China's LineShine tops TOP500; SpaceX demos AI device
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- LineShine hits 2,198 exaflops at 42.2 MW on ~45,000 LX2 CPUs (304 cores at 1.55 GHz) — no GPUs, returning China to TOP500 #1 for the first time in nearly a decade.link ›
- SpaceX showed a slim AI handset prototype to investors pre-IPO; Musk called reporting 'utterly false' as the device reportedly runs xAI on a custom OS.link ›
- Trump lifted export controls on Anthropic's Fable 5 after 18 days; Commerce's CAISI tested the new jailbreak defense, which Anthropic says blocks attacks 99% of the time.link ›
- Anthropic's Claude Science launched as a workflow product — same Claude models, 60+ scientific database connectors, no gating; beta to Pro, Max, Team, and Enterprise.link ›
- Cursor Mobile debuted Monday, joining Anthropic and OpenAI in shipping phone-first coding-agent apps; Claude Code's Boris Cherny says he now does 'most of my coding on my phone.'link ›
- AMI Labs (Yann LeCun) raised $1bn+ from Nvidia and Jeff Bezos's fund to build JEPA — betting LLMs 'are not particularly smart' beyond regurgitation.link ›
- Weird Al Yankovic publicly turned down an undisclosed AI ad, telling Syracuse.com 'I'm not a fan of AI' and 'I can't be the poster boy for AI.'link ›
China just knocked El Capitan off the TOP500 throne with LineShine, a 2,198-exaflop monster running on roughly 45,000 domestic CPUs — no GPUs, no foreign silicon, no apologies. Built at the National Supercomputing Center in Shenzhen using the LingKun processor platform, LingQi interconnect, and Kylin OS, the system draws 42.2 megawatts and beats America's flagship by more than 20%. That's the cleanest proof yet that US export controls haven't stopped China from reaching the top of high-performance computing — they've just forced Beijing to engineer around them. SpaceX also showed investors a slim AI handset ahead of its planned IPO, which Elon Musk called 'utterly false' even as the company preps a Starlink Mobile-tied consumer play to challenge OpenAI's Jony Ive hardware effort.
The stories behind this week
China Defies US Restrictions and Builds the World’s Fastest SupercomputerChina's ability to surpass El Capitan's capacity by more than 20% using entirely domestic hardware — and no GPUs at all — demonstrates that US export controls on advanced chips have not stopped China from reaching the top of high-performance computing. With AI now underpinning most of the world's most powerful supercomputers, the result shows that designing around restrictions now rivals raw chip access as a competitive lever.
SpaceX demos AI device; Musk calls report falseIf SpaceX enters consumer AI hardware, it would directly challenge OpenAI's own device effort with Jony Ive while leveraging Starlink Mobile to compete with Verizon and AT&T. Musk's combination of SpaceX/Tesla manufacturing scale and xAI integration gives him a vertically integrated stack that failed AI device makers like Humane and Rabbit lacked — though their track records show that building the device is easier than convincing consumers to buy it.
Trump Lifts Export Controls on Anthropic's Fable 5AI developers regain access to Fable 5 after an 18-day freeze, but the gatekeeping model — also now applied to OpenAI's GPT-5.6 and Anthropic's Mythos 5 — means U.S. frontier releases remain dependent on case-by-case government approval until an August executive-order deadline forces a standardized framework, even as China's 360 Security claims to match Mythos-class capability.
Anthropic's Claude Science: Workflow Over New ModelThree of the largest AI labs are now converging on scientific research with fundamentally different distribution models: Anthropic goes wide with subscription access, OpenAI gates its biology model behind enterprise review, and Google leans on proprietary foundational models like AlphaFold nobody else has. The fact that Novo Nordisk and Allen Institute appear as customer case studies for BOTH Anthropic and OpenAI suggests the scientific AI market is fragmenting into multi-vendor relationships rather than winner-take-all competition.
Building tech in the world’s secret R&D hubFor AI companies building specialized capabilities, Zurich offers a talent flywheel — ETH, EPFL, IMD's #1 talent ranking for a decade, and 210+ companies founded by ex-Google Switzerland staff — that rivals Silicon Valley in researcher density (110.5 vs 64.8 per 100K) at a fraction of US salary costs. The 60%+ deep-tech VC share signals where capital is concentrating for the next generation of AI products tied to robotics, healthcare, and manufacturing.
Cursor Launches Mobile App for Coding AgentsWhen the head of Claude Code says most of his coding now happens on his phone, that signals the desktop IDE is losing its monopoly as the center of developer workflows. With Cursor joining Anthropic and OpenAI in shipping mobile-first agent interfaces, developers who stay tethered to multi-monitor setups risk working against how the tools are actually being used.
AI is 'not smart' so what's next in artificial intelligence?The $1bn backing from Nvidia and Bezos for LeCun's post-LLM approach signals serious institutional doubt that scaling current AI alone reaches superintelligence. If World Models succeed, robotics — long stuck on household tasks like ironing — gets a viable path forward, redirecting R&D away from pure LLM scaling.
Weird Al Turned Down 'Nice Pile Of Money' For AI AdYankovic's public refusal underscores the reputational risks AI companies face in recruiting mainstream celebrity ambassadors, even for significant sums. As the Trump administration and SAG-AFTRA align on a legislative framework addressing AI's expansion, artists like Yankovic signal that some celebrity talent will actively distance themselves from AI branding, complicating the industry's push for cultural acceptance.
Why it matters: If LineShine holds, US chip export controls stop being a viable throttle on Chinese AI — the bottleneck becomes process nodes and design talent, not GPU access.
