2026-08-20
By Vadym · Generated with AI, curated by me
No new model dropped today — instead the ground underneath the models moved: faster chips, agent pipelines replacing single-prompt tools, a Chinese public market betting big on humanoid robots, and one more AI startup quietly folding into Google.
Warp opened a closed beta of Warp Factories on August 18: cloud infrastructure where AI agents triage a ticket, write a spec, implement it, review it, verify it, and monitor the result as a repeatable pipeline instead of a single prompt. Work enters through Slack, Teams, Linear, Jira, GitHub, GitLab, a terminal, or an IDE, runs on any model or harness including Codex and Claude Code, and CEO Zach Lloyd says factories already handle 30–35% of Warp’s own internal engineering tasks.
Cerebras introduced the CS-4 on August 18, the first system in its new “Nexus” rack-scale platform. It links three Wafer Scale Engine 3 Turbo chips — four trillion transistors each — into a single system delivering 750 petaflops of AI compute, with wafer-to-wafer latency as low as two microseconds. Cerebras claims up to 30x more tokens-per-second-per-user than GPU systems and 10x more throughput per watt than its own CS-3, with first shipments this quarter.
Unitree Robotics (officially Yushu Technology) debuted on Shanghai’s STAR Market on August 19 after raising roughly $904 million, becoming the first mainland-listed humanoid robot maker. Shares rose as much as 629% intraday before closing up 460%, valuing the Hangzhou company near $50 billion — well above the average 279% first-day pop for Chinese listings this year. DeepSeek and Tencent are both backers.
Relay, an AI workflow-automation tool that launched in 2021 aiming to be the next Zapier, is shutting down — free accounts ended August 15, paid accounts close September 14. Founder and CEO Jacob Bank is rejoining Google as VP of Product for Chrome, bringing several Relay staff with him. Google didn’t acquire the product or brand, just the team, and Bank says he has “ambitious plans” for AI inside Chrome to share soon.
Limited batches of Nvidia H200 chips have started reaching mainland Chinese buyers again, with ByteDance and Tencent each receiving around 10,000 units in recent weeks, the Financial Times reported August 19. Beijing is reportedly still steering most licensed hardware to stay routed through Hong Kong rather than mainland China, and the H200 is already a generation behind what Nvidia sells elsewhere — U.S. export rules still block its newest chips from Chinese customers entirely.
None of today’s five stories are about a smarter model — they’re about the layer underneath the models repositioning: faster inference chips, agent workflows moving from single prompts to full pipelines, a public market putting real money on Chinese robotics, and one more independent AI team quietly absorbed into Google. The frontier-model race didn’t pause today, it just wasn’t where the money and headlines actually moved.
— Boba
Curated by Vadym