« I love LLMs, I hate hype » - geohot reminds the only rule that remains
Separate what models actually do from what is attributed to them. Discipline is lacking more than compute power.
Jul 13, 2026 at 02:20 8 8
Quantum's quiet capital wave: Oratomic's $300m Series A and Singapore's QAI cohort
Arch and Khosla just led a $300m Series A into quantum startup Oratomic - a serious sum for pre-utility hardware. In parallel, QAI Ventures picked its first four Singapore quantum cohort startups. The 2026-2028 quantum thesis isn't gate count. It's who owns the tooling layer.
Jul 11, 2026 at 11:41 12 8
Sugon Dawn 8000 at WAIC 2026: 20x density and 100,000 cards full-precision interconnect
Sugon officially unveils Dawn 8000 specs at WAIC: 20x compute density per unit, full-precision interconnect for 100,000 cards. The Chinese compute stack meets industrial standards.
13 h ago 5 8
Huawei's Atlas 950 SuperPoD debuts at WAIC: China's sovereign compute stack goes public
The first time a rack-scale Ascend training pod moves from press release to a physical demo - WAIC 2026 becomes Huawei's shop window.
yesterday 10 9
Apple explores AI chip startup acquisitions: the vertical integration playbook turns to inference silicon
Apple is exploring acquisitions of AI chip startups, sources tell Tech in Asia. The reading: Apple is preparing an inference-silicon path that does not run through Nvidia.
yesterday 7 8
OpenAI imposes hardware keys on its cybersecurity researchers
Hardware-bound passkey mandatory as of September 1st to access cutting-edge cyber models. A standard measure - and that's precisely what makes it interesting.
Jul 14, 2026 at 22:30 10 8
Reflection signs $1M in compute with Nebius: open-weight pays for a factory
A two-year-old startup, valued at $8 billion, spends a billion calculations to train open models. The bet: regulatory restrictions on closed systems make open-source strategic—not just nice.
Jul 14, 2026 at 22:30 10 7
TSMC: June revenue +68% - compute-crunch has a very strong pulse
TSMC reports a 68% year-over-year increase in June 2026 revenue. The message is clear: AI compute demand is neither satisfied nor capped in the short term.
Jul 13, 2026 at 17:27 8 9
Flash-MSA: Train on a million tokens without breaking the compute bank
A sparse attention kernel that makes long-context training finally affordable - if the benchmarks hold up outside the lab.
Jul 13, 2026 at 02:20 10 10
Samsung brings forward the opening of the Yongin 1 plant to 2029 - the global computing calendar is compressed
Samsung brings forward by one year the opening of the first module of its Yongin complex (Korea Times, July 12). The signal is clear: pressure on leading-edge capacity is forcing foundries to cut their schedules.
Jul 12, 2026 at 11:17 7 6
Qwen 3.5-122B runs as a daily driver on Mac Studio - 3 bugs fixed, a silent switch
An engineer documents on his blog three patches that transform a large Qwen 3.5 model (via MLX) into a reliable daily companion on a Mac Studio. Local hardware is making progress in practical applications, patch by patch.
Jul 12, 2026 at 11:17 10 8
Sugon Dawn 8000 - China assembles its first 100,000-card cluster
Sugon announces Dawn 8000, a cluster of 100,000 "ultra-intelligent" accelerator cards. Symbolic + industrial: Beijing catches up with the threshold of Western megaclusters, with domestic chips.
Jul 11, 2026 at 17:15 14 7