News Summaries

From Scooters to Satellites: How a Startup Scored $5 Million to Build Data Centers in Space

A former e‑scooter CEO is swapping city streets for the final frontier. Jeff Poon, the founder of Cowboy Space Company, just closed a $5 million seed round to launch a fleet of “space data centers” – tiny satellite servers that can run artificial‑intelligence tasks far above Earth’s atmosphere. Backed by Andreessen Horowitz’s a16z and a host of other investors, the startup plans to start small, using its own rockets to test the concept while waiting for SpaceX’s Starship to become a regular launch service. Poon believes the exploding demand for AI computing creates dozens of business models, from specialized workloads to different hardware designs. His vision is to eventually field 10,000 satellites, each delivering about 100 kilowatts of power, adding up to a distributed gigawatt of computing capacity in orbit. Competitors like Starcloud already have a GPU‑equipped satellite in space and aim to scale up once larger launch vehicles become available. The funding round also signals a shift in venture capital: investors are now willing to back capital‑intensive space projects even if the founders lack aerospace experience. With the money in hand, Orbital (the new name for Cowboy Space Company) will begin piece‑wise AI inference services, generating revenue each time a satellite is launched, and paving the way for a new era of cloud computing that literally reaches for the stars.

Ex‑SpaceX Engineers Turn Solar Power into AI’s New Fuel

Two former SpaceX veterans, former Starlink engineers, have launched Ambrosia Energy with a bold mission: power the exploding demand for artificial‑intelligence computing using clean solar and battery technology. Drawing on their experience building satellite constellations, co‑founders Andrea Spangelo and Ben Longmier treat each solar‑plus‑storage module like a launch—deploy a few, learn fast, then scale up. In January they broke ground on a 20‑30 MW pilot plant in West Texas, and within weeks the first sections were humming at full capacity. Their plan is to replace off‑the‑shelf components with custom‑designed hardware and to open a factory in Austin that can crank out larger plants faster. The ultimate goal? Deliver gigawatts of renewable power by the end of the decade, giving AI data centers a reliable, carbon‑free alternative to natural‑gas generators. Ambrosia was initially self‑funded, but a recent investment from DFJ Growth has bolstered the effort. With the AI boom racing ahead, these ex‑SpaceX engineers hope their solar‑battery “constellation” will keep the world’s most power‑hungry machines running cleanly and continuously.

Tiny Radar Tags Uncover How Mosquitoes Zip Through Fields and Parks

Scientists have attached ultra‑light harmonic radar tags to mosquitoes to watch their real‑world flight paths for the first time. In a coffee plantation, the insects were seen cruising about two metres (roughly six and a half feet) above the ground before dipping down to land within a metre of the soil. Dense coffee trees forced the bugs to stay low, while open, shaded parkland let them travel farther and higher. The tags, barely heavier than a grain of sand, transmit a unique signal that radar stations pick up, allowing researchers to map each mosquito’s route without disturbing its natural behavior. This breakthrough reveals how landscape features—like trees, crops, and shade—shape mosquito movement, which could help predict where disease‑carrying insects are likely to appear. By understanding these patterns, public‑health officials can design smarter control strategies, such as targeted spraying or habitat modification, to curb the spread of illnesses like dengue and malaria. The study also shows the promise of radar tagging for tracking other tiny, fast‑moving creatures, opening a new window into the hidden lives of insects that affect human health and agriculture.

Why AI’s ‘Flying Pigs’ Reveal Its Biggest Weakness: The Quest for Real‑World Physics

Since last year a flood of AI agents and research tools has hit the market, but most users feel the hype outpaces the usefulness. The buzzword of 2025 is “world model” – a technology that promises to let machines understand and predict the physical world, not just generate text. OpenAI’s Sora, Stanford’s Fei‑Fei Li, and NVIDIA’s Cosmos series have all claimed breakthroughs, yet the industry is still scrambling over definitions, routes, and standards. Beijing Institute of Artificial Intelligence’s Wang Zhongyuan warns that today’s so‑called world models are a mishmash: language‑centered models that map everything to text, pixel‑centered video generators that can conjure pigs flying in the sky, 3D‑reconstruction tools, and visual‑representation systems. None truly predict real‑world physics. Video models merely stitch together frames from sci‑fi data; they don’t grasp gravity, friction or cause‑effect. The real challenge is shifting from “next‑token” prediction to “next‑physical‑state” prediction – the ability to see a cup teetering on a table and know it will fall. To get there, the field needs three things: massive, high‑quality physical‑world data; robust evaluation metrics that test understanding of laws of nature; and consensus on training routes. While agents remain the hottest commercial track, their bottlenecks lie in foundational models and cost, not architecture. The next three to five years will be a crucible as researchers try to turn AI’s imagination into genuine physical reasoning.

China’s ‘Starlink’ Takes Off: New Satellite Launch Paves Way for Global Internet Coverage

On June 4, 2026, China successfully sent the Qianfan Polar Orbit Group 11 satellites into space aboard a Long March 6 rocket from the Taiyuan Satellite Launch Center. The mission marks a major step toward completing China’s low‑orbit internet constellation, often dubbed “China’s Starlink.” The Qianfan system, overseen by Haiying Hu of the Chinese Academy of Sciences, aims to deploy 324 satellites by July 2026, joining the roughly 10,000‑satellite Starlink network already operating worldwide. Unlike traditional high‑orbit satellites, these low‑orbit “airborne base stations” sit just 300‑2,000 km above Earth, delivering fast, low‑latency connections that can reach remote mountains, deserts, seas and even moving ships. Hu explains that China’s strong ground‑based 4G/5G infrastructure has delayed urgency for space‑based internet, but critical gaps remain for scientific stations, ocean buoys, and aviation that rely on satellite links. He also warns that reliance on foreign constellations leaves vital communications vulnerable in crises—citing past GPS disruptions that spurred China’s own Beidou navigation system. With orbital slots and radio frequencies being limited resources, China’s accelerated launch schedule seeks to secure its share of the sky before they’re exhausted, positioning the Qianfan constellation as a strategic asset for both civilian connectivity and national security.

Why Advanced Chip Packaging Is the Next Big Leap for AI – Insights from OxiNming’s Summit

At a high‑profile summit in Shanghai on May 27, industry leaders explained how the AI boom is moving from the data‑hungry training stage to a massive demand for fast, low‑power inference at the edge. That shift is turning the spotlight on advanced chip packaging, which can stitch together memory, compute and even light‑based communication into a single, highly efficient system. The speakers warned that traditional chip‑making is hitting Moore’s Law limits – masks are getting too big, yields are falling and costs are soaring. By combining clever design with sophisticated packaging, manufacturers can bypass these roadblocks and deliver far more powerful devices without shrinking transistors further. Four key trends are driving this change: 1️⃣ Bigger, denser package substrates that break old size limits. 2️⃣ Stacks of high‑bandwidth memory growing from 8 to 12‑16 layers, boosting data speed. 3️⃣ New interconnect methods like hybrid bonding and flux‑less thermal compression bonding, which promise higher yields and lower costs by 2028‑2030. 4️⃣ Co‑packaged optics that bring light‑based links into AI data centers, expected to become mainstream by 2027‑2030. Supply‑chain tightness for materials such as hydrogen and specialty chemicals adds urgency for more resilient, diversified sourcing. OxiNming, backed by ASMPT’s global expertise, is positioning itself to help Chinese customers adopt these advanced packaging solutions, aiming to improve energy efficiency, total cost of ownership and deployment speed across a wide range of AI applications.

China’s Home‑Made ‘Muscles and Bones’ Power Nuclear Reactors – 70% of Key Materials Now Domestic

A research team from the Institute of Metal Research, part of the Chinese Academy of Sciences, has achieved a breakthrough that could make China’s next‑generation nuclear power plants far more self‑sufficient. Over the past year the team spent countless hours in Jiangsu monitoring a metal strip thinner than a human hair (just 0.046 mm) to create a domestic version of the REBCO superconducting alloy substrate – the “muscle” that supports high‑temperature superconducting tapes used in advanced reactors. Previously, China imported this ultra‑thin, ultra‑strong metal, but the new home‑made version meets the demanding specs: it must be hundreds of meters long, mirror‑smooth, and able to withstand extreme stresses. Achieving this required juggling more than 20 production parameters at once, from precise temperature control to uniform rolling force, and solving a stubborn surface‑defect problem that appeared when scaling up from kilogram‑scale to ton‑scale production. The effort paid off: about 70 % of the critical materials for cutting‑edge nuclear technologies—such as sodium‑cooled fast reactors and lead‑bismuth reactors—are now produced domestically. The team’s work earned the prestigious “Keyuan Craftsman” award, underscoring its importance for China’s energy security and its push toward fully indigenous nuclear power infrastructure.