In January 2025, OpenAI unveiled a joint venture with SoftBank and Oracle, with the ambitious plan to invest a staggering $500 billion over a four-year period to construct data centers. However, the project has encountered a major obstacle. There have been substantial disagreements among the three partners regarding the division of labor and structural arrangements. These disputes have ultimately led to the shelving of OpenAI's grand vision of building its own data centers. As it stands, the project has yet to kick off; the team responsible for its execution has not been fully formed, and the cooperation between the parties remains largely one-sided.

In the meantime, Musk's xAI is forging ahead at a breakneck speed. Its inaugural factory, Colossus I, was constructed with remarkable efficiency. Moreover, Colossus 2 has emerged as the world's first GW-scale supercomputing cluster, a testament to xAI's rapid technological advancements. In addition to its terrestrial achievements, xAI is venturing into the realm of space computing. It has acquired xAI (presumably a different entity or a strategic move within the company's ecosystem) and is contemplating renting out Tesla's idle computing power, further expanding its technological footprint.

Nevertheless, the frantic pace of data center construction has not been without its drawbacks. xAI's data centers have become the subject of complaints due to pollution and noise issues. Furthermore, the surge in electricity demand from these data centers has put some regions in the United States at risk of power outages. At a recent event, Altman addressed the pressing issue of data center energy consumption. He proposed a comparison between the energy expended to answer questions after model training is complete and the energy humans use to provide the same answers. According to this metric, AI energy efficiency may have already reached or even surpassed that of humans, a thought-provoking insight into the evolving landscape of artificial intelligence.