Musk isn’t just doing aerospace here—he’s bringing the same fierce drive he uses to build rockets to disrupting semiconductors and computing infrastructure. elonmusk
At a SpaceX earnings call, Musk said that while the entire industry had been scrambling for GPUs, he saw the real bottlenecks right away: power, cooling, and the speed of engineering and construction are the crucial constraints. $SPCX
Bringing aerospace engineering down to Earth is a classic case of overwhelming the competition with a higher-dimensional approach. Rocket teams deal every day with enormous heat loads in extremely tight spaces and with demanding power requirements. Apply their millisecond-level tolerance for failure to building data centers, and their speed will be in a league of its own. Adding 10 GW of computing capacity in a year would compress infrastructure projects that used to take years into just a few months, completely rewriting the rules of computing supply.
The goal of developing chips in-house is to drive computing costs as low as possible. Starting with SRAM to test its chipmaking capabilities, with Terafab expected to reach mass production in 2028, may seem aggressive—but Musk is building a fully integrated, closed-loop operation. If he manages to slash the price per unit of computing power by an order of magnitude, the entire ecosystem for training and running AI models will be reshaped.
From Starship’s high-frequency reuse, to massive computing power on the ground, to an in-house chipmaking loop, Musk is putting together another massive piece of his empire.
In the short term, Nvidia $NVDA and ASML $ASML can still reliably reap the benefits of equipment demand. But in the long run, if SpaceX’s Terafab model works, traditional data centers and chip-design giants will face brutal pressure from its engineering efficiency.
If Starship really manages to launch twice a week next year, combining low-Earth-orbit satellites with massive ground-based supercomputing will give Musk an end-to-end level of control that nobody else can match.
DYOR
#股票财报季
At a SpaceX earnings call, Musk said that while the entire industry had been scrambling for GPUs, he saw the real bottlenecks right away: power, cooling, and the speed of engineering and construction are the crucial constraints. $SPCX
Bringing aerospace engineering down to Earth is a classic case of overwhelming the competition with a higher-dimensional approach. Rocket teams deal every day with enormous heat loads in extremely tight spaces and with demanding power requirements. Apply their millisecond-level tolerance for failure to building data centers, and their speed will be in a league of its own. Adding 10 GW of computing capacity in a year would compress infrastructure projects that used to take years into just a few months, completely rewriting the rules of computing supply.
The goal of developing chips in-house is to drive computing costs as low as possible. Starting with SRAM to test its chipmaking capabilities, with Terafab expected to reach mass production in 2028, may seem aggressive—but Musk is building a fully integrated, closed-loop operation. If he manages to slash the price per unit of computing power by an order of magnitude, the entire ecosystem for training and running AI models will be reshaped.
From Starship’s high-frequency reuse, to massive computing power on the ground, to an in-house chipmaking loop, Musk is putting together another massive piece of his empire.
In the short term, Nvidia $NVDA and ASML $ASML can still reliably reap the benefits of equipment demand. But in the long run, if SpaceX’s Terafab model works, traditional data centers and chip-design giants will face brutal pressure from its engineering efficiency.
If Starship really manages to launch twice a week next year, combining low-Earth-orbit satellites with massive ground-based supercomputing will give Musk an end-to-end level of control that nobody else can match.
DYOR
#股票财报季