Ternary/binary quantization (1.58 bits) with <5% accuracy loss enables frontier models on smartphones. Sub-1-bit quantization via sparsity and low-rank factorization going mainstream in 12…

Chinese lab Moonshot's 2.8T parameter Kimi K3 matches GPT-4.5-class performance on the cost-performance Pareto frontier and will release weights open-source. Combined with Murati's Inkling,…

Kimi K3 proves open models can match frontier performance at 1% of training cost. Frontier intelligence is now a perishable asset with weeks-long shelf life. Value shifts from model weights…

Prism ML's Bonsai 27B runs at ternary (1.58-bit) with 5% accuracy drop; Tencent's binary compression and Samsung's sub-1-bit 'nano quant' prove the trajectory. Imad predicts sub-1-bit mains…

100x software efficiency gains don't reduce silicon demand—they expand total compute consumption. Nvidia Blackwell/Rubin sold out for years. Inference providers (Fireworks, Modal, Base10) r…

Kimi K3 and Inkling enable internal fine-tuning on proprietary data without vendor lock-in. Companies need crash programs with best advisors to decide build vs buy. Inference providers (Fir…

Frontier intelligence is now a perishable asset (shelf life: weeks). Value shifts to 'interfaces' — orchestration layers that hot-swap models (Kimi, Inkling, Fable, proprietary) per task. E…

Government review delays cut value 50%, open-weight parity cuts another 50%. Frontier labs become one ingredient; competitive moat moves to integrated tools, security, ease of deployment. J…

Frontier models (Fable, Opus) trained on CVE data enable cyber weapons. Chinese models lack this data but can be fine-tuned. Open-source cyber-capable models emerging in 1-2 quarters. Every…

As foundation model layer commoditizes (open weights, 100x cheaper inference), every application-layer company, enterprise, and inference provider gains. Only pure-play frontier labs (OpenA…

Export controls on H800s/Blackwell forced Chinese labs (Moonshot, DeepSeek) to innovate on data curation, Muon optimizer, linear attention, and quantization — achieving frontier performance…

H800 embargo was 'worst case scenario'—enough to irritate but not stop China. Forced quantization, mixture-of-experts, and data curation innovations that are now permanent. China now full-s…

70% of elite AI researchers are Chinese, Indian, Taiwanese, UK. Stapling green cards to PhDs is zero-friction win. China's ecosystem now thriving—80% of Chinese PhDs return vs Indians stayi…

150 Chinese humanoid companies, Unitree at 11k units heading to 11M/year. MMA combat demos stress-test engineering but set dangerous inductive priors. West (ProRL) needs economically produc…

Starship rapid reuse demonstrated. SpaceX AI (xAI) partnered with Anthropic for Colossus orbital compute. Unit economics cross over by early 2030s. OpenAI dismissing space now but will pivo…

The administration is exploring a FINRA-like self-regulatory organization under the SEC to govern frontier model deployment. This could make it economically infeasible for US public corpora…

Open-weight models (Kimi K3, future Chinese releases) will gain cyberattack capability within 1-2 quarters via fine-tuning on CVE data. Every company needs AI-native cyber defense. Imad: 'm…

70% of top AI researchers are non-US citizens (Chinese, Indian, Taiwanese, UK). Stapling green cards to PhDs is the lowest-friction fix. Moonshot founder Yang Zhilin started his first AI st…

AI models now match human superforecasters on ForecastBench. When hyper-forecasting connects to algo trading (already dominant by volume), markets will pre-price human collective actions fa…

150+ Chinese humanoid companies, state-backed, staging MMA combat demos (Engine AI T800s punch 4x Tyson). Alex warns PLA infantry integration is likely. West needs competitive humanoid prog…

Starship launch cadence and Colossus 2-scale compute make space-based data centers inevitable. Anthropic's partnership with SpaceX AI for orbital compute validates the thesis. Sam Altman's…