear the 545 grams (1.2 pounds) headset for even two hours. That's even a little heavier than the Quest 3, which has an integrated battery. However, both pale in comparison to the 800 g (1.7 pounds) second-generation Vision Pro.

General Motors is ending production of its Chevy BrightDrop electric delivery vans after sluggish demand and the expiration of key EV tax credits. "This is not a decision we made lightly because of the impact on our employees," GM CEO Mary Barra said during the company's third quarter earnings call Tuesday. "However the commercial electric van market has been developing much slower than expected, and changes to the regulatory framework and fleet incentives has made the business even more challenging." The Verge reports: Brightdrop first launched in 2021 as GM's effort to capture a large portion of the commercial EV market, starting with a pair of electric vans, as well as fleet management software and electric-powered carts for goods delivery. The automaker made deals with Walmart, FedEx, and other major retailers to add the van to their delivery fleets. But after trying to make a go of it as a standalone brand, GM reabsorbed BrightDrop in 2023, and then later assigned it to Chevy in order to tap into the brand's sales and service dealer network.

Now the van will stand as yet another casualty of the expiration of the $7,500 federal EV tax credit, which ended on September 30th. In addition to the consumer credit, there was also a $7,500 discount for commercial EVs under 18,000 lbs -- which Brightdrop was eligible for. The van was a range leader, but also was more expensive than its most prominent competitor. Brightdrop's vans started at $74,000, while Ford's E-Transit van with extended battery range sold for $51,600.

British Columbia is permanently banning new cryptocurrency mining operations from connecting to its power grid to conserve electricity for industries that generate more jobs and tax revenue. The province is also capping power allocations for AI and data centers, while launching a competitive allocation process in January 2026. CoinDesk reports: The move from the government of Canada's third-most populous province is part of a broader legislative and regulatory overhaul unveiled Monday [...]. "Government will also implement several regulatory and policy changes in fall 2025 that will ... permanently ban new BC Hydro connections to the electricity grid for cryptocurrency mining to preserve the province's electricity supply and avoid the overburdening of the electricity grid," the government said in a post on its website

The province said the restrictions will help prevent grid strain and ensure industrial development is powered by clean electricity. "We're seeing unprecedented demand from traditional and emerging industries," Charlotte Mitha, the president and CEO of power utility BC Hydro, said in the web post. "The province's strategy empowers BC Hydro to manage this growth responsibly, keeping our grid reliable and our energy future clean and affordable." Crypto mining operations often consume large amounts of electricity without creating many local jobs or tax revenue, according to the statement. By contrast, projects like mines or liquefied natural gas (LNG) facilities are seen as more beneficial to the economy.

alternative_right shares a report from the Internet Archive: This October, the Internet Archive's Wayback Machine is projected to hit a once-in-a-generation milestone: 1 trillion web pages archived. That's one trillion memories, moments, and movements -- preserved for the public and available to access via the Wayback Machine.

We'll be commemorating this historic achievement on October 22, 2025, with a global event: a party at our San Francisco headquarters and a livestream for friends and supporters around the world. More than a celebration, it's a tribute to what we've built together: a free and open digital library of the web.

Filipino workers in Manila are remotely operating robots that restock convenience store shelves across Tokyo. The partnership represents a new economic model where physical labor can be offshored through telepresence. Around 60 workers at Astro Robotics monitor the machines and intervene when problems occur about 4% of the time. They earn between $250 and $315 per month. Japan faces severe labor shortages but has resisted expanding immigration. Offshoring the work through robots solves this while dramatically reducing costs.

Filipino workers are also training the AI systems designed to eliminate the need for human operators entirely. Tokyo-based Telexistence has collected extensive data from its workers and is providing it to a San Francisco startup building fully autonomous robots. The combination of automation and offshoring creates what one University of Michigan professor called a "double whammy" for workers in developed nations. It also exploits workers in developing countries who build the tools meant to replace them. The market for AI agents is expected to grow eightfold to $43 billion by 2030. Human-only work is forecast to drop 27% over the next five years.

Amazon executives believe the company can avoid hiring more than 160,000 workers in the United States by 2027 through robotic automation. Internal documents viewed by The New York Times show the automation would save approximately 30 cents on each item the company picks, packs and delivers. The documents reveal that executives told Amazon's board last year they hoped automation would allow the company to flatten its U.S. workforce growth over the next decade.

Amazon expects to sell twice as many products by 2033. That projection translates to more than 600,000 positions Amazon would not need to fill. Amazon opened its most advanced warehouse in Shreveport, Louisiana last year as a template for future facilities. The site uses a thousand robots and employed a quarter fewer workers than it would have without automation. The company plans to replicate this design in approximately 40 facilities by the end of 2027. A facility in Stone Mountain, Georgia currently employs roughly 4,000 workers. After a planned robotic retrofit, internal analyses project it will process 10% more items but need as many as 1,200 fewer employees. The documents show Amazon's robotics team has set a goal to automate 75% of its operations.

Alibaba Cloud claims its new Aegaeon GPU pooling system cuts Nvidia GPU use by 82%, letting 213 H20 accelerators handle workloads that previously required 1,192. The advancements have been detailed in a paper (PDF) at the 2025 ACM Symposium on Operating Systems (SOSP) in Seoul. Tom's Hardware reports: Unlike training-time breakthroughs that chase model quality or speed, Aegaeon is an inference-time scheduler designed to maximize GPU utilization across many models with bursty or unpredictable demand. Instead of pinning one accelerator to one model, Aegaeon virtualizes GPU access at the token level, allowing it to schedule tiny slices of work across a shared pool. This means one H20 could serve several different models simultaneously, with system-wide "goodput" -- a measure of effective output -- rising by as much as nine times compared to older serverless systems.

The system was tested in production over several months, according to the paper, which lists authors from both Peking University and Alibaba's infrastructure division, including CTO Jingren Zhou. During that window, the number of GPUs needed to support dozens of different LLMs -- ranging in size up to 72 billion parameters -- fell from 1,192 to just 213. While the paper does not break down which models contributed most to the savings, reporting by the South China Morning Post says the tests were conducted using Nvidia's H20, one of the few accelerators still legally available to Chinese buyers under current U.S. export controls.

Investigators recovered the OceanGate Titan sub's underwater camera nearly intact, discovering a SanDisk SD card that survived the 2023 implosion and still contained 12 images and 9 videos. TechSpot reports: Scott Manley, the science communication YouTuber, gamer, astrophysicist, and programmer, posted about the latest find: a hardened SubC-branded Rayfin Mk2 Benthic Camera containing the undamaged SD card. The titanium and synthetic sapphire crystal camera is rated to withstand depths of up to 6,000 meters (19,685 feet) -- the Titan imploded at around 3,300 meters (10,827 feet). The casing is intact, though the lens is shattered and the PCBs are slightly damaged.

Incredibly the SD card inside the camera was undamaged. Tom's Hardware reports that it's almost certainly a SanDisk Extreme Pro 512GB, which costs around $62 on Amazon. The camera's SD card was found to be fully encrypted, divided into a small partition for operating system updates and a larger one for user data. Due to impact damage from the accident, several components of the system-on-module (SOM) board -- including connectors and the microcontroller -- were broken, complicating the data extraction process. [...] After determining the data wasn't encrypted beyond the file system level, they successfully accessed the SD card contents using the manufacturer's proprietary equipment and procedures.