Kimi K3 is live: Moonshot's 2.8T-parameter model graduates from benchmark hype to production
After weeks of benchmark videos, Kimi K3 is generally available. Now the real test begins.
19/08/2026 à 09h34 12 13
Discovered Materials is hunting cooler chips with AI - because the heat wall is now as real as the compute wall
A startup is using AI to search the space of possible semiconductor and cooling materials. At 1,000W per chip, thermal management is becoming the binding constraint on datacenter density.
10/08/2026 à 16h30 10 11
China's optical modules and domestic AI chips: building the sovereign compute stack from the bottom up
China's optical module manufacturers and PCB producers are reporting strong demand as domestic AI chip clusters scale up. It's the unglamorous infrastructure layer that makes Huawei Ascend clusters viable without Nvidia - and the supply chain is materializing.
05/08/2026 à 16h39 11 9
Sugon Dawn 8000 à WAIC 2026 : densité ×20 et interconnect 100 000 cartes en full-precision
Sugon officialise à WAIC les specs du Dawn 8000 : densité par unité de calcul multipliée par 20, interconnect full-precision à 100 000 cartes. La pile compute chinoise passe l'étalon industriel.
17/07/2026 à 14h19 19 8
Sugon Dawn 8000 - la Chine assemble son premier cluster 100 000 cartes
Sugon annonce Dawn 8000, un cluster de 100 000 cartes accélératrices "ultra-intelligentes". Symbolique + industriel : Pékin rattrape le seuil des mégaclusters Occidentaux, avec des puces domestiques.
11/07/2026 à 17h15 6 7
AI Boosted Homework Scores. Then Exam Scores Dropped.
A study found that students who used AI tools improved their homework grades - but their exam performance declined. The gap between assisted performance and retained knowledge is the real learning crisis no one is measuring.
24/08/2026 à 21h00 8 8
Context engineering: why 300 well-chosen tokens beat 100k noisy ones
An InfoQ architecture presentation makes the case that context quality is the new latency - and that most teams are over-indexing on window size instead of signal density.
14/08/2026 à 18h59 9 11
More incidents doesn't mean less reliability - and measuring it wrong will make your systems worse
An InfoQ analysis challenges the reflex to treat incident count as a reliability proxy. The teams with the best uptime often have the highest incident frequency - because they're counting the things that matter.
14/08/2026 à 18h58 6 6
Why Opus 5 feels worse to work with - and what it says about model evaluation
A widely-shared post breaks down the subjective regression many developers are experiencing with Claude Opus 5, even as benchmarks hold or improve. The gap between measured and felt capability is widening.
14/08/2026 à 18h58 8 5
Huawei's Atlas 950 SuperPoD goes public at WAIC 2026 - China's sovereign compute stack gets a flagship demo
Huawei demonstrated its Atlas 950 SuperPoD at China's flagship AI conference. It's the most visible proof point yet that Huawei's compute stack competes at the datacenter level, not just the chip level.
10/08/2026 à 16h32 11 16
ZTE unveils an AI supernode to help Chinese enterprises run large AI workloads without Nvidia
Export controls have cut off H100/H200 access for Chinese enterprises. ZTE's response: stack available domestic compute into dense cluster nodes that can handle enterprise-scale AI.
10/08/2026 à 16h31 9 10
How OpenAI keeps ChatGPT fast as agentic coding rewrites the load profile
A QCon performance engineering talk from OpenAI surfaces the infrastructure decisions behind ChatGPT's latency targets - and how the product's shift from chat to agent changed every assumption.
09/08/2026 à 00h59 11 10