In a paper titled "Financing the AI Buildout" presented at the 2026 BPEA conference, the Brookings Institution estimated that U.S. AI infrastructure construction will require $10.3 trillion from 2025 to 2032. It pointed out that tech giants have isolated approximately $970 billion in lease commitments and over $2.4 trillion in off-balance-sheet liabilities from their financial statements through SPVs, project bonds, and long-term leases. Using Meta's Hyperion project as an example, it demonstrated how 90% debt leverage, residual value guarantees, and lease options support lightweight reporting while amplifying transmission risks. The paper warned of multiple mismatches between power, chips, construction, and long-term debt, and called for the establishment of a comparable disclosure framework for economy-wide capital stacks.
PrivOS is an integrated AI-native operating system for enterprise work, featuring chat, documents, Kanban tasks (supporting list/table/Kanban views), and custom MCP applications. AI agents can execute operations in sandboxed containers by @-mentioning lists, files, folders, or apps. It offers a free Community Edition (Beta, supporting self-hosting for up to 10 users) and EU-hosted cloud services, compatible with models using Anthropic and OpenAI interfaces (including local models).
The article points out that as single-rack power consumption in AI computing exceeds 132kW to over 190kW, air cooling is approaching its limits, making full liquid cooling an inevitable trend; the value of liquid cooling systems is concentrated in three key segments: cold plates (accounting for 32%-40% of costs), CDUs, and quick connectors, corresponding respectively to heat dissipation limits, flow control precision, and online maintenance reliability; the high-end global market is monopolized by Taiwanese firms (cold plates) and Western firms (CDUs/quick connectors), with their moats lying in co-design capabilities, mass production consistency, and liability acceptance mechanisms; Chinese manufacturers have achieved breakthroughs through three paths: cross-industry restructuring (such as Sanhua Intelligent Controls and Envicool), single-point efficiency iteration (such as Inspur Information partnering with JD Cloud), and accelerated system-level delivery (such as Kehua Data's 45-day POD delivery), and have entered NVIDIA's MGX ecosystem and Google's custom supply chain.
NVIDIA has released the Vera CPU, designed specifically to support AI Agent workflows, emphasizing high single-core performance, LPDDR5X memory bandwidth, and Scalable Coherency Fabric; cloud providers such as Nebius, Nscale, and Verda, server manufacturers including Bull, Dell, and HPE, data processing companies like ClickHouse, Elastic, and SAP, developer platforms such as Atlassian, Inherent, and Nous Research, and sandbox service providers including Baseten-Blaxel, Modal, and Vercel have announced adoption or testing of Vera.
Amazon sent an email to US marketplace sellers requiring confirmation of whether the Importer of Record (IOR) used for goods is a US entity or a foreign entity. The background is Executive Order No. 14411, "Strengthening Customs Enforcement," signed in June. This order requires IORs to disclose information such as import volume, establishment date, and ownership; raises thresholds for bonds and domestic assets; restricts informal customs declarations by foreign IORs; and purges inactive IORs. If an IOR fails verification or becomes invalid, it may lead to customs clearance delays or inability to pick up goods. Channels relying on shared IORs for double-clearance tax-inclusive shipping are most severely impacted.
On October 14, the PancakeSwap core team submitted 'The Flip' proposal to the governance forum, reducing CAKE daily emissions from 22,500 to 7,000 tokens, a 69% decrease; simultaneously, the CAKE burn mechanism for V3 and Infinity pools was canceled, with fee distribution changed to 88% for LPs and 12% for the team treasury, while only V2 pools retain a 23% burn ratio.
Zhiyuan Robotics, together with institutions such as the Anhui Automotive Innovation Center and Guoyuan Equity, completed the Pre-A round financing of RISC-V chip company Xinji Wuxian. The funds will be used for the research, development, and industrialization of full-stack RISC-V chips for embodied intelligence. Xinji Wuxian has already achieved mass production of automotive-grade MCUs (with 5.2 million new orders signed in 2026), and its XC64AE robot chip is currently undergoing mass production verification with leading embodied intelligence enterprises. During the same period, Zhiyuan took controlling stakes in Shangwei New Materials, established an industrial fund, and advanced its 'one split into four' business separation, forming a four-layer capital structure of direct investment—fund—holding—separation.
moyuge has developed and open-sourced MirrorProxy based on Rust. It supports proxying for various mirror sources including GitHub, Docker Hub, npm, PyPI, Cargo, Go Modules, Maven, NuGet, RubyGems, Homebrew, WinGet, Rustup, NVM, Debian, Ubuntu, Fedora, Arch Linux, Alpine, and OpenWrt. It is suitable for users with overseas servers or VPNs who wish to self-host.
This project released the reverse-engineered source code of Sid Meier's Civilization (1991 MS-DOS version), built using Microsoft C 5.1, MASM 5.10, and LINK 5.01; the generated CIV.EXE is byte-identical to the original version 474.05. The project includes the technical book "Reading Civilization from the inside" and Docco narrative-style code comments, supports local builds via DOSBox-X and Dev Container, and utilized LLMs from OpenAI, Anthropic, and Google to assist in reconstructing the code and documentation.
forzen has officially released the macOS version of DbCat, a database client developed in Rust. It supports 12 databases including MySQL, PostgreSQL, SQL Server, Redis, and MongoDB. The interface uses GPUI for GPU rendering and offers features such as inline table editing, visual schema modification, ER diagrams, SQL autocompletion, and execution plans. Passwords are stored only in the macOS Keychain with no telemetry or tracking. Currently in 0.x beta, it supports only macOS 12+ (Apple Silicon and Intel). The official website is https://dbcat.forzens.com, and the code is hosted at https://github.com/dbcat-app/dbcat
Plex Media Server 存在可被串联利用的4个安全漏洞，攻击者无需登录即可远程执行代码，在未升级至1.43.3.10861及更高版本的公网服务器上部署挖矿程序与代理软件；AOMEI Backupper 和 AnyViewer 的官方更新服务器在2026年4月至7月间多次遭黑客短暂劫持（单次最长10分钟），用户正常检查更新时可能下载并运行恶意程序，Google与Mandiant已披露该 ephemeral 供应链攻击。
lidedongsn released FrameCut, a local Mac vlog editing software optimized exclusively for Apple silicon, weighing only 2.3MB and watermark-free. It supports HDR processing, color grading and transitions via Metal GPU acceleration, and speech-to-subtitles via the Neural Engine; on an M2, exporting a 3-minute vlog takes about 26 seconds for 1080p and about 95 seconds for 4K. Open-sourced under MIT, written in Swift + SwiftUI, it requires manual permission to run upon first use.
Papertrade announced a contract upgrade and temporarily suspended PAPER token transfers to prevent "minting strangle attacks" that threaten the solvency of external market short liquidity pools; this move led platforms such as Aster and Paragon to delist PAPER pre-launch perpetual contracts. The protocol accumulated $62.45 million in revenue within ten days of launch, with 84% coming from LP surplus (i.e., net trader losses), but active addresses dropped by 42% from their peak, with a net outflow of $11.35 million over the past 7 days. The PAPER minting rate has decreased from an initial 100 tokens per $1 loss to approximately 10.3 tokens.
The Brains is an open-source (MIT) tool that converts public speeches, articles, interviews, and posts by founders, CEOs, and investors into searchable MCP servers. It allows direct questioning in Claude, Cursor, ChatGPT, or Codex, with every answer accompanied by source citations and links. Currently, it integrates 37 individuals, providing individual URLs for each person or a unified entry point; the project is unofficial and not affiliated with the featured individuals.
Strata 0.2.0-beta has been released, supporting GUI installation on Windows/macOS/Linux; local large models can run with a combined minimum of 27.6GB of RAM + VRAM. New features include a Chinese interface, AI coding Agent, and image understanding support. Recommended tiers for quantized models have been lowered (e.g., IQ3_XXS, which previously required 64GB, is now available for 48GB). Text generation speed on NVIDIA RTX 30 series increases by up to approximately 13%, prompt processing speed on AMD RX 9000 series nearly doubles, and Intel Arc A750 sees an improvement of about 42%.
bingoAI released an interactive single-page application for Conway's Game of Life based on native HTML, Canvas, and Vanilla JS. It includes visual breakdowns of rules, an interactive simulator (supporting drag-to-seed, one-click placement of Gosper's Glider Gun from the gallery, spacebar control for play/pause, and step-by-step evolution), background information on the Turing completeness of cellular automata, and classification of 1D rules. The page itself is a practice in AI collaborative development. Online experience address: https://jiangbin-ai.pages.dev/apps/life/.
Independent developer pmer has released Laohuangli (Huangli Chacha), an ad-free, transparent-algorithm modern traditional Chinese almanac tool that supports RFC 5545 .ics exports. The web version is currently available online, WeChat Mini Program v1.12.2 is live, and Android v0.3.1 is in closed beta testing. The project has accumulated over 240 commits, with algorithm documentation publicly available at /methodology/, and the frontend built on Astro 6.
Aero, the new protocol resulting from the merger of Aerodrome and Velodrome, launched on October 22, 2026, covering seven chains: Base, Ethereum Mainnet, OP Mainnet, Arc, Ink, Arbitrum, and Robinhood Chain. Old AERO is exchanged for new AERO at a 1:1 ratio, while 1 VELO is exchanged for 0.044 new AERO. The ve model has been replaced by the Predictive Allocation mechanism, where sAERO holders can adjust reward allocations multiple times, and rewards are minted only upon claim. Emissions are tied to protocol revenue, with Gauge Caps setting the reward limit for each pool at 1.2 times the projected revenue. Slipstream dynamic fees and MEV taxes were introduced simultaneously. Following the launch of Coinbase tokenized stocks, Aero's share of EVM tokenized stock trading volume rose to 34%.
BrbiwsFtd9zDGZqB released Condr, a personally developed agent management tool. Built with Rust, it features a native GUI terminal, supports the OSC 7501 protocol for monitoring agent status, cross-device agent orchestration, and SSH/TCP/Noise/P2P remote connections; currently used as a daily primary tool, available for download on macOS/Windows/Linux platforms, with iOS/Android versions under native development; the project is open source at https://github.com/condrdev/condr.
hanguokai developed and released WeChat MD Studio, a WeChat official account typesetting tool focused on enhancing Markdown layout. It supports 10 high-aesthetic themes, GitHub Alert semantic callout syntax, 1:1 replication of the WeChat dark mode algorithm preview, title numbering and alignment toggling, automatic conversion of external links to end-of-article footnotes, and code block highlighting with Mac window styles. Two access addresses are provided for domestic containers and overseas deployment.