2025-12-20 Tech Briefing | NVIDIA's Blackwell architecture-based RTX PRO 5000 72GB GPU is now available, empowering local AI workflows.

Saturday

Welcome to this edition, featuring 10 articles.

📚 Contents/📝 TL;DR

📚 Contents

📝 TL;DR

Primary

Another NVIDIA Blackwell GPU Launch, Enabling Complex AI Workflows Locally

NVIDIA's Blackwell architecture-based RTX PRO 5000 72GB GPU is now available, empowering local AI workflows.

Another NVIDIA Blackwell GPU Launch, Enabling Complex AI Workflows Locally

NVIDIA's Blackwell architecture-based RTX PRO 5000 72GB GPU is now available, empowering local AI workflows.

Innovating Against Streaming Abuse - And Our Latest Transparency Report

Cloudflare releases its first-half 2025 report, focusing on streaming abuse and content blocking.

Innovating Against Streaming Abuse - And Our Latest Transparency Report

Cloudflare releases its first-half 2025 report, focusing on streaming abuse and content blocking.

What is Nuqs? GitHub 10,000 Stars, 6KB Size, Yet Makes URL a State Container for React

Nuqs makes URL a state container for React, enabling teleportation and time travel.

What is Nuqs? GitHub 10,000 Stars, 6KB Size, Yet Makes URL a State Container for React

Nuqs makes URL a state container for React, enabling teleportation and time travel.

December 19, 2025

Multidimensional updates on the global economic and political environment.

December 19, 2025

Multidimensional updates on the global economic and political environment.

Personalized Matching + Real-time Response, 'Baoli Feiqiao' Helps Reconstruct 'AICon Beijing' Attendee Experience!

Baoli Feiqiao showcased an intelligent attendee mini-program at AICon and introduced its industrial internet platform.

Personalized Matching + Real-time Response, 'Baoli Feiqiao' Helps Reconstruct 'AICon Beijing' Attendee Experience!

Baoli Feiqiao showcased an intelligent attendee mini-program at AICon and introduced its industrial internet platform.

Essential

Efficient Retrieval for Agents: Full Explanation of Apache Doris HSAP Unified Hybrid Search Architecture

Apache Doris 4.0 introduces HSAP unified hybrid search architecture to empower AI Agents with efficient retrieval.

Efficient Retrieval for Agents: Full Explanation of Apache Doris HSAP Unified Hybrid Search Architecture

Apache Doris 4.0 introduces HSAP unified hybrid search architecture to empower AI Agents with efficient retrieval.

Orange Code: Localized Failures — Our Resilience Plan Following Recent Incidents

Cloudflare launched the 'Orange Code' initiative to enhance configuration change controls and system resilience, preventing widespread outages.

Orange Code: Localized Failures — Our Resilience Plan Following Recent Incidents

Cloudflare launched the 'Orange Code' initiative to enhance configuration change controls and system resilience, preventing widespread outages.

Vercel: Our Million-Dollar Hacker Challenge for React2Shell

Vercel successfully defended against React2Shell attacks and shared its security experience.

Vercel: Our Million-Dollar Hacker Challenge for React2Shell

Vercel successfully defended against React2Shell attacks and shared its security experience.

General

Introducing vercel.ts: Programmatic Project Configuration

Vercel has introduced vercel.ts to support TypeScript-based project configuration.

Introducing vercel.ts: Programmatic Project Configuration

Vercel has introduced vercel.ts to support TypeScript-based project configuration.

Pixel Portraits: AI-Generated Trading Cards

Vercel uses AI to generate trading cards, enhancing meeting interaction experiences.

Pixel Portraits: AI-Generated Trading Cards

Vercel uses AI to generate trading cards, enhancing meeting interaction experiences.

Primary

Another NVIDIA Blackwell GPU Launch, Enabling Complex AI Workflows Locally

InfoQ - AI & LLM•Published on December 19

News•AI and Frontier Technologies•Score: 8/10

NVIDIABlackwellRTX PRO 5000desktop AIhigh memory

⚠️

Notable

Although the 72GB memory is tempting, the article does not mention the specific price range of this professional-grade GPU. For most domestic AI developers or small teams, price is often a key factor in deciding adoption. If the price is too high, the 'desktop-level' accessibility will be greatly reduced, potentially remaining exclusive to a few 'well-funded' research institutions or large studios. Additionally, the description of 'full release' seems a bit exaggerated, as it mentions that 'broader global system vendor channels are expected to launch complete workstation products based on this GPU in early next year,' implying that at the time of 2025 December, the actual products available may not be numerous, requiring further waiting, which is a waiting cost for teams in urgent need of computing power.

📈

Perspectives

In my view, NVIDIA's move is a precise capture of the AI trend moving from the cloud to 'edge' and 'local'. In 2025, the complexity of Agentic AI makes pure cloud inference cost and latency unacceptable, while the demand for local development and testing is growing. The launch of the RTX PRO 5000 72GB is essentially giving teams that want to run large models locally without burning data center costs a powerful tool. In the domestic AI hardware market, besides NVIDIA, companies like Huawei and Horizon Robotics are also making efforts. However, currently, NVIDIA still maintains a significant lead in the ecosystem and performance of desktop-level professional GPUs. This product will further solidify its dominant position in the AI developer toolchain. However, for domestic users, supply stability and channel pricing will be crucial factors affecting its market performance. I hope it doesn't end up as another case of 'the waiting crowd always wins'.

Source

◆

Innovating Against Streaming Abuse - And Our Latest Transparency Report

Cloudflare Blog•Published on December 19

News•General News•Score: 7/10

CloudflareTransparency ReportStreaming AbuseCopyright InfringementContent Blocking

⚠️

Notable

While Cloudflare highlights the effectiveness of its technical solutions, it still faces a dilemma regarding content blocking. On one hand, it opposes the 'one-size-fits-all' IP blocking approach used by Spanish LaLiga, arguing that it leads to over-blocking and internet disruption. On the other hand, it must comply with court orders in Europe (Belgium, France, Italy, UK) for geographical blocking. This 'both-and' stance leaves the boundary between commercial interests and user rights unclear. The description of AI tool abuse in the report is somewhat broad, lacking specific case studies and technical details, such as 'LLMs also contribute to the proliferation of low-quality and malicious abuse reports,' which raises questions about how these issues are identified and addressed in practice.

📈

Perspectives

In my view, this report once again underscores Cloudflare's core value proposition as a global internet infrastructure leader—not just speed and security, but also as a participant in internet governance. Especially in 2025, as AI-generated and distributed content becomes increasingly accessible, copyright protection and content abuse will become routine challenges. Cloudflare's 'technology + compliance' dual strategy allows it to meet the needs of rights holders and regulatory bodies while striving to maintain its 'open internet' vision. This is crucial for building its competitive moat. Domestic companies expanding into the global internet market, particularly content distribution and UGC platforms, must closely monitor such reports. Cloudflare's practices indicate that platform governance complexity has surpassed the technical realm, with legal compliance, international policy, and even AI ethics becoming unavoidable 'pitfalls.' Understanding these 'industry unwritten rules' in advance can help mitigate future operational risks.

Source

◆

What is Nuqs? GitHub 10,000 Stars, 6KB Size, Yet Makes URL a State Container for React

InfoQ - Frontend•Published on December 20

Insight•Frontend Development•Score: 7/10

NuqsReactURL State ManagementuseQueryStateType Safety

🔬

Tech Insights

In my view, Nuqs has three core strengths: 1. Core Concept: Directly embedding application state into the URL, an idea that's not new, but Nuqs implements it with type safety and engineering rigor, replacing useState with useQueryState, seamlessly integrating with mainstream React frameworks, which is impressive. 2. Teleportation and Time Travel: These concepts are very vivid, directly hitting product pain points. For applications that need to share specific views or allow frequent state rollback, this is a game-changing feature. Think about complex filters or multi-step forms, being able to directly share the current state can lead to a quantum leap in user experience. 3. Engineering Considerations: It addresses issues like type safety, throttling and debouncing, server-side rendering integration, and URL key mapping, indicating it's not a toy project but one that considers production complexity. The 6KB size is also very appealing.

⚠️

Notable

Of course, this approach isn't without its drawbacks: 1. URL Length Limitation: A well-known issue, URLs aren't databases, so stuffing complex states into them can make the links extremely long, which can degrade user experience. The article mentions this, but it remains a major limitation for this type of solution. 2. Risk of State Uncontrol: Once state is in the URL, it's hard to control its modifications. Users can tamper with it freely, and if the backend doesn't strictly validate or the frontend doesn't implement sufficient error handling, problems can easily arise. Especially when frequently modifying the URL structure, migration strategies become a significant challenge. 3. Limited Applicability: Although impressive, not all states are suitable for being placed in the URL. For example, user-sensitive information or transient UI states (like whether a popup is open) shouldn't be stored here. It's more appropriate for global query or filter states that need to be persisted and shared.

📈

Perspectives

In my opinion, Nuqs is an interesting addition to the frontend state management landscape, but not the silver bullet for replacing existing React state management solutions. It's more of a high-level tool for specific scenarios. For SaaS products and data analysis platforms that require strong sharing and historical rollback capabilities, Nuqs provides an elegant and type-safe solution. However, in China, most business scenarios haven't deeply utilized URL states yet, and teams are more accustomed to using Redux/Zustand/Jotai and other internal state management systems. The emergence of Nuqs may push some teams to reconsider the value of URL as an application entry point and state carrier, especially in scenarios emphasizing SEO and user experience, where it has the potential to become a standard component.

Source

◆

December 19, 2025

Global Vision•Published on December 19

News•Comprehensive News•Score: 7/10

Global EconomyU.S. EmploymentBeijing EnvironmentEU PolicyHainan Free Trade Port

Source

◆

Personalized Matching + Real-time Response, 'Baoli Feiqiao' Helps Reconstruct 'AICon Beijing' Attendee Experience!

InfoQ - AI & LLM•Published on December 20

News•AI and Frontier Technologies•Score: 7/10

AIConBaoli FeiqiaoIndustrial InternetLarge ModelCloud-Edge Collaboration

📈

Perspectives

Baoli Feiqiao, backed by Poly Group, has a 'major advantage' in the domestic industrial internet market, especially in securing large central state-owned enterprise projects. With the group's endorsement and resources, it is easier to enter the market than pure tech SaaS companies. The combination of 'cloud-edge collaboration + large model + domestic innovation compliance' targets the 'highest value area' in China's industrial digital transformation: large state-owned enterprises and central enterprises. However, industrial internet implementation has long cycles and high difficulty, and even with background support, it tests their 'delivery capability and ecosystem building'. Now in December 2025, this sector is extremely competitive, with everyone focusing on 'real implementation' and hard skills.

Source

Essential

Efficient Retrieval for Agents: Full Explanation of Apache Doris HSAP Unified Hybrid Search Architecture

InfoQ - Backend•Published on December 19

Insight•Software Architecture and Engineering Practices•Score: 9/10

Apache DorisHSAPAI AgentHybrid SearchRAG

🔬

Tech Insights

The article accurately identifies the core pain points of databases in the AI era: the traditional 'vector database + search database + OLAP database' multi-system approach has severe issues in data redundancy, query latency, consistency, and unified orchestration, forming 'data silos'. Doris' HSAP unified architecture resolves these issues via a single engine, which is a key breakthrough in the current RAG and Agent application deployments. Doris 4.0's technical implementation is solid, with features like high-performance inverted index-based text analysis (supporting BM25 relevance scoring), semantic search expansion via ANN index (HNSW/IVF), and pre-filtering optimization, virtual column mechanisms, and TopN lazy materialization—all of which are 'killer features' for improving query performance in practice.

📈

Perspectives

In my view, Apache Doris 4.0's HSAP architecture is a strong boost for domestic teams still building 'lego-style' data infrastructure for RAG and Agent applications. Previously, we often mentioned 'storage and computation separation'; now, Doris offers 'search and computation integration', which is the true path to cost reduction and efficiency enhancement. In 2025, Agent and RAG have moved from concepts to practical implementation, demanding high real-time performance and complex query capabilities. Doris' one-stop solution will be highly competitive in the domestic market. It not only meets the performance demands of large companies but also enables SMEs to quickly build AI applications with hybrid retrieval capabilities under limited resources. Teams still tinkering with Elasticsearch + vector database + ClickHouse should seriously evaluate Doris—wasting time is wasting life.

Source

◆

Orange Code: Localized Failures — Our Resilience Plan Following Recent Incidents

Cloudflare Blog•Published on December 19

Insight•Software Architecture and Engineering Practices•Score: 9/10

CloudflareIncident Post-MortemResilient ArchitectureConfiguration ManagementCode OrangeHMD

🔬

Tech Insights

Cloudflare's post-mortem and improvement plan is packed with valuable insights, offering several key takeaways for all architects and SRE teams to reflect on: 1. Strict Configuration Change Management: Cloudflare acknowledges that both incidents originated from rapid, unverified global configuration deployments. They will now treat configuration changes as code deployments, introducing Health Mediated Deployment (HMD) for canary releases and health checks. This is a hard-learned lesson, emphasizing that configuration changes can pose risks equal to or greater than code changes, as they can take effect globally in an instant. 2. Design Philosophy of 'Assume Failure': The article highlights the need to review service-to-service interface contracts and to 'assume failure will occur', then handle it in the most reasonable way. This is a mature resilience design mindset. For instance, when the Bot Management module fails, traffic should not be discarded by default, but rather allowed to pass with a default classification. This is more advanced than simple 'circuit breaking', as it considers business continuity. 3. Incident Response and 'Break Glass' Mechanism: Cloudflare recognizes that security systems and cyclic dependencies can become obstacles during emergencies. They will optimize the 'break glass' process to ensure rapid access to necessary tools and eliminate cyclic dependencies in critical systems. This is crucial for any critical infrastructure that requires quick responses.

⚠️

Notable

Although Cloudflare's post-mortem and plan are detailed, as someone within the industry, I can't help but voice a few criticisms: 1. 'Why didn't we think of this before?': As a leading CDN and security service provider, Cloudflare's negligence in configuration management is surprising. It's hard to imagine that many mature internet companies would not treat configuration changes with the same rigor as software deployments. This exposes a blind spot in their internal SRE practices. 2. 'Hindsight is 20/20': While the improvement plan is comprehensive, the issues of configuration management, failure modes, and cyclic dependencies are well-known in the industry. Cloudflare, as an industry leader, only 'realized' these problems after two major outages, indicating that their engineering culture and processes still have room for improvement. 3. Commitment vs. Execution: The plan is ambitious, but achieving full HMD coverage for all production system configurations by the end of Q1 2026 (i.e., early 2026) is a massive workload. Execution capability will be the biggest test.

📈

Perspectives

Cloudflare's two outages have certainly served as a wake-up call to the global SRE and architecture community. This is not just an internal issue but a common challenge in large-scale distributed systems resilience. Many companies in China, while chasing rapid iteration, often overlook system stability and the rigor of configuration management. In my view, Cloudflare's post-mortem is a must-read SRE textbook. It underscores the importance of configuration management and highlights the central role of 'assuming failure' and 'fast response' in building resilient systems. Looking ahead, I believe more companies will follow Cloudflare's lead in incorporating configuration changes into more stringent deployment processes and deeply contemplating the boundaries of system resilience. For Cloudflare, this incident is a test of brand trust, but if they can execute the plan as promised, it could further solidify their reputation in reliability.

Source

◆

Vercel: Our Million-Dollar Hacker Challenge for React2Shell

Vercel News•Published on December 19

Insight•Cloud Computing and Backend•Score: 8/10

VercelReact2ShellWAFRCESecurity Challenge

🔬

Tech Insights

1. Importance of WAF: Vercel's Seawall WAF intercepted the majority of attacks initially, demonstrating the power of edge security. This again proves that while a WAF is not a silver bullet, it is absolutely essential as the first line of defense, especially for known vulnerabilities. 2. Depth Defense: In addition to the WAF, Vercel implemented extra RCE defenses at the compute layer, which is the key to architectural resilience. Since WAF bypasses are inevitable, runtime protection becomes crucial afterward. 3. White Hat Collaboration: By offering a million-dollar bounty, Vercel collaborated with 116 security researchers to quickly identify and fix 20 WAF bypass techniques. This community-driven security model is more efficient and resilient than internal, closed development.

Source

General

Introducing vercel.ts: Programmatic Project Configuration

Vercel News•Published on December 19

Insight•Frontend Development•Score: 6/10

Vercelvercel.tsconfiguration as codeTypeScriptDX

Source

◆

Pixel Portraits: AI-Generated Trading Cards

Vercel News•Published on December 22

News•AI and Frontier Technologies•Score: 5/10

AI-generatedTrading CardsVercelNext.js ConfUser Interaction

🔬

Tech Insights

In my view, the highlight of Vercel's AI application is as follows: 1. AI Use Case Expansion: This demonstrates that AI is not just about writing code or generating images; it has significant potential in marketing interactions and user experience. Integrating AI-generated content into event registration and on-site experiences is a very clever approach. 2. User Engagement Boost: Allowing users to generate their own personalized cards greatly enhances engagement and the desire to share. People always enjoy showcasing 'exclusive' items, and Vercel has captured this effectively. 3. Technology and Fun Integration: As a technology company, Vercel showcases its platform capabilities (possibly Next.js + Vercel AI SDK) in a way that is both technically sound and fun, making it an excellent example of brand marketing.