OpenClaw's influence continues to grow, and CMSC reiterates its recommendation for AI infrastructure in its weekly computer sector review dated March 8, 2026. OpenClaw has become the fastest-growing open-source software project in GitHub's history, with several Chinese large language models ranking among the most frequently utilized globally. Additionally, OpenAI has launched GPT-5.4, and ByteDance has disclosed the pricing structure for Seedance 2.0. Cloud computing services remain the most definitive growth area, and a strong buy recommendation is maintained.
OpenClaw is now the fastest-growing open-source project on GitHub. According to reports, as of March 3, 2026, OpenClaw's number of stars on GitHub surpassed that of the long-dominant front-end framework React, making it one of the most-starred "non-aggregated software" projects on the platform. It took less than four months for OpenClaw to ascend to the top of this category. Citing technology media Wccftech, a report notes that Jensen Huang, speaking at a Morgan Stanley conference, praised the open-source software OpenClaw highly, calling it "the most important software release of our time." Huang believes that AI agents, using only a series of prompts, can execute tasks that previously required significant time and expertise. This has led to an approximately 1,000-fold surge in token consumption, directly creating a "computing power vacuum": as long as agentic AI continues to permeate human workflows, existing hardware deployments, regardless of scale, will face computing constraints.
Chinese large language models are leading in OpenClaw usage. Data from OpenRouter indicates that over the past 30 days, OpenClaw's usage ranked first globally, consuming a total of 8.69T tokens and utilizing 344 different large models. Among the most frequently called models, three of the top five are Chinese: Kimi K2.5, Step 3.5 Flash from Step Fun, and MiniMax M2.5. Four of the top ten models are Chinese, with DeepSeek V3.2 ranking eighth.
OpenAI has officially released GPT-5.4, its first general-purpose large model with native computer operation capabilities. On March 6, OpenAI launched the new GPT-5.4 model, which includes the GPT-5.4 Thinking version for ChatGPT and API access, and the GPT-5.4 Pro version for complex tasks. The company stated this is its most capable and efficient frontier model for professional work to date. GPT-5.4 can autonomously execute keyboard and mouse commands based on screen visuals, working across devices and applications to complete complex workflows, marking a shift from passive response to active execution. The model also supports a context window of up to 1 million tokens, enabling the planning, execution, and full-process verification of long-term tasks. Its tool search efficiency is further enhanced, allowing for more precise invocation of various tools and APIs.
ByteDance has announced the pricing standard for Seedance 2.0. According to information from the Volcanic Engine website, pricing for Seedance 2.0 has been set. The cost is 28 RMB per million tokens for tasks including video input (video editing), and 46 RMB per million tokens for tasks without video input (pure video generation). Generating a 15-second video with Seedance 2.0 consumes 308,880 tokens. Under the pricing for pure video generation at 46 RMB per million tokens, the cost for a single 15-second video is approximately 15 RMB, equating to 1 RMB per second.
Risk warnings include potential delays in AI technology development and intensifying industry competition.
A review of the past week's tech highlights follows.
1. OpenClaw's Momentum Continues OpenClaw has solidified its position as GitHub's fastest-growing open-source project. As of March 3, 2026, its star count reached over 24.8k (exceeding 27.7k by March 8), surpassing React to become one of the most-starred non-aggregated software projects. It achieved this top position in under four months, whereas React accumulated 243k stars over 13 years. This growth rate is exceptionally rare on GitHub. Currently, OpenClaw has attracted contributions from over 900 developers, and its Discord community maintains a real-time online user base of 18,000, establishing it as the world's fastest-growing open-source AI project.
Jensen Huang stated that Agentic AI is at a major historical inflection point. Citing Wccftech, a report from IT Home notes that NVIDIA CEO Jensen Huang, speaking at a Morgan Stanley conference, indicated that Agentic AI is reaching a significant historical turning point. He highly praised OpenClaw, calling it the most important software release of our time, and emphasized that while the Linux operating system took about 30 years to achieve its current level of adoption, OpenClaw surpassed it comprehensively in just three weeks, becoming the most downloaded open-source software in history. Agents can execute tasks requiring substantial time and expertise using only prompts. Huang believes this has caused token consumption to surge by about 1,000 times, creating a computing power vacuum where existing hardware will be constrained as Agentic AI continues to integrate into human workflows.
Chinese LLMs lead in OpenClaw usage. Data from OpenRouter shows OpenClaw's usage was highest globally over the past 30 days, consuming 8.69T tokens across 344 models. Three of the top five most-used models are Chinese: Kimi K2.5, Step Fun's Step 3.5 Flash, and MiniMax M2.5. Four of the top ten are Chinese, with DeepSeek V3.2 ranking eighth.
2. OpenAI Releases GPT-5.4 OpenAI officially launched the new GPT-5.4 model on March 6. The release includes the GPT-5.4 Thinking version for ChatGPT and API, and the GPT-5.4 Pro version for complex tasks. It is described as the most capable and efficient frontier model for professional work. Pricing for the API is set at $2.5 per million input tokens and $15 per million output tokens for GPT-5.4. GPT-5.4 Pro is priced at $30 per million input tokens and $180 per million output tokens. GPT-5.4 Thinking is currently available to Plus, Team, and Pro subscribers.
GPT-5.4 is OpenAI's first general-purpose model with native computer operation capabilities. It can autonomously execute keyboard and mouse commands based on screen content, working across devices and applications to manage complex workflows, transitioning from passive response to active execution. Specifically, GPT-5.4 supports dual operation modes: it can automate computer tasks by writing code using libraries like Playwright, or it can directly issue mouse and keyboard commands based on screenshots. Developers can also customize model behavior via "developer messages" and configure custom safety confirmation policies based on their risk tolerance. The model supports a context window of up to 1 million tokens, facilitating the planning, execution, and verification of long-cycle tasks. Its tool search efficiency is improved, enabling more precise calling of various tools and APIs.
3. Seedance 2.0 Pricing Standard Announced ByteDance has announced the pricing standard for Seedance 2.0. According to the Volcanic Engine website, pricing is set at 28 RMB per million tokens for services including video input (video editing), and 46 RMB per million tokens for services without video input (pure video generation). Generating a 15-second video consumes 308,880 tokens. The service content differs between the two price points; the latter, pure video generation, consumes more computing power and is priced higher. Calculating based on the 46 RMB rate, a single 15-second video costs about 15 RMB, or 1 RMB per second.
For market performance review, during the first week of March 2026, the computer sector declined by 5.29%. The top gainers were Keyuan Wisdom, Zhongke Star Map, Huashi Technology, Southern Power Grid Technology, and ST Diweixun.
Comments