Beijing Municipal Government released "Several Measures to Promote Steady and Progressive Economic Development," proposing three supporting policies to promote artificial intelligence applications, including financial subsidies for computing power rental costs and high-quality solutions, as well as support of up to 50 million yuan for demonstration projects.

Gate News: On March 18, Baidu Intelligent Cloud released a price adjustment announcement for AI computing power, storage, and other products. Among them, AI computing power-related product service prices will increase by approximately 5%-30%; parallel file storage prices will increase by approximately 30%. The above price adjustments will officially take effect starting from April 18, 2026 at 00:00 (Beijing Time).

Gate News: On March 18, ShapeShift founder Erik Voorhees made public comments regarding the SEC's explanatory guidance on crypto assets, believing the significance of this guidance lies in the SEC acknowledging its own power boundaries. Erik Voorhees contrasted Gensler's tenure's large-scale enforcement actions against utility tokens, pointing out that these actions significantly decreased entrepreneurs' willingness to design utility tokens at that time, instead catalyzing the emergence of a large number of

On March 18, Alibaba Cloud announced price increases of up to 34% for its AI computing power and storage products due to the global explosion in AI demand and supply chain price increases. Computing cards and file storage products also saw significant price increases. This round of price increases is primarily driven by the surge in Token call volume, with Alibaba Cloud prioritizing AI computing power demand for Token business.

Decentralized GPU computing infrastructure platform Aethir officially launched its managed Kubernetes service on March 17, aiming to expand into enterprise-grade AI infrastructure. The service supports rapid cluster deployment, is compatible with the latest GPUs, and starts at $1.45/hour with enterprise-grade security. Through cost optimization, Aethir has been providing computing power services across 95 countries globally, serving 150 enterprise customers.

HIVE Digital is accelerating business restructuring, reducing bitcoin mining in Sweden and pivoting toward Canadian AI data centers and high-performance computing to address tax pressures and market volatility. The new business model will shift revenue generation from block rewards to computing power service fees to improve cash flow stability, though it also faces intensifying competitive challenges. This transformation reflects industry trend shifts and may reshape the profitability logic of the mining ecosystem.

OpenAI announced on March 17 a major adjustment to the Stargate project, abandoning plans to build its own data centers and switching to third-party cloud services to alleviate financing pressure. The company lowered its expected computing power expenditure for 2030 from 1.4 trillion dollars to approximately 600 billion dollars. A new partnership was reached with SoftBank, which will be responsible for facility development.

Nvidia unveiled Vera CPU at GTC conference, its first processor designed for Agent AI and reinforcement learning, featuring 88 proprietary cores with multi-task parallel processing support. Its high-bandwidth memory reduces power consumption, and multiple cloud service providers and hardware manufacturers have already planned integration. It is expected to enhance AI programming and streaming data processing performance, with shipments anticipated in the second half of the year.

Nvidia announced at its GTC conference the launch of the Space-1 space computing module, which integrates 2 Rubin GPUs and 1 Vera CPU, delivering AI inference computing power 25 times that of the H100, and supports on-orbit data processing. Founder Jensen Huang pointed out that thermal dissipation challenges in space remain to be solved. Initial partners include multiple space enterprises, but the launch timeline has not been determined.

Nvidia CEO Jensen Huang announced at GTC 2026 that the Vera Rubin platform will enter full-scale production, integrating seven new chips designed to build a supercomputer specifically for agentic AI. Compared to the previous generation, the GPU count has been reduced to one-quarter while inference capabilities have improved significantly. Combined orders are projected to reach $1 trillion between 2025 and 2027. Multiple cloud service providers and system vendors will launch related products in the second half of the year.