Elon Musk dissolved xAI and merged it into SpaceX. Meanwhile, the Colossus 1 supercomputer cluster originally under xAI, which contains more than 220,000 NVIDIA GPUs (including H100, H200, and next‑generation GB200), is exclusively leased to AI company Anthropic for inference services of the Claude series LLMs. This move marks SpaceX’s deep entry into the global AI computing infrastructure sector.
The SpaceX‑Anthropic collaboration includes a long‑term vision for orbital AI computing power. Anthropic has expressed interest in co‑developing several gigawatts of orbital computing capacity, believing that space‑based computing can provide near‑infinite clean energy and significantly reduce environmental impact on Earth.
NVIDIA announced an expanded partnership with data center operator IREN, including an equity investment opportunity of up to 70 per share) and a five‑year GPU cloud service agreement valued at $3.4 billion. The two parties also plan to deploy up to 5 gigawatts of AI data center infrastructure globally. After the announcement, IREN’s stock surged as much as 27% in after‑hours trading.
SoftBank is in talks with NVIDIA and Foxconn to evaluate the feasibility of “Made in Japan” AI servers. The plan initially involves assembling systems using externally sourced components, with the goal of mastering the entire manufacturing process by 2029. This could further reshape the global landscape of AI computing hardware production.
GPU‑related computing power stocks in A‑share markets (including Baidu) rallied strongly. Memory chips and CPU‑centric computing chips also surged, driving the STAR 50 Index up by 5.47%. This rally is led by domestic Chinese computing chain players, contrasting sharply with the earlier overseas‑focused computing chain (centered on optical modules), highlighting the accelerating trend of domestic computing power self‑sufficiency.
AMD released its fiscal Q1 2026 earnings, with revenue of 1.383 billion (up 95% year‑on‑year). Data center revenue reached 120 billion, boosting stocks such as Hygon (a Chinese GPU maker) in the A‑share market.
DeepSeek officially released and open‑sourced the DeepSeek V4 large language model, which is fully adapted to eight domestic chips including Huawei Ascend. For the first time, the model is fully deployed (from training to inference) on domestic chips, marking a new logic of self‑reliant industrial competition in China’s AI sector.
Hong Kong Broadband announced the launch of an AI+ domestic computing platform, investing HK$100 million to bring in domestic GPU computing resources from M‑Thread and Biren Technology. All hosting facilities are located in data centers within the Cyberport campus, making it the first telecom operator in Hong Kong to offer enterprise‑grade AI computing power as a paid service.
China’s National Development and Reform Commission and three other departments jointly issued the “Action Plan on Promoting Two‑Way Empowerment between Artificial Intelligence and Energy,” deploying 29 key tasks focused on secure and reliable energy supply for computing facilities, green and low‑carbon transformation, and integrated computing‑electricity coordination. Meanwhile, the “Computing Power Supermarket – SME Zone” went live on the China Computing Platform, providing affordable computing services for small and medium‑sized enterprises.
The four major North American cloud providers (Google, Microsoft, Meta, Amazon) recorded a combined capital expenditure of 710 billion. At the same time, global GPU rental prices rose from 2.35/hour in March 2026, an increase of nearly 40%, with high‑end GPU utilization rates exceeding 90%.