China US AI race 🤖, Nvidia’s weakness ⚡, Thinking Machines exodus 👋

Some elite Chinese AI researchers estimate that the country's chances of catching up to the US are slim in the short run due to a bottleneck in chips ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌  ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ 

Sign Up |Advertise|View Online TLDR Together With TLDR AI 2026-01-16 Launch beautiful, prod-ready sites on Framer — 1 year free for early-stage startups (Sponsor) Vibe-coding your homepage is the easiest way to look exactly like everyone else. With Framer, hundreds of YC-backed startups have built polished and unique sites in hours, not weeks.  And if you're an early-stage startup on Framer, you get: One year free: Save $360 with a full year of Framer Pro, free for early-stage startups. Tools to grow: Scale your site from MVP to full product with CMS, analytics, and AI localization. One-click publish: Make changes instantly to keep your site current. 👉 Apply to claim your free year (Pre-seed and seed-stage startups new to Framer are eligible.) 🚀 Headlines & Launches Chinese AI Developers Say They Can't Beat America Without Better Chips (5 minute read) Some elite Chinese AI researchers estimate that the country's chances of catching up to the US are slim in the short run due to a bottleneck in chips. US rules block the direct sale of Nvidia chips to China. Chinese companies can rent Nvidia chips in third countries, but this typically leaves developers with fewer chips and more inconvenience compared with well-funded competitors in the US. The US recently unblocked sales of Nvidia's H200 chip to China, but this is not likely to be a game-changer in helping Chinese companies catch up. Mobile Apps on Replit (1 minute read) Replit launched a new AI-powered coding assistant designed to enhance coding efficiency and collaboration. This tool integrates seamlessly into its platform, providing real-time code suggestions and intelligent code completion. It aims to streamline the coding process for developers and teams. 🧠 Deep Dives & Analysis Claude Economic Index Report (51 minute read) Anthropic published its January 2026 Economic Index, offering new data on how Claude is used across geographies and sectors. The report introduces five metrics to track Claude's economic footprint and includes the most detailed usage dataset released to date. Nvidia's Achilles Heel: Inference (5 minute read) Benchmark's Eric Vishria shares his email to investors about Nvidia from 2024. Nvidia's dominance holds in training, but GenAI inference is emerging as the weak spot. Purpose-built SRAM architectures from players like Groq and Cerebras deliver order-of-magnitude faster inference than GPUs, unlocking larger models, lower latency, and new use cases. As inference becomes a growing share of spend, performance gaps could shift AI compute winners away from incumbents. 🧑‍💻 Engineering & Research Solve the reliability problem of agents and chatbots (Sponsor) Turn any knowledge base into an AI that people trust. Scroll.ai's platform provides accuracy and context-adherence that generic models can't match. Your users will feel the difference from the first message. Thousands of teams use Scroll for documentation, RFPs, sales enablement, agency work, and more. Get 2 free months ($158 value) with code TLDR-AI-26 On neural scaling and the quanta hypothesis (99 minute read) Humanity started an experiment in scaling up deep neural networks several years ago. The experiment is on track to be the most expensive ever attempted, and we don't yet know what the results of the experiment will be. Despite its importance, there isn't a mature theory backing the experiment. We don't really know how to think about what neural networks are doing internally. The number of people working on such a theory in public is relatively small. Google's TranslateGemma (12 minute read) Google's TranslateGemma is a set of open machine translation models built on Gemma 3 and fine-tuned using synthetic and human-translated data, followed by reinforcement learning with quality-focused reward models. These models outperform Gemma 3 baselines across 55 language pairs and retain strong multimodal capabilities. Sequence Distillation for Efficient Reasoning (GitHub Repo) DASD is a distillation pipeline that combines techniques like temperature-scheduled learning and divergence-aware sampling to train compact models for reasoning tasks. Its 4B and 30B variants achieve strong results on coding, math, and science benchmarks. Nvidia Speeds up AI Reasoning with Fast-ThinkAct (5 minute read) Fast-ThinkAct introduces a vision-language-action framework that compresses textual reasoning into latent plans, achieving up to 9.3x faster inference for embodied AI while maintaining high reasoning performance through action-aligned visual plan distillation. 🎁 Miscellaneous How Scientists are Using Claude Code? (8 minute read) Anthropic detailed progress on Claude for Life Sciences and its AI for Science program, highlighting Claude's expanding role as a research collaborator. The update showcases Opus 4.5's improvements in scientific reasoning and new use cases across labs accelerating discovery through AI-powered experimentation and analysis. AI is everywhere, but nowhere in recent productivity data (5 minute read) Forrester's J.P. Gownder argues that AI hasn't yet boosted productivity, similar to how PCs didn't immediately impact productivity metrics. His research suggests AI could replace 6% of jobs by 2030, with many companies yet to see tangible ROI from AI projects. He contrasts AI's slow job-replacement with previous shifts like outsourcing, highlighting financial decisions rather than direct AI replacements driving current job losses. ⚡ Quick Links Two cofounders leave Thinking Machines Lab to return to OpenAI (1 minute read) Thinking Machines' cofounders, Barret Zoph and Luke Metz, along with Sam Schoenholz, are returning to OpenAI. Apple is Fighting for TSMC Capacity as Nvidia Takes Center Stage (10 minute read) Apple faces competition from Nvidia for TSMC's chip production capacity, as Nvidia's demand for AI chips rises with the tech boom. Investing in Merge Labs (2 minute read) OpenAI has invested in Merge Labs, a research lab with the long-term mission of bridging biological and artificial intelligence to maximize human ability, agency, and experience. Love TLDR? Tell your friends and get rewards! Share your referral link below with friends to get free TLDR swag! https://refer.tldr.tech/44fee20c/2 Track your referrals here. Want to advertise in TLDR? 📰 If your company is interested in reaching an audience of AI professionals and decision makers, you may want to advertise with us. Want to work at TLDR? 💼 Apply here, create your own role or send a friend's resume to jobs@tldr.tech and get $1k if we hire them! TLDR is one of Inc.'s Best Bootstrapped businesses of 2025. If you have any comments or feedback, just respond to this email! Thanks for reading, Andrew Tan, Ali Aminian, & Jacob Turner Manage your subscriptions to our other newsletters on tech, startups, and programming. Or if TLDR AI isn't for you, please unsubscribe.