SpaceX launches two back-to-back Starlink batches from California and Florida
March 15, 2026, 2:10 AM EDT. Two back-to-back SpaceXStarlink launches ferried satellites to low Earth orbit on Friday and Saturday, from California then Florida. A Falcon 9 first stage lifted off from Vandenberg Space Force Base on Friday, delivering 25 satellites (Group 17-31) into orbit about an hour after liftoff at 10:57 a.m. EDT. On Saturday, another Falcon 9 launched from Cape Canaveral Space Force Station, delivering 29 Starlink units (Group 10-48) after a 7:37 a.m. EDT liftoff. Both boosters, B1071 and B1095, landed on droneships Of Course I Still Love You and Just Read the Instructions. The tally lifts Starlink network toward 9,985 active sats, per Jonathan McDowell; SpaceX marked its 625th mission and 585th landing, and its 32nd flight in 2026.
Open-Source GreenBoost Aims to Extend NVIDIA GPU Memory With System RAM and NVMe for Bigger LLMs
March 15, 2026, 1:36 AM EDT. In a Linux kernel module called GreenBoost, independent developer Ferran Duarri is assembling a multi-tier memory extension for NVIDIA GPUs. The GPLv2 project does not replace NVIDIA's drivers; it works with CUDA via a user-space shim to access system RAM and NVMe as GPU memory, enabling larger models than VRAM alone. The system uses a pinned RAM pool exported as DMA-BUF file descriptors and imported into CUDA as external memory, with PCIe 4.0 x16 moving data. A watchdog tracks RAM/NVMe pressure. The CUDA shim intercepts allocations and frees; large allocations are redirected to GreenBoost. A workaround addresses Ollama's symbol resolution, which otherwise limits GPU visibility. The code is experimental and available on GitLab.
Perplexity launches Personal Computer, AI agent on M4 Mac mini servers
March 15, 2026, 1:34 AM EDT. Perplexity unveiled Personal Computer, an AI agent that runs continuously on M4 Mac mini servers and links local apps with the Perplexity platform. The system operates in a secure environment, requiring approval for sensitive actions and producing a full audit trail for each session. A kill switch gives users immediate control, and every query runs in its own sandbox. Macworld notes the approach continues a trend of using Mac minis for AI workloads. Perplexity has not disclosed Mac mini configurations or pricing. Details on whether Apple supplies hardware or Perplexity uses off-the-shelf units remain unclear. The company has not announced a price and only offers a waitlist. Rumors have circulated about an upcoming M5 Mac mini; no chip or spec is confirmed.
BYD unveils 1,500 kW charger, aims to reshape global EV charging leadership
March 15, 2026, 1:32 AM EDT. BYD unveiled a 1,500 kW charging station and said its flash charger will be rolled out outside China, according to Electrek. The 1,500 kW unit doubles last year's capability and dwarfs typical U.S. and European fast chargers, which top out around 250-350 kW. BYD aims to install 20,000 chargers within a year, a sharp push beyond about 35,000 U.S. charging bays that still operate at slower speeds. The company also introduced a next-generation blade battery, claiming it can charge from 10% to 97% in nine minutes. Critics say the U.S. has fallen behind, but supporters argue expanded EV charging infrastructure could accelerate adoption, improving air quality and lowering costs for drivers.
North Korea-linked AI scams impersonate workers to target European firms, researchers say
March 15, 2026, 1:30 AM EDT. Security researchers warn that North Korea-linked actors are using AI to impersonate legitimate workers, then approach European companies with convincing proposals. The technique blends AI-generated text, synthetic voices and manipulated images to create a credible hire or partner persona. The aim: social engineering to harvest credentials, access proprietary data or win fraudulent contracts. The operation relies on rapid outreach, coordinated multi-step messaging and short timelines that pressure recipients. First use explanation: social engineering is deception-based manipulation rather than hacking software. Companies are urged to tighten identity verification, require secondary confirmation for payments, and train staff to spot anomalies in tone, language and requests. The report underscores the widening role of AI in fraud and the challenge for compliance and risk teams.
iFixit: MacBook Neo is the most repairable MacBook in about fourteen years
March 15, 2026, 1:10 AM EDT. The MacBook Neo is not only the most affordable MacBook by far, but also the most repairable in about fourteen years, according to iFixit's teardown. The battery sits in a tray secured with 18 screws, a design iFixit says beats adhesive and should simplify replacement. iFixit notes other fixes: a flatter disassembly tree, Repair Assistant that accepts replacement parts, and easier-to-replace display and keyboard. Yet RAM and storage are soldered, reducing repair prospects; the teardown assigns the Neo a 6/10 repairability score. The verdict: a strong result for a MacBook, but not a full restoration of serviceability.
Nvidia faces pressure as Meta and Google weigh cross-use of AI chips; new LPU on horizon
March 15, 2026, 12:54 AM EDT.Meta is said to be considering renting Google-designed TPUs and possibly integrating them into its data centers by 2027, a development that rattled Nvidia after it had cited strong demand from Google's cloud. Nvidia defended its lead, but the company now appears to acknowledge rising pressure as the Financial Times reports Nvidia is preparing to unveil a new chip focused on AI inference, not training. The as-yet-untitled language processing unit (LPU) would use SRam instead of HBM, a shift tied to cost and availability. The product, expected at Nvidia's next GTC, follows its $20 billion Groq acquisition and marks a move away from Jensen Huang's idea that one GPU can handle all workloads. Analysts see growing inference share, with four players pushing into this space and Nvidia facing a less dominant phase.
India smartphone market to shrink more than 13% as memory prices spike, retailers warn
March 15, 2026, 12:48 AM EDT. New Delhi – India's smartphone market could shrink by more than 13% this year as memory prices rise 15-40% and drive device costs higher, pressuring budget buyers and small retailers. February sales fell about 35% and March is expected to worsen as fiscal-year payments bite. IDC India analyst Navkendar Singh says shipments could drop to 132 million units in 2026 from 152 million last year. Retailers say margins tighten for local brands as the absence of the usual quarterly 10% discounts coincides with price hikes up to 30% since November. AIMRA notes offline 'general trade' has stalled, with stock shortages and limited credit. Consumers delay purchases or turn to second-hand markets; if trend persists, an entry-level phone could approach ₹20,000 by year-end, up from around ₹10,000 today.
Samsung moves up foldable iPhone display production, report says
March 15, 2026, 12:38 AM EDT. Apple plans a foldable iPhone with the iPhone 18 Pro line later this year. Citing a translated source, Samsung has pushed mass production of foldable iPhone displays to as early as May, earlier than the July timeline cited. Samsung remains the sole supplier for the first generation; LG Display and BOE are not supplying first-gen units. The shift could be to avoid interfering with Samsung's own foldables or to test a crease-free panel for its Galaxy line. Samsung is also expected to roll out the Galaxy Z Fold 8 in July alongside a new Galaxy Z Wide Fold, a response to Apple's design. The foldable iPhone marks a milestone; its success remains uncertain.
Meta weighs up to 20% workforce cut to trim AI costs, sources say
March 15, 2026, 12:36 AM EDT. Three sources familiar with the matter told Reuters that Meta is weighing a cut of as much as 20% of its workforce to offset rising AI infrastructure costs. The timing and size are not finalized, the report said. A Meta spokesperson told FOX Business the piece is speculative. If enacted, the move would be the company's largest restructuring since 2022-23 and would follow earlier rounds that reduced roughly 30,000 roles. Meta employed nearly 79,000 at year-end. The company's AI investments come as rivals like Amazon also announce sizable layoffs. The report underscores pressure to improve efficiency as AI costs mount.
Galaxy S26 series launches; S25 FE pricing and Pixel 10 Pro XL in focus
March 15, 2026, 12:34 AM EDT. Samsung launched the Galaxy S26 series this week, with Wednesday's window for a free 512GB storage upgrade now closed; customers must pay full price for that tier. The S26 Ultra is driving US preorder momentum, led by the Privacy Display, horizon lock and generative edit features, with a $200 Amazon gift card on top. The S26+ remains above $1,000, effectively $1,000 after a $100 Amazon gift card, though it lacks Ultra's high-end cameras and charging tweaks. The base S26 adds a 4,300mAh battery and 256GB storage with a 6.3" display. Samsung also promotes the S25 FE at roughly half the price, while the older S25+ remains available. The Pixel 10 Pro XL is a major rival, boasting a 50MP main and 48MP telephoto and Google's Pixel UI alongside Samsung's One UI.
AI tools push software engineering back toward math and physics, says Perplexity AI CEO
March 15, 2026, 12:18 AM EDT. Aravind Srinivas, CEO of Perplexity AI, joined a growing chorus that LLMs are quietly automating the grunt work of coding and pulling computer science back toward its math and physics roots. The comments followed a high-engagement post that drew thousands of likes and views. Industry figures say the shift is real: Anthropic's Dario Amodei says the field is 6 to 12 months from AI handling most software tasks end to end; some engineers there no longer write code. Replit's CEO echoed the idea that the traditional software engineering role may disappear. Data backs the trend: GitHub Copilot speeds tasks, and the AI Exposure Index shows high task coverage by LLMs. Yet junior developers benefit most while senior engineers remain essential for verification; Code.org's founder urges less syntax, more logical reasoning.
GDC survey finds only 7% view generative AI as good for gaming
March 15, 2026, 12:16 AM EDT. At GDC, a year-end survey of game-industry workers shows generative AI remains contentious. Only 7 percent of respondents said generative AI is "good for the industry." PC Gamer reports that Moritz Baier-Lentz of Lightspeed Venture Partners called the near-unanimous skepticism 'shocked and sad,' noting Lightspeed's stakes in firms such as Anthropic. He argues pessimism is tied to layoffs-something some executives cite after COVID-era digital-entertainment bets-and warns that investment in data farms drives a RAM crunch, lifting hardware costs and making PC gaming a wealthier hobby. Microsoft's Project Helix console is slated for a late-year reveal, with the company's Copilot AI client arriving on Xbox later this year, a restrained note amid broader AI enthusiasm.
