Musk's AI power crisis warning at G20

Musk told G20 technology ministers that electricity, not chips or algorithms, is now the binding constraint on AI growth ^[1]. What went less remarked in the framing is that SpaceX is a direct beneficiary of exactly the scarcity he was describing. Google is paying SpaceX $920 million a month for bridge compute through mid-2029 ^[2], and Anthropic has leased the entirety of SpaceX's Colossus 1 data center in Memphis for $1.25 billion a month through 2029, a contract worth more than $40 billion in total and roughly half Anthropic's current annualized revenue ^[3]. Combined, the two deals are worth more than $26 billion a year, more than SpaceX's entire 2025 revenue ^[3]. That does not make the underlying shortage untrue, but it does mean the loudest global messenger for the crisis is also its most immediate commercial winner, arriving just as SpaceX and xAI merge into a $1.25 trillion entity ahead of a planned IPO ^[3].

Musk's specific claim was a projected shortfall of at least 15 gigawatts of power for AI chips by 2027 ^[1]. His reasoning: AI chip production is rising roughly 40-50% a year, while power availability outside China is growing only 10-20% a year ^[4], a gap that compounds fast enough to outpace new data-center buildout well before the decade is out. SpaceX's own Colossus 1 facility, built in a record 122 days with 200,000 GPUs and 300 megawatts of capacity, is effectively Musk's proof of concept that compute can be stood up faster than traditional utility-scale power projects can catch up ^[3]. Video coverage of the same G20 appearance corroborates the same figures verbatim, along with Musk's aside that China has abundant electricity but remains blocked from the most advanced chips by export controls, and that combined digital-plus-robotics AI value could run "10x or more" beyond his headline $20-30 trillion estimate.

The power-shortage framing did not stand alone at the G20 session; it was paired with an explicit push against AI oversight. Musk argued that new technologies should be presumed legal absent explicit restriction, and singled out the EU by name: "the regulation level is extraordinarily high, and things are generally default illegal, and this inhibits progress with new technologies" ^[5]. David Sacks, the Trump administration's AI czar, spoke at the same session opposing the creation of strict AI regulatory bodies ^[5]. Not everyone in the room agreed: Google DeepMind's Demis Hassabis used his remarks to call for pre-release testing of powerful AI systems, a more cautious counterpoint delivered at the exact same event ^[5]. The sequencing is notable - an infrastructure-scarcity argument for building faster, immediately adjacent to an argument for regulating less.

Public reaction split along two tracks. On X, Musk's own recap of his G20 remarks was widely amplified alongside straight news relays of the Musk-Zuckerberg data-center plea, with little visible pushback in the immediate response. On Reddit, the tone was sharper: threads discussing the power-shortage claim leaned heavily skeptical of Musk personally, citing his track record on prior predictions and his evident financial interest through SpaceX and his other energy holdings. Underneath the distrust, though, a more substantive technical argument played out - whether the real bottleneck is power generation at all, or grid transmission and interconnect capacity, with commenters pointing to Texas alone fielding requests for 474 gigawatts of new grid interconnects, far beyond what the state can currently produce. That debate, over whether more power plants or more transmission lines are the actual fix, unfolded alongside broader distrust of Musk's own predictive track record.