Musk: The World Seriously Underestimates the Scale of Computing Power Needed in the Future
Elon Musk stated that the world is seriously underestimating the scale of computing power needed in the future. He emphasized in response to related discussions that people do not understand how much computing power the world will require, reiterating that "they really don't understand." Musk also pointed out that due to land-based power supply and licensing issues, orbital computing power may become the only way to expand AI around 2029. SpaceX plans to exceed 2 GW of computing power by the end of this year, aiming to accumulate close to 10 GW or even higher by the end of next year, and has decided to exclusively adopt NVIDIA architecture. Musk believes that AI revenue will soon surpass SpaceX's other businesses, and long-term demand for computing power will far exceed current expectations. On the market mechanism level, the significantly revised expectations for computing power demand are driving capital to accelerate towards chip, power, and data center construction, with buyers increasing their stakes in NVIDIA and related infrastructure, leading to a rise in the event-driven orbital computing concept, while the land-based computing bottleneck is under pressure, benefiting early layout participants. Supplementary data shows that SpaceX currently has about 1.4 GW of computing power, with a goal to achieve higher scales through a series of projects. Source: Public Information
ABAB AI Insight
As the head of SpaceX and xAI, Musk has previously warned multiple times that computing power will become a core bottleneck for AI expansion, prompting the company to build large-scale computing clusters. Historical behavior shows that he views computing power as a strategic asset equally important as rockets. On the capital path, SpaceX is investing huge capital expenditures into computing power construction and plans to resell surplus capacity, motivated by the goal of seizing the high ground in AI training and inference infrastructure, strategically avoiding power and licensing restrictions through both land and orbital paths. Similar cases can be seen in public statements from tech giants like Microsoft and Google, which have been unable to fully monetize due to computing power shortages. The current AI industry is transitioning from a model competition to a computing power arms race. The structural judgment belongs to technological substitution: because land power and licensing bottlenecks limit scale expansion, this mechanism makes orbital computing power shift from a marginal concept to a necessary option, forcing capital to transition from a single ground data center to a multidimensional energy and computing layout, with pricing power shifting from chip supply to power and deployment freedom. ABAB News · Law of Cognition
- Computing power demand is always systematically underestimated.
- Land bottlenecks force orbital options.
- Those who stockpile computing power in advance hold pricing power.